Conference Digest

Europa Clipper, launched by NASA in 2024, is currently on its cruise to the Jupiter system, where it will perform a comprehensive investigation of Europa through a series of close flybys. Achieving the mission's science objectives depends on the reliable return of large volumes of data collected by its instrument suite during each encounter. To support these requirements, the spacecraft employs a high-rate Ka-band telecommunications capability using its High Gain Antenna (HGA) for science data return and precision radiometric tracking. Because opportunities for high-rate downlink are constrained by mission geometry, Deep Space Network (DSN) availability, spacecraft operational activities, and the demanding communications environment associated with Jupiter operations, maximizing data return requires careful coordination between the telecommunications and mission operations teams. This paper summarizes the Ka-band telecommunications performance observed during the mission's early cruise phase, including in-flight antenna calibration results, comparison of measured link performance with pre-launch predictions, operational considerations for scheduling Ka-band contacts with the DSN, and lessons learned from initial mission operations. These early experiences provide insight into Ka-band operations for long-duration deep-space missions and offer guidance for optimizing future high-rate communications throughout Europa Clipper's science mission and for upcoming planetary exploration missions employing similar telecommunications architectures.

Launched in October 2024, Europa Clipper will explore Jupiter’s icy moon to investigate its potential habitability. The spacecraft follows a Mars-Earth Gravity Assist (MEGA) trajectory and is scheduled to arrive at Jupiter in April 2030 after a 5.5-year interplanetary cruise. During this period, when the spacecraft is closer to the Sun at < 2 AU, Clipper is required to fly with the High Gain Antenna (HGA) pointed toward the Sun as a sunshade during the challenging thermal environment. This pointing constraint limits the data rates of the spacecraft based on the Sun-Probe-Earth (SPE) geometry over time and the locations and fields of view for the remaining antennas. The periods when the data rates fall below 1000 bits per second due to this thermal constraint are designated as low data rate (LDR) periods. To maintain spacecraft visibility via telemetry and enable critical commanding during LDR periods, the team developed an “HGA burst” communication strategy. This approach involves temporarily slewing the HGA away from its Sun-pointing thermal safe attitude to point towards Earth for brief communication windows when the spacecraft is between 1.3 – 2 AU. During these bursts, the spacecraft can achieve nominal data rates, enabling uplink of time-critical commands and downlink of essential telemetry and health status data. Each burst is carefully planned to balance thermal constraints with communication needs. This paper describes the mission’s planning, development, implementation, and adaptation of activities to increase visibility to the spacecraft during these LDR periods, including the operational constraints, thermal risk management, and ground system coordination required to execute HGA bursts successfully.

Dragonfly is a NASA New Frontiers mission that will send a rotorcraft lander to Titan, a world with low gravity and a dense atmosphere, conditions that make flight an ideal method for surface exploration. The lander will be powered by a Multi-Mission Radioisotope Thermoelectric Generator (MMRTG), whose heat will be distributed throughout the lander using ducting and a small fan. Because the MMRTG does not provide enough instantaneous electrical power to fly, its energy will be stored in a large lithium-ion battery for use during flight and science operations. For most of the mission, the lander will be in hibernation, during which power is conserved by limiting activity to essential components needed to maintain functionality and battery charge. Effective temperature regulation on Titan is critical to ensure the longevity and performance of the battery and sensitive electronics. While Titan’s ambient surface temperature is extremely stable, and cold at 94 +/- 2 K the average ambient wind can vary considerably, from 2 m/s at the surface down to dead calm conditions. To modulate heat rejection from the lander in response to changes in the external wind, the Lander will be equipped with a cold duct trim device. Diverter valves, located fore and aft of the Lander ducting, will shunt warm gas through the bypass duct at the bottom of the lander, allowing for dynamic thermal control based on both internal and external conditions. During both active operation and hibernation, a Field Programmable Gate Array (FPGA) embedded with a LEON Processor will remain active to regulate the inlet and battery temperatures by adjusting the diverter valves using a Proportional-Integral (PI) control algorithm. Feedback from both the battery cells’ temperature and the internal air temperature will enable the thermal controller to keep the battery and electronic components' temperatures at their prescribed setpoints during hibernation. This paper presents a design overview of Dragonfly's thermal control system and the Thermal Control Algorithm (TCA) developed for thermal regulation. The thermal modeling approach used to support TCA development and evaluation is described, followed by simulation results demonstrating closed-loop performance under wind and heat disturbances. Experimental results obtained using a full-scale thermal model of the lander in a thermal vacuum chamber further validate the effectiveness of the proposed thermal control approach.

The Skyfall helicopter mission, led by NASA’s Jet Propulsion Laboratory in collaboration with NASA Ames Research Center and commercial partner AeroVironment, aims to deliver three Ingenuity-class helicopters to the surface of Mars via a mid-air deployment that ejects the helicopters off a platform after which they will land themselves. Once on the surface, the helicopters will conduct an unprecedented science survey, traveling significant distances to map subsurface ice deposits while collecting other surface and atmospheric data that will enable scientists to characterize Martian geology and weather. This flight project is on track for a launch readiness date in 2028, a fast development cycle that leverages heritage, commercial components, and, where possible, a test-heavy/hardware-rich approach to manage scope and risks. The Skyfall helicopter design is largely pulled from Ingenuity but includes a handful of higher-maturity improvements that were baselined for the Sample Recovery Helicopter (SRH), including an updated blade design and larger not-to-exceed (NTE) mass of 2.7 kg (compared to Ingenuity’s 1.8 kg). However, due to its novel egress approach, the Skyfall helicopter must operate in flight conditions that represent a significant expansion from Ingenuity’s flight envelope, driving a need for innovation in the helicopter’s guidance, navigation, and control (GNC) systems along with an understanding of aero- and flight- dynamics at these extremes in altitude, velocity, and disturbances. In particular, the Skyfall helicopters will be mounted to a deployment platform that in its deployed configuration is suspended below a backshell on a bridle, with the backshell attached to a parachute. This system is expected to be traveling ~30-40 m/s vertically at its terminal velocity, and in this condition the helicopters will need to spin up to full rpm (at an angle of attack design to minimize thrust generation) while fully attached to the deployment platform. From here, the system will trigger retro rockets on the backshell to deliver the helicopters to a selected drop-off vertical velocity and altitude where the bridle is cut. It is then that the helicopters will simultaneously receive a spring-assisted push to separate from the deploy platform. As the helicopters leave the platform, they begin a “Skyfall maneuver” designed to allow each heli to handle the initial conditions of its flight and execute the maneuver necessary to land on the surface. The specifics of the maneuver design, as well as the algorithms and dynamics modeling that supports its design/execution are described in the paper. This paper will also summarize the preliminary GNC design of the Skyfall helicopter, including key interfaces from the entry/descent spacecraft stage, GNC algorithms/software design, the GNC sensor suite, and GNC verification and validation (V&V) campaign design. This document has been reviewed and determined not to contain export controlled technical data. Pre-Decisional Information – For Planning and Discussion Purposes Only

The Europa Clipper mission, launched in October 2024, is designed to conduct detailed reconnaissance of Jupiter’s icy moon Europa and assess its potential habitability. A critical enabler of its science objectives is the spacecraft’s telecommunications subsystem, which provides high‑rate data return over extreme distances while supporting precise navigation and robust spacecraft health monitoring. The subsystem forms the link between flight and ground, providing command reception, science and engineering data return, navigation data (including ranging, 2‑way Doppler, and ΔDOR), and supports the gravity and radio science investigation. Since launch, telecom operations have evolved through several key phases and activities: initial acquisition, flight system characterization including a RF checkout, two low date rate operations, Mars flyby, high rate Ka-band operations, first solar conjunction, fan-beam and low gain antenna characterizations, high gain calibrations, Earth flyby planning, and routine cruise operations. While some of these phases are unique, the success of each one continues to build confidence and preparedness in the system’s ability to support the intense flyby campaign planned in the Jovian system. This presentation will cover these in detail and review lessons learned from the mission so far.

NASA’s Earth Surface Mineral Dust Source Investigation (EMIT) mission has been operating on the International Space Station (ISS) since July 2022. The primary mission has been to characterize dust sources to provide more accurate inputs to climate models’ treatment of airborne dust. To meet the cost and schedule constraints of the NASA’s NASA Announcement of Opportunity dated August 16, 2016 “Second Stand Along Missions of Opportunity Notice – Program Element Appendix S: Earth Venture Instrument-4,”, EMIT was proposed to maximize the reuse of designs from previous airborne and space flight missions. These elements needed to be packaged to be compatible with the unique requirements of flying on the ISS. This was most challenging in the thermal design – specifically the need to survive an unpowered robotic transfer lasting up to six hours. Early in the mission development, a requirement to move for the 28 Volt to 120 Volt power service created additional packaging and thermal challenges. To minimize the cost impact and maintain the development schedule, these new requirements were likewise met by adapting additional heritage designs. At launch, EMIT incorporated elements ranging from significant design heritage to build-to-print copies to spare flight units deriving from 2 airborne missions (Next Generation Imaging Spectrometers (NGIS) and Airborne Visible/Infrared Imaging Spectrometer (AVIRIS)) and five space missions (Moon Minerology Mapper (M3), OPALS, OCO-2, OCO-3 and ECOSTRESS). Further, the mission operations planning system was inherited from OCO-3 and ECOSTRESS while the Science Data System was inherited from the AVIRIS-family of airborne spectrometer. Given such a heterogenous set of hardware – all designed for different mission requirements - preflight performance predictions for how they would perform together were sometimes difficult. Adding to the challenge was the need to use high-performance Field Programmable Gate Arrays (FPGAs) with a known radiation sensitivity as these were the only technical solution that could meet the data processing and power requirements simultaneously. In this paper, we will examine some of the trade studies that led to the final design, present preflight predictions, and review the inflight performance of the first four years of operations. Specific examples include the predicted and observed Single Event Upset (SEU) rates and the impact these cause on operational efficiency, the preflight and observed thermal performance for the robotic transfer, survival mode, and operations, and the radiometric performance. For each of these, we will also examine the expected versus observed rates of degradation. Where significant observed performance margin is observed, we will reexamine the decisions that created this margin with an emphasis on lessons for future mission development. Finally, we will briefly review the science results obtained from EMIT and expectations for future performance and science investigations.

The Explorer for Artemis Geology, Lunar and Earth - Visible to Short Wave Infrared (EAGLE-VSWIR) mission is poised to revolutionize Earth observation by providing new global sub monthly imaging spectroscopy measurements, while also laying a foundation for future Moon and Mars exploration architectures. Originally conceived as the Surface Biology and Geology (SBG) designated observable from the 2017 Decadal Survey, the mission reformed in early 2026 into EAGLE-VSWIR to enable a streamlined, faster delivery for a 2028 launch. This paper details the complex formulation efforts, risk reduction investments, and the critical shift from a NASA Class C to a Class D risk classification, which relaxes traditional standards to increase programmatic agility. A central enabler for this mission's efficiency and affordability within a strict cost cap is the intentional integration of commercial space capabilities. This includes utilizing the IM-300 flight system platform, executing a Firm Fixed Price (FFP) contract strategy, and leveraging KSAT for Mission Operations and Ground Data Systems. By adopting an "instrument-first" approach built upon SBG investments and leaning on proven commercial practices, EAGLE-VSWIR demonstrates a new, rapid paradigm for Earth Science mission implementation.

NASA’s PACE observatory launched on 8 Feb 2024, and has been on-orbit for just over three years of its 38 month primary mission. This “in-house project” was conceived, designed, built, tested, and is operated by NASA’s Goddard Space Flight Center in Greenbelt Maryland. PACE’s first three years in orbit have demonstrated the power of hyperspectral ocean color measurements and polarimetry from its three instruments, enabling novel science products. This paper describes the flight experiences of the PACE mission as it nears its fulfillment of primary mission requirements, including an overview of science achievements to date and a detailed presentation of routine operations, orbital maneuvers, in-flight anomalies, safe-hold recoveries, and risk mitigation maneuvers to avoid orbital debris.

The Autonomous Ion Mass Spectrometer Sentry (AIMSS) is a next-generation space flight sensor currently operational aboard the International Space Station (ISS), providing real-time, high-resolution measurements of the space environment in low-Earth orbit (LEO). Launched on May 15, 2026, AIMSS delivers critical data for three mission-essential capabilities: (1) scientific validation of space weather forecasting models, (2) autonomous detection of hazardous spacecraft charging events that endanger astronauts and electronics, and (3) real-time identification of human-made contaminants from propellant leaks, thruster plumes, and thermal system venting. Operating at 400 km altitude (approximately 250 mi), AIMSS provides high-mass-resolution measurements to advance real-time understanding of the space environment in LEO. The payload features a simple design, making it cheaper and faster to build and integrate. AIMSS provides high-mass-resolution measurements and real-time alerts, delivering high-quality science and contamination-monitoring data for space-based assets. Over the last two decades, the International Space Station (ISS) has served as a critical test bed for research and technology development, on-orbit study of astronaut health and performance, and operational practices required for long-duration space missions. Launched on May 12, 2026, as part of the Space Test Program Houston (STP-H11) mission, AIMSS provides ion composition measurements to advance scientific understanding of basic ionospheric science and the complex interactions between the natural space environment and space systems. The AIMSS payload is a double focusing mass spectrometer and utilizes electric and magnetic field geometries to focus in both direction and energy. This design allows for multiple ion species to become spatially distributed by mass-to-charge ratio (M/q) along the focal plane. Specifically, lighter ions entering the analyzer will have a shorter path length than heavier ions, creating a series of concentric paths almost like a rainbow. The instrument is comprised of: (1) a laminated collimator to set the field-of-view (FOV); (2) a laminated electrostatic analyzer (ESA) to selectively filter ions by the energy-to-charge ratio (E/q); (3) a magnetic sector analyzer (designed and built by Plasma Controls at Colorado State University) to separate ions by M/q; and (4) a microchannel plate (MCP) followed by position sensitive cross delay anode (XDL) assembly to detect the location of the ions on the detector plane. We will present an overview of the instrument design, electro-optic modeling, calibration results, and initial on-orbit measurements.

The Habitable Worlds Observatory (HWO) is NASA’s next great observatory. It will search exoplanets for biosignatures; it will truly look for life beyond our planet! HWO will use a coronograph to filter out the light from the exoplanet’s star, overcoming a ratio that is approximately ten billion to one. This requires stability in the picometer range, which is beyond today’s state of the art. To burn down the development risk and prepare for a successful program, NASA has issued contracts to several companies for technology development and architectural studies. Northrop Grumman was awarded one of these studies with our System Technologies for Architecture BaseLinE Studies (STABLE-S) program. Our STABLE-S program will mature structural stability capabilities and will gather test data to develop methods of protecting the telescope from micrometeoroid and stray light pollution. In parallel, our larger focus is on our Multi-disciplinary Architecture Study (MAS). Our MAS evaluates a wide parameter space against a broad range of criteria. We have identified over fifty specific parameters in broad areas such as environmental shielding, vibration isolation, mission lifetime, Observatory servicing, ConOps, and launch vehicle compatibility. The first phase is a wide but quick look to identify cliffs and plateaus to rule out a large set of architectures. Our second phase focuses on the coupling between parameters to increase our model fidelity. Our third and final phase will be to perform detailed modeling of the chosen architecture options to help optimize NASA’s final design.

Occulters are essential tools for blocking starlight to enable the direct imaging and spectroscopic study of exoplanets. Mission concepts such as the Hybrid Observatory for Earth-like Exoplanets aim to combine the capabilities of large ground-based telescopes with orbiting star shade occulters to detect and characterize exoplanets. A primary challenge in these missions is maintaining precise positional and orientational alignment between the starshade and the telescope during observations. Alongside the requirements for individual observations, multi-target observation campaigns must design efficient trajectories to maximize the number of observable target stars without exceeding the starshade's maneuverability and propellant constraints. Finally, maintaining the sensitive structure and shape of the starshade itself poses an additional layer of difficulty for the mission In this case study, we model in simulation station-keeping requirements for a star shade paired with a ground-based observatory and assess key observation campaign design attributes. Both individual observations and campaigns of multiple observations were investigated. Furthermore, we also explore flight system design and operational strategies informed by the results of the observation and campaign simulations, which are necessary to implement multi-star observation campaigns effectively. Translating the simulation assumptions, constraints, and outcomes into a feasible mission architecture required addressing many engineering challenges. These challenges included both operational considerations, such as strategies for orbital debris mitigation, and architectural considerations, such as propulsion system design to achieve the required total delta-v and acceleration requirements. Implementing a ground based telescope paired with an orbital starsahde presents a unique interplay between flight dynamics, mission architecture, and operational constraints. Our simulations found that mission performance is highly sensitive to the duration of individual observations and expanding the initial pool of valid target stars significantly increases the total number of observations achievable within a fixed propulsion budget. The simulations also demonstrated that the constraints imposed by the structural limitations of the starshade lead to a measurable reduction in campaign performance relative to theoretical models.

The Earth system is rich with new high-value science and application opportunities uniquely addressed through imaging spectroscopy in the visible to short-wavelength infrared (VSWIR) portion of the spectrum. This range captures spectral signatures produced by atomic and molecular electron transitions, combined with vibrational, scattering, and emission processes of the land, water, and atmospheric constituents of our planet. Building upon decades of investments, advances, and precursor projects such as the Earth Surface Mineral Dust Source Investigation (EMIT), NASA is developing EAGLE-VSWIR, a next-generation global imaging spectrometer mission. The core state-of-the-art instrument delivers a 190-km swath with 30-m spatial sampling and contiguous 10-nm spectral sampling from 380 to 2500 nm. The mission implementation provides complete land and coastal coverage every 16 days. EAGLE-VSWIR closely aligns with NASA's Earth Science to Action strategy by addressing a suite of cross-cutting objectives: Economic Resource Development and Food Security: Enabling field-scale precision agriculture, mapping critical mineral resources, and supporting the blue economy through aquatic characterization. Natural Hazards and Resilience: Quantifying wildland fire fuel conditions, assessing post-fire recovery, and monitoring geologic hazards like volcanic activity and landslide susceptibility. Water and Ecosystems: Improving estimates of snow accumulation and melt for water resources, while quantifying vegetation composition and physiological dynamics across terrestrial and aquatic ecosystems. Environmental Quality: Identifying suspended sediments, harmful algal blooms, and acid-generating minerals to monitor ecosystem integrity. The core instrument captures over 1.5 million spectra per second and is validated through a Dyson imaging spectrometer demonstration unit developed in 2025. EAGLE also views the Moon for direct spectroscopic measurements and advances technologies and reduces risks for future Moon and Mars missions. This paper details the application and science objectives, the VSWIR instrument, and breakthrough observational capabilities of the EAGLE-VSWIR mission

Powered flight on Mars, first demonstrated by NASA JPL’s Ingenuity helicopter, has opened up new possibilities for exploration on Mars in greater access, range, and modalities previously not available with landers and rovers alone. Chopper is a next-generation science-driven hexacopter concept that expands on Ingenuity’s technology demonstration to carry over 3 kg of science payload and support traverses over 3 km/flight over a wide variety of terrain types. However, to fit a hexacopter of sufficient size into a heritage aeroshell, half of the rotors are on one vertical station and the other half is on a lower vertical station, with significant overlap of the rotor disks. This tightly-packed configuration means that the wake of the upper rotors will impinge on the lower rotors, and each rotor influences the flow (and thus the aerodynamic performance) of its neighbors. Severe wake impingement can lead to a loss of thrust, increase in power demand, and/or development of unsteady loads that impact controllability in ways that can impact the feasibility of this design. However, the aerodynamics of wake impingement in a CO2-dominant, very low Reynolds number atmosphere is poorly characterized, and relevant physical data in this regime is limited. Computational fluid dynamics (CFD) runs were used to estimate the impact of the effect, which, when included in our flight dynamics model, ultimately predicts that the Chopper configuration concept would still able to perform over its flight envelope. However, without relevant experimental data, it is difficult to validate the aerodynamic models. The experiment that is the subject of this paper was conducted in order to provide that validation dataset. The water tank experiment, performed at Caltech’s Earth Surface Dynamics Laboratory, consisted of two commercial rotors with blade lengths scaled to fit into the test area and a vertical and horizontal offset scaled to match that of the Chopper design of neighboring rotors. The rotors were towed and rotated at speeds to provide the necessary dynamic similarity to generate Reynolds matching to expected Mars conditions at around 0.02 kg/m3. Although the loading in water is much higher than it would be on Mars, a water tank proves to be a reasonable proxy for the wake dynamics and provides a much cleaner flow environment at the necessarily slow speeds than many wind tunnels could achieve. Force and torque measurements at the hub of each rotor were collected in three rotor conditions (on/off, off/on, on/on) in order to extract the difference that the wake impingement induces on the system in the on/on configuration. The experiment was run over a variety of test conditions spanning Reynolds number, rotor attitude relative to the flow, and rotor horizontal spacing. We also collected dye visualization data for a selected set of test conditions. This paper summarizes the water tank experiment design, test articles, configurations, and experimental results, setting the stage for future work to discuss the model validation and comparison to CFD.

The Skyfall mission plans to fly three autonomous helicopters on Mars to further the understanding of ice deposits below the Martian surface. Despite inheriting design components from previous Mars helicopter missions, Skyfall’s unique mid-atmosphere deployment introduces the additional challenge of navigating extreme aerodynamic regimes, ambient conditions, and descent patterns. This work presents a testing infrastructure for a Captive Carry Cube (CCC), an avionics unit modeling the Skyfall helicopter sensor box and modified from the Mars Sample Recovery (SRH) CCC testbed, to assess the robustness of the Skyfall helicopter sensor suite. To help retire risks associated with the Skyfall maneuver, an adapted model of the CCC is mounted to a drone for flight testing on Earth. This test explores the performance of the helicopters’ on-board sensor package and navigation algorithms under the Skyfall-like flight conditions. To address these concerns, two important additions are introduced to the SRH CCC. The primary addition is the introduction of the Time Sync System (TSS) which consists of a Global Positioning System (GPS) module, an inertial measurement unit (IMU), and a Raspberry Pi 5. The purpose of the TSS is to ensure that the adapted CCC system is totally independent of the drone GPS and IMU, and therefore not tied to any specific drone. The TSS provides time stamps and ground truth flight data via the GPS and IMU, respectively. The second addition is a mounting structure capable of securing the CCC to the test drone at a ground-adjustable angle with respect to the ground to allow for a variable navigation camera perspective. This work results in a test-ready system that assesses the robustness of the Skyfall helicopter sensor packages, paving the way for a successful science mission that advances exploration objectives for Mars. The paper presents the design of the test system and preliminary operations results in advance of field testing.

The Skyfall Mars exploration architecture employs aerial vehicles to enable rapid access to geographically distributed science targets and sample recovery locations. A critical challenge for such architectures is the characterization of the aerodynamic environment encountered during mid-air deployment and initial flight operations. Wake interactions, rotor inflow distortion, and large-scale flow-field nonuniformities can influence vehicle controllability, flight performance, and guidance, navigation, and control (GNC) behavior during these mission phases. Accurate prediction of these effects requires aerodynamic models and methodologies that bridge the gap between high-fidelity models and system-level flight simulation. This paper presents the development of a batch of system-level aerodynamic models and scenarios for Skyfall using STAR-CCM+ under representative Martian atmospheric conditions. The framework is designed to characterize the flow environment surrounding a multi-rotor flight system during the Skyfall maneuver and to generate aerodynamic data products that can be integrated into higher-fidelity rotorcraft analysis and flight dynamics tools. Virtual-disk rotor representations are employed to efficiently capture system-scale wake interactions and rotor-induced flow phenomena while being computationally feasible across a broad range of operating conditions and con-ops. The models are used to investigate aerodynamic effects relevant to rotorcraft deployment and autonomous flight, including rotor wake interactions, inflow nonuniformity, flow-field disturbances from the mid-air deployment system, and the development of transient aerodynamic keep-out regions. CFD-derived velocity fields and inflow distributions are processed into reduced-order representations suitable for integration with helicopter flight simulation environments and GNC analyses. Within the Skyfall multi-fidelity modeling framework, the system CFD model provides the aerodynamic context required to connect vehicle-scale flow phenomena with mission-level flight conditions as seen by the helicopters. Representative results are presented demonstrating the generation of deployment-environment flow maps, rotor inflow distributions, and aerodynamic disturbance metrics under Mars atmospheric conditions. The analysis highlights regions of elevated aerodynamic interaction and illustrates how system-level CFD can be used to identify operational boundaries and support aerodynamic risk retirement. The resulting methodology provides a scalable approach for incorporating high-fidelity aerodynamic simulations with flight dynamics and GNC models and supports the development of future planetary aerial systems operating in complex atmospheric environments.

Planetary surface reconstruction is a key capability for autonomous missions supporting precision landing, hazard assessment, surface mobility, and scientific exploration. Traditionally, Digital Elevation Models (DEMs) are generated from orbital stereo imagery, enabling large-scale topographic mapping of planetary bodies. However, these approaches rely on dedicated stereo imaging systems and orbital infrastructure, which may be unavailable for resource-constrained missions or newly explored targets. In addition, the spatial resolution of orbital reconstructions is limited. A cost-effective alternative is to reconstruct terrain from imagery captured during the entry, descent, and landing (EDL) phase. Such imagery can bridge the spatial-resolution gap between orbital observations and local surface measurements, while also enabling the extraction of new topographic information from archival planetary mission datasets. Classical Structure-from-Motion (SfM) and Multi-View Stereo (MVS) methods, for example MicMac and Metashape, have been widely used for planetary surface reconstruction, but their performance often degrades under imaging conditions that commonly arise in the descent phase. Recent machine learning (ML)-based approaches, such as Neural Radiance Fields (NeRFs) and Gaussian Splatting, enable high-quality dense 3D reconstruction and realistic appearance model- ing. However, planetary descent imagery remains particularly challenging due to strong fisheye distortion, limited viewpoint diversity, and weak parallax, all of which increase depth ambiguity. Moreover, these methods typically provide little information about the reliability of the reconstructed geometry. For safety-critical tasks such as autonomous landing and terrain-relative navigation, uncertainty quantification is therefore as important as reconstruction accuracy. This work introduces a 3D probabilistic framework for planetary surface reconstruction using descent imagery. By combining ML-based 3D reconstruction with variational Bayesian inference, the proposed approach generates uncertainty-aware DEMs. Existing approaches typically generate a single average surface reconstruction from the observed imagery, providing no estimate of the associated uncertainty or of alternative surface configurations that are consistent with the imagery. This limitation is critical in safety-sensitive applications. Instead, our method models both surface elevation and scene appearance as posterior distributions conditioned on the observed imagery, enabling uncertainty quantification alongside 3D reconstruction. This is the first application of uncertainty-aware probabilistic reconstruction to planetary surface modelling. Specifically, each point on surface is represented by latent variables corresponding to its elevation and RGB colour channels. We investigate three variational posterior approximations. The first employs a mean-field approximation, in which the elevation and colour variables are modelled as independent distributions for each point. The second uses a flow-based posterior to jointly model the four variables (elevation and RGB) associated with each point, capturing their mutual dependencies. The third extends this formulation by jointly modelling all latent variables across the entire surface, enabling dependencies both within individual points and between neighbouring points. The proposed probabilistic approaches are evaluated against both conventional photogrammetry techniques and modern ML-based reconstruction methods using simulated lunar and Martian descent sequences derived from high-fidelity terrain models. Experimental results demonstrate its ability to jointly reconstruct detailed surface geometry and scene appearance while providing spatially varying estimates of reconstruction uncertainty, enabling confidence-aware terrain assessment for autonomous planetary landing applications.

Indian Space Research Organisation, ISRO is planning a lunar lander mission to study the lunar volatile materials and lunar water in the vicinity of a permanently shadowed region in the lunar South pole. A bi-propellant propulsion system is proposed for the propulsion. The propellants, fuel and oxidiser, are to be stored in four identical 1025L tanks. With these tanks being partially filled, study of propellant sloshing occurring in it due to external disturbances during launch vehicle and landing phases is of paramount importance from control stability point of view. Slosh behaviour in propellant tanks is mathematically represented as pendulum mass model for control stability studies. Various slosh parameters are essentially required for design of a robust control system during lander operation. Numerical and experimental methods are used to evaluate these slosh parameters for different fill fractions in the tank. In numerical method, a finite element-based slosh code is used whereas a model tank is used for experimental studies by using principle of dimensional similitude. Reynolds model and Froude models are used for evaluating these parameters experimentally. Based on the control stability studies, the damping available, ie. only the viscous damping, is not adequate during landing phase. The additional damping requirement is finalised based on these studies and it can be achieved by introducing a suitable anti-slosh baffle system. A retention cone and propellant acquisition system are provided in the lower hemispherical dome in all the tanks to supply propellants to the thrusters even in low gravity and retardation conditions for the propulsion requirement. A feasibility study has been carried out to evaluate damping contributed by the retention cone and propellant acquisition system. Numerical slosh analysis carried out considering the retention cone and propellant acquisition system. But numerical studies cannot reveal information about critical parameters like slosh pendulum angle, saturation pendulum angle and slosh damping provided in presence of retention cone and propellant acquisition system. Slosh experiments are conducted in a scaled-down model tank in which retention cone and propellant acquisition system are modelled. Experimentally evaluated saturation pendulum angle, which is a critical parameter determines the occurrence of two-phase flow during landing phase on lunar surface and propellant mass budget. From the experimental studies, slosh damping offered by retention cone and propellant acquisition system is evaluated for different fill levels in the vicinity of landing phase. It is observed that retention cone and propellant acquisition system together offer significant influence on damping and other slosh parameters. Slosh disturbances are reduced in presence of retention cone and propellant acquisition system. Damping contributed by retention cone and propellant acquisition system is adequate for control stability requirements and can manage without any anti-slosh baffles, by which considerable mass can be saved and contribute for the mission requirements. This work is a key milestone for the optimised design for the lunar lander mission.

High-performance parachute systems present significant complexity, cost, and risk. This paper investigates a robust tensegrity platform as a parachute-free alternative for surface payload delivery. Tensegrity structures, which consist of compression struts suspended within a network of tension members, can rapidly redistribute impact loads to protect internal payloads. Their open, web-like geometry also generates high aerodynamic drag relative to the apparent cross-section, resulting in a low terminal velocity, though these dynamics are not well characterized. This work describes experiments to quantify impact mitigation and aerodynamic performance: instrumented prototypes are deployed in free-fall tests from high-rise structures and an unmanned aerial platform, capturing payload accelerations, structural strains, and aerodynamic forces. This effort builds on prior >100 km drop tests from a suborbital rocket that demonstrated system survivability and avionics protection but were limited by low-resolution data. The higher-fidelity measurements acquired here are used to validate simulations and support the development of future parachute-free airdrop and reentry concepts.

NASA is developing a robotic excavator known as the Infrastructure Pilot Excavator (IPEx), a 40 kg-class vehicle that seeks to demonstrate robotic collection, transportation, and emplacement of up to 20 metric tons of lunar regolith over a 9-day period. IPEx’s uniquely designed hollow excavation mechanisms, known as bucket drums, are designed with staggered scoops around its circumference and an internal baffle for collecting regolith. Once collected, the material can be transported to any location and dispensed by reversing the drum’s direction. Two pairs of bucket drums on IPEx dig simultaneously in opposing directions, resulting in counter-acting excavation forces that enable this low mass robotic excavator to effectively dig in reduced gravity environments. This is a significant departure from terrestrial excavators that rely on the tractive forces produced by high mass to counteract the forces of excavation. Extensive experimentation of the IPEx’s bucket drum behavior was conducted at Kennedy Space Center (KSC) to quantify the flow of regolith and the efficiency of the bucket drum’s baffle during collection and deposit of excavated material under vacuum conditions and 1g. In February 2025, the IPEx team flew two payloads called the IPEx Bucket Drum Flow experiment aboard a Blue Origin New Shepard suborbital vehicle configured to spin at 11 RPM to induce lunar gravity levels. The experiments each consist of a custom-made small vacuum enclosure that houses three independently actuated rotating cylinders representative of scaled down bucket drum geometries. Each bucket drum analog is fully enclosed with clear polycarbonate side walls and contains one of three different grain size distributions (fine, mid, or course) of lunar regolith simulant or glass beads. Three cameras were mounted above the clear polycarbonate lid of the vacuum enclosure to capture real time video image data of the regolith-baffle flow. Upon confirmation of lunar-g, video of the regolith flow was recorded and post processed to measure the flow rate and any jamming effects. Multiple internal geometries for the bucket drums and regolith simulant size distributions were tested. To account for the entrance/exit of regolith from the bucket drums, a “bubble” volume was protruded from each scoop. These “bubble” volumes also allow for repeated trials, as all regolith is retained and the system can be “rewound”. After the successful flight of NS-29, the IPEx team received both payloads back at the Blue Origin payload processing lab. Visual inspections of the payloads were completed, and no visible damage or anomalies were found. Upon thorough review and post processing, there were three key findings and several design lessons learned. The volumetric flow rate was not significantly decreased, no regolith jamming or bridging was observed, and an average 20% reduction in regolith density was observed. This publication will further detail the design of the bucket drum flow experiment, the variables under test, and the results from the flight. This information will support future mechanisms that will interact with lunar regolith and payload developers for future suborbital experiments.

Subsurface access on planetary bodies is a prerequisite for in-situ resource utilization (ISRU), which is a key enabling technology for a sustainable human presence on the Moon. While previous Mars rovers have successfully demonstrated the value of in-situ rock drilling and coring, those systems rely on large turret assemblies and complex sample-handling chains that are difficult to accommodate within strict mass and volume constraints. Consequently, lightweight and low-cost rover platforms increasingly targeted for future planetary missions often lack the capacity for such instrumentation. To support these constrained platforms, we present a prototype of a lightweight, modular drilling and sample collection tool designed for ~40 kg rovers with limited payload capabilities. The tool consists of three modular components: a drill, a sample container, and a payload box. Each payload box stores up to two soil containers. The system is operated by a single robotic manipulator using a standardized docking interface. The operational workflow proceeds as follows: the manipulator first mounts the drill and retrieves an empty container from the payload box. The drill then penetrates the terrain through a central aperture in the container's base, capturing soil samples that fall directly into the container upon drill retrieval. Once the drilling process is complete, the manipulator detaches the filled container and stores it back into the payload box. Finally, the rover detaches the drill and mounts the payload box to the manipulator for transport. This design offers four primary contributions. First, the tool demonstrates efficient modularity: by standardizing the docking interface, a single manipulator can be reconfigured to perform diverse tasks, including drilling, container management, and payload transport. This eliminates the need for multiple robotic arms or a large, complex turret, thereby reducing system complexity. Additionally, the tool can be tested independently as a standalone module. Second, the tool is lightweight (<2 kg), making it suitable for resource-constrained rovers. Third, the scalable sampling mechanism enables cumulative soil sample collection through repeated drill-and-pull cycles, potentially filling the container to its capacity of a few hundred cubic centimeters (adaptable to mission requirements). Fourth, both the tool and its sampling mechanism are simple; the tool comprises only a few modular parts, and the passive sample-capturing process contributes to reduce failure risks. We experimentally evaluate this prototype using our lightweight LRU2 rover at a dedicated Moon-Mars analog test site. A reactive torque-control algorithm guides the drilling process, adjusting axial feed based on torque feedback to ensure consistent operation across varying terrain profiles. Our tests successfully demonstrate sample extraction, validating force-driven manipulation for constrained platforms. The samples serve in-situ resource utilization: characterizing the subsurface resources that a sustainable presence on the Moon would mine and process. We conclude with practical lessons learned and outline directions for future design improvements.

Propulsive drones are a promising new technology for generating high-resolution terrain models of the Moon. In future lunar missions, propulsive drones will complete a series of hops across the lunar surface while capturing images with downward-facing cameras. We present an autonomous motion planning framework that uses a vision-guided optimal control strategy for performing safe, efficient hops between specified waypoints. The drone plans parabolic hops between waypoints generated by a global-level planner; if a nominal waypoint is unsafe or unreachable, the drone then completes a diversion maneuver to safely land. To evaluate the validity of potential landing sites, we develop a custom semantic segmentation model for identifying obstacles from aerial images, and we introduce an approach for rapidly calculating reachable terrain. We then use model predictive control (MPC) to robustly track hop trajectories that guide the drone towards safe landing sites. We test the optimal hop planner in a high-fidelity lunar simulation environment. Our results demonstrate that our unified motion planning framework yields safe, fuel-efficient hop trajectories which can be tracked rapidly and with low error via MPC.

NASA is developing the Infrastructure Pilot Excavator (IPEx) to excavate lunar regolith and construct berms at the lunar South Pole. IPEx is a high-uptime, 40 kg-class robot whose mission energy demand exceeds the practical limits of onboard energy storage, solar recharge, or radioisotope power. The mission concept addresses this through repeated recharge from surface infrastructure, with the lander providing the charging interface. While previous extraterrestrial rovers have relied on onboard power sources, no extraterrestrial rover has demonstrated return- to-lander docking for battery recharge on the surface. Throughout its mission, IPEx will recharge over 30 times at 180 W of power. The power delivered to IPEx by the lander during each of these charge sessions is a key driver of performance and directly affects how much regolith IPEx can excavate during its mission. This recharge requirement motivates the need for a novel contact-based charger design that is highly efficient, dust-tolerant, and tolerant to misalignment in multiple degrees of freedom. The charger design leverages IPEx’s unique modalities of motion, including the ability to lift its chassis off the ground using its arms. This paper presents the charger design process from initial concepts through lunar surface analog testing. Three competing designs were evaluated before down- selection to a single design. The selected design underwent several rounds of optimization and testing. This test campaign utilized several capabilities of the Kennedy Space Center’s Granular Mechanics and Regolith Operations Lab, including a dusty thermal vacuum chamber and representative robotic excavators.

This work presents a modular octahedral tensegrity robot for extraterrestrial exploration that combines impulsive tilting locomotion with basic manipulation and payload transport capabilities. The system builds on the mechanical advantages of Tensegrity structures, in particular their ability to distribute loads efficiently while remaining lightweight and inherently compliant. These properties make them attractive for planetary environments where impacts, irregular terrain, and uncertain landing conditions are expected. A major motivation for this work is prior research in tensegrity space robotics, most notably NASA’s SuperBall Bot, which demonstrated that highly deformable structures can survive rough landings and still maintain mobility. The approach presented here extends this idea by not only focusing on locomotion, but also on how the structure itself can be used for simple interaction tasks such as grasping and transporting small objects. Unlike conventional planetary rovers based on wheels or articulated legs, the proposed system does not rely on continuous contact or complex kinematic chains. Instead, motion is achieved through bistable actuation units placed at the frame nodes, which store elastic energy and release it in a controlled snap-through event. This results in an impulsive torque that drives the robot into a tilting motion between stable equilibrium configurations. The choice of an octahedral geometry is driven by its balanced tipping behavior and relatively simple structural layout. Each reorientation step allows movement in multiple directions without requiring additional degrees of freedom. At the same time, the tensegrity frame serves as both a load-bearing structure and a passive damping system, which helps absorb landing shocks and reduces the need for dedicated landing hardware. This makes direct surface deployment feasible, even in scenarios where a separate lander would typically be required. Beyond mobility, the structure can be actively deformed to partially enclose objects, enabling basic grasping and transport of samples. This is achieved by locally adjusting the pretension of one tensioned member, effectively turning the robot into a compliant, reconfigurable structure without adding dedicated manipulators. The modular nature of the design allows multiple units to be combined into larger assemblies. Such configurations could increase payload capacity, improve stability, or enable cooperative transport strategies. The same modularity also provides robustness, as individual units can fail without completely disabling the system. Overall, the presented concept explores a lightweight and mechanically simple alternative for planetary robotics, where mobility, impact tolerance, and functional adaptability are combined in a single structural framework. It is particularly relevant for environments such as lunar caves, asteroid surfaces, or Martian terrain, where conventional robotic platforms often face limitations in both mobility and survivability.

Asteroids and dwarf planets have been frequent targets of robotic exploration missions over the past decades. Their low gravity, lack of atmosphere, and regolith-covered surfaces impose requirements on locomotion systems fundamentally different from those encountered on Earth, demanding paradigms that may be infeasible terrestrially yet excel on such bodies. Under low gravity, even small vertical contact forces render an exploration system ballistic. While this effect can be exploited to cover large distances, it results in uncontrolled tumbling and precludes a precise approach to points of interest. Since frictional forces are likewise reduced, sliding along the surface constitutes a viable alternative. Sliding locomotion can complement ballistic locomotion, enabling a precise final approach on locally flat terrain after landing near a point of interest. This paper studies a locomotion system that slides across flat, moderately compliant surfaces by actuating an internal mass, thereby generating reaction forces on the robot body. Stiction is exploited by accelerating the internal mass slowly and braking it abruptly, which induces forward motion of the robot. Since the actuation is parallel to the contact plane, vertical reaction forces remain small and ground contact is maintained. The fully enclosed actuation prevents regolith from jamming mechanical components, yielding a design of low complexity and high robustness. While an internal mass can equally drive a rolling cylindrical or spherical system, a sliding platform maintains a defined payload orientation and a stable resting pose without continuous attitude regulation. The control law is formulated as a model predictive controller based on a piecewise-smooth system description that detects when stiction is overcome and sliding is initiated. Enabling the controller to autonomously discover the switching sequence of the sliding modes yields robust and energy-efficient translation. Constraints keep the overturning moment below the restoring moment to prevent tipping and bound the velocity by the tangential escape velocity. The modeled dynamics, however, are highly sensitive to the contact friction coefficient. Uncertainty in its estimate causes the controller either to fail to escape stiction or to induce uncontrolled oscillatory motion. Since exploration robots are frequently the first sensing devices to touch the surface of a celestial body, this coefficient is unknown a priori and may vary as the robot traverses regions of differing soil composition. This paper introduces a model identification approach that estimates the contact friction coefficient online. Employing a piecewise-smooth model analogous to that of the controller, a moving horizon estimator is formulated that exploits the complete envelope of the nonsmooth dynamics. The complete framework is validated in simulation with high-fidelity contact dynamics, including nonlinear stiction and a compliant surface, where ground-truth friction coefficients enable a quantitative assessment of the identification performance. A goal-reaching task traversing regions of varying friction, emulating changing soil conditions, evaluates the framework's robustness to frequent parameter variations. The experiments span gravitational accelerations representative of celestial bodies targeted by recent exploration missions, as well as variations in system mass and contact-plane dimensions, delineating the surface conditions and mission scenarios for which the approach is applicable.

The Perseverance Mars Rover is quickly approaching the Opportunity Mars Rover’s interplanetary distance traveled record of 45.16 km, as established in 2019 and published in [Verma et al, Science Robotics 2023]. But how is that distance measured? In orbital images, by commanded wheel motion, using in situ knowledge of vehicle slip as measured by Visual Odometry flight software? Perseverance and Curiosity Mars rover flight software systems maintain over a dozen derived odometers (each tracking distance using five distinct metrics), and their operations teams track overall distances employing still more strategies to estimate mechanism lifetime and provide positional information for science target localization and strategic drive planning. In this paper we document the multiple strategies used on NASA Mars Rover missions for measuring rover progress and estimating position uncertainty and explain their relative benefits, both to onboard realtime mobility behaviors and also to Earth-based tracking of resources and mobility progress. The design of future Lunar and other planetary rovers’ flight software and ground analysis tools can hopefully benefit from these observations.

Monocular visual odometry on resource-constrained platforms such as UAVs, ground rovers and orbital inspection vehicles must deliver metric pose, yet imagery on such platforms arrives at a cadence set by power and downlink budgets rather than by the algorithm. Learning-based odometry typically absorbs the cadence of its training corpus, and monocular systems recover scale only up to a factor fitted to ground truth at evaluation time. We present SAGE-VO, a sparse geometric egomotion solver in which learning never estimates the pose. A frozen front end supplies correspondences and samples a zero-shot metric depth prior at them; a 34.6 k-parameter network scores those correspondences; the pose is initialised in closed form and refined by iteratively reweighted Gauss–Newton under those scores; and the translation magnitude is voted from the depth ratio log‖t‖ = log z_metric − log z_rel. No component holds temporal state, and no map, bundle adjustment or test-time alignment is used. On held-out KITTI sequences SAGE-VO reduces translation error by 37 % and 25 % against the same pipeline with its learned component removed, and by 46 % and 51 % at 2 Hz, reporting absolute trajectory error unaligned — within 1–5 % of its similarity-fitted value, confirming metric scale. A single model spans 10–2 Hz and outperforms per-rate specialists. The magnitude is structurally insensitive to focal error, moving 2.2 % across a ±5 % sweep against rotation's 20.6 %, so surrendering the intrinsics costs 5 % of translation error.

Wind-driven spherical rovers can provide a low-power method for long-duration persistent-missions in remote environments by using available wind for propulsion. The Spherical-Sailing-Omnidirectional-Rover (SSailOR) is a wind-driven rover concept that uses sails as lifting-surfaces to generate aerodynamic forces and achieve motion with an upwind-component. Along with terrestrial remote-surveillance applications, this concept will also be useful for planetary exploration where environmental winds can support rover-mobility. Titan is considered as one possible planetary application because of its dense-atmosphere and low surface-gravity. The purpose of this work is to develop a model which can predict the upwind feasibility, validated with Earth-based experiments. The validated model is further used to show upwind feasibility under different terrestrial and planetary environmental conditions specifically for Titan in this study. This paper presents development of a custom wind-chamber, a numerical-model for predicting upwind-motion, and experimental demonstration of upwind-mobility using a 0.4m-diameter SSailOR prototype. The wind-chamber was developed to conduct rover’s translation and transient-motion studies. The developed suction-type wind-chamber has overall dimensions of 8ft×7ft×2ft, with a 6ft×6ft×2ft test section and three axial-fans, each having rated free airflow of approximately 9000-CFM. The chamber produced wind speeds of approximately 3m/s. The wind flow inside the chamber was characterized using Pitot-tube measurements at approximately 4000-spatial locations. Two wind-chamber configurations were tested, giving mean wind-speeds of approximately 3.0m/s and 3.2m/s and standard deviation of 0.377m/s and 0.800m/s, respectively. Upwind feasibility was predicted using a dynamic model of SSailOR integrated with a reduced-order aerodynamic-model. The aerodynamic forces were calculated using identified-coefficients from previous wind-tunnel testing of the rover. The main model inputs were rover heading, main-sail angle of attack, jib-sail angle of attack, and local wind speed. The model calculates the net torque about the rover center due to aerodynamic and contact forces, where a positive net torque indicates feasible upwind motion. Rover heading was constrained by the rail and wheel carriage, while the sail angles were fixed for each test-case. Approximately 500 test-cases were conducted for each wind-chamber configuration for headings ranging from 65° to 87.5°, with main-sail and jib-sail angles of attack ranging from 10° to 24°. The local wind speed for each test condition was obtained by interpolation of the measured wind-profile. The experiments provided the first physical demonstration of upwind motion for SSailOR. The experiments have shown good agreement with the prediction model for upwind motion, validating the model. The validated model was further implemented for Titan environmental conditions to demonstrate SSailOR’s upwind capability for planetary studies. Simulations were conducted for 0.4meter-diameter, 1meter-diameter, and 2meter-diameter rover configurations with masses of 1.1kg, 6.25kg, and 20kg, respectively. Out of 500 cases, upwind motion was predicted for 26 cases for 0.4meter-rover, 66 cases for 1meter-rover, and 205 cases for 2meter-rover. The feasible upwind-heading range increases with rover-scale, with minimum-heading 82.5° for 0.4meter-rover, 77.5° for 1meter-rover and 70° for 2meter-rover, showing SSailOR has potential to be able to operate effectively on Titan. These results show the experimentally validated model can be used to study rover-sizing, operation-conditions, and upwind-feasibility for both terrestrial and planetary applications.

Hardware validation of lunar legged-locomotion controllers requires testing under reduced effective gravity and on representative terrain, yet facility-scale planetary analog environments are poorly suited to the rapid iteration required during controller development. We present a portable reduced-gravity locomotion testbed intended to provide an intermediate testing stage between simulation and large analog campaigns. The system combines a software-commandable active gravity-offload winch with an automatically tracked single-axis carriage and a tipping granular-terrain bed for flat and inclined locomotion experiments. The offload system is separable from the terrain module and can be installed over other test environments, while software-commandable force and tracking parameters allow different effective-gravity settings to be tested without hardware changes. The gravity-offload system was characterized using an in-line six-axis force/torque sensor on a 34 kg prototype lunar quadruped. At the 288 N lunar setpoint, the measured force at rest was 289 ± 13 N. During flat-terrain walking, the mean delivered force was 8% below the command, with a standard deviation of 23 N. Parasitic horizontal force associated with carriage tracking was approximately 10 N RMS and could be estimated from the measured tether angle without the external force/torque sensor. Dynamic experiments indicate that force variation during nominal locomotion is dominated by friction in the winch and tether path, while reflected actuator inertia becomes important during rapid vertical motion. The integrated testbed has supported repeated reduced-gravity locomotion testing on flat and inclined granular terrain, and the offload hardware was deployed at the LUNA lunar analog facility. With the carriage unpowered, the same system can additionally provide a drawbar-pull-like measurement of the horizontal traction sustained by the robot.

A proposed Mars aerosol release process experiment would disperse submicron particles into Mars’ planetary boundary layer (PBL), and track them in order to study near surface atmospheric transport, dust storm initiation, aerosol dispersion relevant to planetary protection, and the feasibility of warming Mars, in support of terraforming research. Within this Mars surface payload, that includes the particle handling unit (PHU) together with remote and in-situ diagnostics instruments, the PHU is at the lowest technology readiness level (TRL). The PHU must store, meter, deagglomerate, and disperse submicron particles while meeting size, power, contamination, and gas consumption constraints of a Mars surface payload. This paper presents the technology development approach and test results for the PHU. Many laboratory systems exist that release either small flow rates of monodisperse nanoparticles or substantial flow rates of polydisperse nanoparticles, but, to the authors’ knowledge, no existing aerosol generation system simultaneously combines high particle mass flow rate, a narrow particle size distribution, and predominantly submicron output. The key challenge lies in developing and demonstrating a system that can do both high throughput release and narrow particle size distribution, while meeting the constraints of a Mars surface payload. The development approach presented combines analytical sizing, system level breadboarding, and testing with surrogate powders and aerosol diagnostic tools. The integrated PHU has demonstrated controlled dry powder mass flow rates above 5 g/s. Particle size distribution measurements show a clear sensitivity to pneumatic operating conditions: at a constant particle flow rate of 2.1 g/s, increasing gas flow reduced the measured D50 from approximately 4.3 μm to 3.0 μm, while the finest distribution measured to date had a D50 of approximately 1.5 μm at a particle flow rate of 4.2 g/s. A coarse agglomerate population remains above the final D90 requirement. The test campaign will progress from ambient field operation toward chamber testing at Mars atmospheric pressure. The resulting particle handling architecture, nanoparticle properties, preliminary breadboard/component results, dominant risks, constraints, and a maturation plan for advancing the subsystem from TRL 3 toward TRL 6 are presented.

Establishing a persistent presence on the moon requires the construction of safe, reliable infrastructure under uniquely challenging environmental and geotechnical conditions. Lunar infrastructure must withstand extreme temperature fluctuations, radiation exposure, dust hazards, and seismic loading from moonquakes while maintaining long-term structural integrity. At the scales required for persistent infrastructure, astronaut-led construction is impractical because of the cost, risk, and logistical burden of sustained human activity on the lunar surface. Autonomous robotic construction, therefore, represents a critical enabling technology for future lunar missions. While NASA and industry have advanced robotic construction and site preparation technologies, the development of anchoring solutions that can be installed autonomously while maintaining structural performance remains a critical challenge. Variations in lunar regolith composition and mechanical properties can influence anchor behavior, requiring designs that remain effective across a range of site conditions. Furthermore, anchoring systems must tolerate both environmental loading, including moonquakes and thermal cycling, and operational loads introduced during robotic construction and subsequent infrastructure use. Robust leveling and anchoring systems are essential for ensuring the stability and long-term performance of safety-critical structures such as habitats, towers, power systems, and other surface construction. In this paper, we investigate an actuated leveling concept designed for integration with modular robotic construction systems. This work builds on existing systems for autonomous off-world assembly, including the ARMADAS (Automated Reconfigurable Mission Adaptive Digital Assembly Systems) project at NASA Ames Research Center, which uses teams of autonomous robots to assemble large-scale lattice structures from modular unit-cell building blocks. While prior work focused primarily on robotic assembly of structural components, we extend this framework by developing compatible leveling solutions. Through iterative design, prototyping, and testing, we evaluate a functional leveling mechanism tailored to the operational constraints of autonomous robotic deployment. In this approach, footer voxels are connected in a centralized system and controlled to self-level, with control algorithms theoretically extended toward a fully distributed architecture. Balancing structural robustness with efficient robotic deployment, this work provides a practical surface-interfacing framework that advances scalable autonomous construction and supports the development of a sustained human presence on the moon.

Sustained lunar operations require lightweight and flexible power distribution between spatially separated generation and loads. This paper proposes wireless power transfer (WPT) in which processed lunar regolith acts as a dielectric transmission path that couples an electric near field excited by resonant electrodes. In contrast to conventional antenna-to-antenna WPT, the concept is intended to turn a road or work area into a powered zone, i.e., a “surface-as-an-antenna” physical layer. In prior experiments, sand with a relative permittivity of approximately 3 increased average transfer efficiency by about a factor of four compared with the no-sand condition, and a batteryless rover carrying a 100 mm × 100 mm receiver antenna moved continuously over a 500 mm × 1500 mm sand bed. In the present study, complex-permittivity measurements of consolidated FJS-1 and LHS-1 simulants showed relative permittivity values of up to approximately 10 for dense, melted-and-solidified specimens at 10 and 100 MHz. These measurements were used to examine trends associated with material processing. HFSS simulations then identify a maximum transfer efficiency of approximately 60% to a receiver 3 m away for an idealized path with relative permittivity 10 and loss tangent of 10^-5. These results do not yet establish a flight-ready system but support feasibility of an innovative in-situ resource utilization (ISRU) approach that transforms a specific area of the lunar surface itself into a wide power-distribution network. Accurate loss-tangent measurements, high-voltage testing in the lunar environment, fabrication-tolerance assessment, thermal design, and scaled demonstrations remain necessary. The proposed method is positioned as a complementary local-distribution layer rather than a replacement for long-distance DC or beamed links. This positioning also defines a practical path for staged validation in representative lunar conditions.

Model-based systems engineering (MBSE) toolchains and artificial intelligence (AI) assistants are increasingly introduced into early mission design to improve traceability, interdisciplinary coordination, and design iteration, but evidence of how far their intended workflows are realized in short concurrent-design settings remains limited. This paper assesses an AI-augmented MBSE toolchain used during the 2026 Space Station Design Workshop (SSDW), in which two independent teams of twenty students and young professionals each had six days to develop a lunar south-pole in-situ resource utilization (ISRU) infrastructure concept producing at least 20 t of propellant per month by 2045. Both teams received the same mission statement, toolchain, and AI assistance. The assessment measures the requirements exports, SysML v2 structural models, and parametric workflow archives delivered at fixed review deadlines, and uses a post-workshop survey of nine of the 40 participants and one follow-up interview to interpret them. The teams developed distinct architectures that converged on polar water-ice extraction and combined fission and photovoltaic power. The delivered models showed limited end-to-end integration. Both teams produced complete requirement hierarchies, but no more than 3 of 84 and 22 of 142 requirements were linked to architecture elements, verification methods were recorded for at most 5 % of requirements, and neither model encoded requirement values in machine-readable fields. Neither team delivered a parametric model in which a change propagated across subsystem boundaries to a system-level output. One team built an integration framework around subsystem templates without calculations, whereas the other developed subsystem calculations without connecting them into a system model. The only reported integrated budget was kept in a spreadsheet, and both teams' central ISRU analyses remained outside the parametric environment. Toolchain operation was concentrated among a few participants, and median System Usability Scale scores of 56 for the requirements platform and 39 for the parametric platform fell below the commonly cited average of 68. Reported AI use centered on research and drafting rather than on integration. Three conditions were associated with this partial realization: front-loaded learning and integration cost, continued architectural change during model development, and incomplete connections to discipline-specific analysis tools. The paper contributes artifact-derived measures for evaluating such toolchains and six testable implementation hypotheses for time-constrained concurrent design.

In December of 2024, NASA’s Mars Exploration Program (MEP) released a high-level plan for a sustainable science program at Mars, covering the years 2024-2044. Supporting that, in FY25 MEP started technology investments meant to enable elements of that plan. These investments included 25 projects grouped across six broad areas: (1) entry, descent, and landing (EDL), (2) mobility, (3) avionics, autonomy, and power, (4) drilling and sample handling, (5) science instruments and planetary protection, and (6) telecommunications. Most of these projects are now complete and will all be complete at the end of FY26. Some results from these investments have already been infused into new flight projects, particularly Skyfall for Mars and Moonfall for Earth’s moon. This paper summarizes these projects, their results, technology transitions achieved to date, and goals of follow-on work now underway. Due to space limitations, this abstract highlights a subset of those. In EDL, work reached TRL 5 in development of a woven thermal protection system (TPS) called MERINO-LD, which is expected to reduce the cost of forebody TPS by approximately 70% compared to C-PICA. Another project developed a prototype of a small thruster throttle valve (STTV), designed to mate with the MR-107 thruster used on the Mars Phoenix and InSight landers, and did a successful water flow test to establish TRL 5. This valve is expected to be applicable to potential future Mars soft landers with landed mass from approximately 35 to 500 kg. In mobility, experimental characterization was done of Mars helicopter rotors with high solidity (0.3) and blade tips that transiently reached supersonic velocity (Mach 1.08) due to gusts. The high solidity test was essential to provide confidence in aerodynamic performance of rotors for potential future Mars helicopters with payloads in the range of 5 kg; the supersonic tip test allows flying with higher RPM to produce more thrust, without concerns about catastrophic effects from transient supersonic tip velocities. In power, commercially available battery cells were tested to identify higher performance cells for potential future Mars helicopter missions than those used on Ingenuity. This included several cylindrical cells in the 21700 format and several pouch cells with varying amounts of silicon in the anode. Cylindrical cell testing included using custom electrolytes to enable charging at lower temperature. This work validated cells with > 1.5x the specific energy and 2x the specific power of Ingenuity cells. In planetary protection, a new workflow was developed and validated for metagenomic characterization of bioburden on spacecraft before launch, enabling quantification of known taxa in 4 hours and comprehensive detection of DNA present in 48 hours, which is a huge step beyond the NASA Standard Assay used to date. In telecommunications, a family of miniaturized UHF radios called Arke was developed to enable future Mars helicopters to communicate directly with relay orbiters, instead of through a nearby rover or lander like Ingenuity did, with printed circuit cards that are under 80 grams. The Skyfall and Moonfall missions in development are baselining versions of this radio.

Various life detection missions are due to take place within the next few decades, targeting the inner solar system with planet Venus and the outer one with the icy moons of Jupiter and Saturn. The current gold standard for organics detection, GC-MS, and other novel techniques are currently under consideration for landed missions on the icy moons with the aim to detect and identify organic molecules related to life as we know it. In this contribution, we present the current measurement capabilities of our laser-based mass spectrometer, as well as technological advances in sample handling of snow and ice samples allowing for the direct detection of organic molecules. The deployed mass spectrometer system consists of a compact time-of-flight mass analyzer (currently in preparation for the Blue Ghost Mission 4 to the lunar surface) connected to a passively Q-switched microchip laser operated at UV irradiation (λ = 266 nm, laser pulse repetition rate of nom. 100 Hz, few µJ per laser pulse only). The system has been deployed recently on the Aletsch Glacier, Switzerland, a test bed for the exploration of the icy moons. Various snow, collected from a 1 m deep pit, and ice materials from a glacier crevasse were collected. The samples were placed in a sample preparation vacuum chamber, allowing the water content of the samples to sublimate under low pressure, resulting in an organic film. The film was subsequently transferred to and analyzed in the main vacuum chamber in which the mass analyzer is operated. The measurement performance of the overall system in combination with the sublimation chamber will be presented in detail, with a special focus on the environmental and technological challenges posed by this extreme environment. Operating flight-like systems and measurement protocols within icy environments mimicking a subset of parameters as observed on planetary objects is the next logical step in preparing for future space exploration missions to the icy moons.

The next era of lunar exploration demands more than isolated sample return missions: scalable, preservation-optimized architecture designed for sustained operations and Earth return integration. As the Moon Base establishes a long-duration presence and robotic sample collection matures, lunar science infrastructure must evolve into a capability that maximizes the scientific return of these investments. FOSSILS with AMBER (Framework for Optimization of Sampling Systems for Integrated Lunar Science with Archive of Material, Banked for Earth-based Research) is a modular, extensible architecture concept standardizing lunar sample acquisition, preservation, and handling across multiple surface actors. It integrates standardized digital and mechanical interfaces while leveraging existing and planned NASA surface assets, including the PROMISE rover, a second-generation VIPER rover, the Lunar Terrain Vehicle (LTV), and Moon Base power, to reduce development costs and accelerate deployment. At the core is a standardized containment hierarchy supporting geological cores, volatile-bearing regolith, loose surface material, and human biomedical samples in one unified system. The Universal Sample Container (USC) and companion Human Research Vials (CHRVs) use a metal-to-metal compression hermetic seal, drawn from Mars 2020 heritage. Filled CHRVs slot into USCs, which integrate into pallets sized for surface transport, long term storage, and packing for Earth return vehicles, including the contracted Lunar Freezer System for Orion. Long-duration surface preservation is handled by AMBER lockers, providing unconditioned, refrigerated, and frozen storage to protect sample integrity, co-located with the Moon Base for crew access and shared power. A digital Sample Procurement Portal links terrestrial science priorities to surface acquisition. Researchers submit retrieval requests specifying location, preservation state, and material characteristics, which are prioritized and manifested to available surface agents for collection. RFID-enabled logging embedded within CHRVs and USCs automates metadata capture, reducing crew time and preserving chain of custody across multiple actors. The concept of operations spans long-range robotic retrieval by PROMISE, PSR volatile and core sampling by a VIPER-derived rover, astronaut-assisted sampling using LTV mobility, and in-habitat biomedical collection at the Moon Base. A Science Add-On module extends this reach further by allowing non-science missions to interface with the containment system and opportunistically acquire samples during commercial prospecting activities, broadening terrain access at marginal cost. Samples flow from request through collection, containment, transport to the Moon Base, and long-term storage in an AMBER locker while surface agents replenish at a Resupply Station before their next voyage out. Preliminary TRL maturation timelines indicate the system could be fully operational no earlier than late 2031, with a development and first year operations cost (100 kg sample capacity) estimated at $ 5.2 billion, including leveraged systems. While this concept relies on the deployment of proposed surface actors, power availability, and temperature management technologies, it highlights an important approach to sustained lunar activity: standardizing interfaces to ensure interoperability, enable collaboration, and eliminate redundant development costs. FOSSILS with AMBER re-frames lunar sample collection and, more broadly, lunar science as a high priority precursor infrastructure integrated with Artemis surface operations rather than siloed into isolated missions, reducing the risk and cost while maximizing long-term scientific yield.

Persistent, scalable power generation is a foundational requirement for sustained human presence and in situ resource utilization (ISRU) at the lunar poles. Solar power is especially attractive at select high-latitude sites that offer near-constant annual illumination, where conventional horizontal solar arrays are severely limited by local topographic shadowing. The NASA Lunar Vertical Solar Array Technology (LVSAT) program was established to address this gap by developing a deployable, mast-mounted solar array capable of operating well above the surface, enabling access to more continuous illumination and supporting networked, mobile power grids in service of Artemis and beyond. This paper presents the design, testing, and TRL 6 verification of Power Tower, a deployable vertical solar array system developed by Honeybee Robotics, a Blue Origin Company, under the LVSAT program. Power Tower is a self-deploying, relocatable vertical solar array system designed to deliver 10 kW end-of-life electrical power from a mast height exceeding 20 meters above the Lunar surface. The system integrates the Deployable Interlocking Actuated Band for Linear Operations (DIABLO) boom: a compact, high-stiffness deployable mast with lightweight, flexible solar panels offering mass efficiency. Supporting subsystems include flight-heritage brushless DC actuators for mast deployment, solar tracking, and auto-leveling; dust-tolerant electrical connectors based on a novel permeable membrane architecture; a four-zone thermal control system designed for the extreme environment of the lunar South Pole; and avionics and S-band communications derived from proven spaceflight hardware. The system is designed for single-lander delivery and rover-assisted relocation. This work advanced Power Tower from to TRL 6 through horizontal and vertical deployment testing, solar array thermal cycling, characterization of mast stiffness and dynamic stability through thermal-structural modeling, explicit dynamics simulation, and hardware-validated tip deflection mitigation. This paper reports on the completion of TRL 6 verification conducted in Thermal Vacuum Chamber A at NASA Johnson Space Center. The test campaign subjected the full engineering unit to thermal and vacuum environments representative of Lunar surface conditions, validating deployment functionality. Results confirm that Power Tower meets LVSAT program requirements, including 10 kW power generation, operability within 2° of the Lunar South Pole, and a 10-year design life. These results mark a critical milestone toward a Lunar flight demonstration and establish Power Tower as a near-term enabling technology for permanent Lunar surface power infrastructure, including the first Lunar-scale solar power microgrid.

Venus's extreme surface conditions (470 degrees Celsius, 90 bars) and dense, global cloud layer severely constrain our ability to perform high-resolution geological and atmospheric investigations. Current orbital near-infrared (NIR) imaging is limited by the clouds to spatial resolutions of approximately 100 kilometers, whereas conventional landers succumb to the thermal environment within hours. While variable-altitude aerobots offer a platform for sustained exploration, they must operate at benign altitudes between 52 and 62 kilometers to survive and maintain solar power generation, leaving them optically shielded from the surface. This paper introduces a novel mission architecture featuring a tethered towbody designed to circumvent these limitations. This architecture deploys a compact payload suspended on a multi-kilometer tether from a high-altitude primary gondola. By lowering the towbody below the main cloud deck to a target altitude of 45 to 47 kilometers, the system achieves unprecedented night-time NIR surface imaging at spatial resolutions approaching 10 meters per pixel over contiguous, wide-area swaths. In addition to surface imaging, the towbody facilitates in-situ atmospheric sampling and meteorological profiling. Operating at 47 kilometers exposes the payload to ambient temperatures near 100 degrees Celsius and relative wind shears over 10 meters per second. Safely operating in this environment necessitates a rigorous feasibility analysis and series of trade studies. The multi-kilometer tether requires careful evaluation of its strength, induced drag, and feasibility of providing power and communication to the gondola. To ensure thermal survivability, this study explores power and cooling trades, contrasting in-situ energy harvesting via an integrated wind turbine against battery-powered phase-change cooling. Furthermore, achieving the attitude stability required for high-resolution imaging drives trade studies on aerodynamic hull designs, passive stabilizers, and active gimbal mechanisms. This paper presents the feasibility study and point-design of a tethered towbody platform for sub-cloud imaging on Venus. Specifically, we focus on foundational engineering trades across four primary research pillars: dynamic modeling of the tether under atmospheric wind shear, aerodynamic stability analysis for the towbody, power and thermal management strategies, and the comprehensive mission concept of operations. By decoupling the survival constraints of the primary spacecraft from the sub-cloud sensing requirements, this tethered platform provides a transformative pathway to revolutionize our understanding of Venusian geologic and atmospheric evolution.

As international space agencies pursue sustained human presence on the Moon, a fundamental challenge is how the infrastructure required to support long-duration surface operations will be constructed. Landing pads, blast berms, equipment shelters, power stations, habitats, and other infrastructure may require construction campaigns extending beyond practical periods of continuous human supervision. Communication latency, limited operator availability, and the need to construct infrastructure before crew arrival therefore motivate robotic systems capable of progressively higher levels of autonomy. This paper formulates the autonomous construction of lunar infrastructure using Regolith Containment Units (RCUs), analogous to terrestrial sandbags, as a multimodal embodied artificial intelligence (AI) problem and establishes a physics-based simulation testbed for its investigation. Recent advances in multimodal robotic policies, including vision-language-action (VLA) models, motivate examining their applicability to autonomous lunar construction. Building upon prior work that conceptualized and system-engineered a mobile-manipulator platform for constructing berms by stacking RCUs, this paper makes four contributions: (1) a task-level analysis identifying where autonomy provides value across a lunar construction campaign; (2) a formal problem definition for autonomous RCU manipulation and placement; (3) a trade study of candidate multimodal autonomy architectures using task-derived evaluation criteria; and (4) a MuJoCo-based simulation testbed for autonomy development and evaluation. The autonomy analysis employs a parametric model that decomposes a construction campaign into constituent tasks and compares teleoperation, supervised autonomy, and full autonomy across ranges of communication latency, operator availability, and operational duty cycle. The RCU manipulation problem is then formalized in terms of perception, manipulation, placement, task sequencing, and safety requirements, including operation in proximity to crew. These requirements provide the basis for evaluating four autonomy architecture classes: end-to-end VLA policies; a vision-language model planner coupled to a low-level skill manager; modular learned skills integrated with classical task planning; and non-language visuomotor policies with scripted sequencing. The resulting trade study identifies a recommended architecture direction for continued system development. A physics-based lunar construction environment implemented in MuJoCo provides a common testbed in which these approaches can subsequently be implemented, compared, and evaluated. The resulting framework connects lunar construction requirements to autonomy architecture selection and simulation-based evaluation. The problem definition establishes measurable requirements for autonomous RCU construction, the architecture trade identifies a pathway for implementing multimodal autonomy, and the MuJoCo testbed demonstrates the framework through an initial autonomous RCU placement task. Together, these contributions establish an engineering framework and evaluation platform for investigating multimodal embodied AI for lunar construction and provide a foundation for subsequent robustness, hardware validation, and simulation-to-real transfer.

This paper presents the design, optical characterization, and thermal performance verification of a compact spectral subsystem developed for space-grade in-situ resource utilization (ISRU) missions. The subsystem employs a Circular Variable Filter (CVF) comprising four sector-shaped bandpass filters, driven by a NEMA 11 hybrid stepper motor (Lin Engineering 211-13-01) to discretize the 400–2500 nm spectral range into 200 contiguous sampling bands, each with a full-width at half-maximum (FWHM) of ≤ 2%. A coaxially packaged dual-sensor photodiode module (Hamamatsu C10439-15) integrates a Silicon sensor (400–1200 nm) and an InGaAs sensor (1100–2500 nm) to achieve seamless broadband coverage without spatial registration error. Quantitative ground characterization confirms a Signal-to-Noise Ratio (SNR) exceeding 30 dB across an operating temperature range of −40°C to +70°C, with signal standard deviation maintained below 0.004 V throughout thermal cycling. The assembled Flight Model (FM) successfully passed all space qualification tests mandated by the Taiwan Space Agency (TASA), including random vibration (14.10 Grms), thermal vacuum cycling (−40°C to +85°C, 8 cycles), electromagnetic compatibility (MIL-STD-461G), and total ionizing dose radiation (15 krad), validating its readiness for lunar rover deployment.

NASA’s Uranus Orbiter and Probe (UOP) mission concept study for the 2023-2032 decadal survey, which was ranked as a top priority, envisions a comprehensive suite of instruments to study the origins of Uranus, the physical and chemical processes governing its internal and external structure, as well as the potential habitability of its moons. In addition to the main orbiter module, the mission plans to deploy an atmospheric probe into the atmosphere of the planet to measure its structure, chemical composition, and winds as a function of depth, in the pressure range between 0.1–10 bar. A key instrument on the atmospheric probe is a mass spectrometer to identify chemical species and accurately quantify their abundances, and to determine isotope ratios of selected elements. Here, we present a prototype time-of-flight (TOF) mass spectrometer that is based on a heritage instrument, comprising an electron-impact ionization source, an ion mirror, and a microchannel-plate detector. We have thoroughly characterized various performance parameters and show that this instrument meets most, and even surpasses some, of the measurement requirements defined in the Science Traceability Matrix (STM) for the UOP mission. Specifically, the mass spectrometer detects mass-to-charge ratios between m/z = 1-300 (extended range: m/z = 1–1000), with a mass resolution of about m/Δm = 1400 for masses above m/z = 28. It covers an instantaneous dynamic range of 5-6 decades and is sensitive to partial pressures of 4.5x10-14 mbar (corresponding to ~1200 particles/cm3). We determine isotope ratios for the noble gases Ne, Kr, and Xe – a crucial measurement during the probe descent – with a relative error of less than 5%, except for the least abundant isotopes Ne-21 and Xe-126. These performance parameters, together with its low mass and modest volume, make our mass spectrometer well suited for the atmospheric measurements on the Uranus atmospheric probe. A gas calibration unit, a gas-handling system, and a gas separation and enrichment system to enhance the detection of low-abundance species, notably the noble gases (except He), are currently under development and we expect these additions to further improve the sensitivity and dynamic range of the mass spectrometer.

Calibrating the ion measurement mode of a space-borne mass spectrometer requires an ion source that emulates the measurement conditions in space: a collimated beam of low velocity spread, with a transverse area of the order of 1 cm² and beam currents in the picoampere range. In the framework of ESA's JUICE mission, a carbon-nanotube-based Low Energy Ion Source (LEIS-CNT) with a novel ion cage has been developed. In this source the neutral test gas is ionized by electron ionization (EI) using field-emitted electrons from a carbon nanotube (CNT) emitter, whose increased emitting surface improves the collimation of the electron beam through the ionization volume compared with conventional hot-cathode ionization. The LEIS-CNT is a continuous ion beam source with a typical ion beam current between 1 pA and 70 pA over a transverse beam area of 1.5 cm², at a chamber background pressure of 10⁻⁷ mbar. The ion beam energy can be tuned from 1 eV to 90 eV, with a typical theoretical energy spread of 1 eV full width at half maximum and measured spread of 5.0 ± 0.5 eV, and emission currents up to 400 µA have been achieved from the CNT emitter. The ion beam is characterized with a customized Faraday cup and a delay-line detector as a function of beam energy, electron beam-shaping configuration, and ion drift gradient in the cage. A linear dependence of the ion yield on the ionization energy is found, attributed to transverse momentum acquired by the electrons at the extraction grid. The characterized LEIS-CNT provides a compact, low-power ion source suitable for calibrating the ion measurement mode of space-borne mass spectrometers such as NIM.

Solid-sample analysis is essential for investigating planetary formation, surface evolution, and alteration processes because rocks and regolith preserve the elemental and isotopic records of these events. Laser-ablation time-of-flight mass spectrometry provides a compact approach for direct in situ analysis of solid planetary materials by detecting ions generated in the laser ablation plume. However, conventional single-pulse laser-ablation measurements are limited to the electrically charged fraction of the plume, while the neutral species in the plume remain inaccessible. This work presents the design and laboratory implementation of a dual-mode approach that enables analysis of both prompt ions and the initially neutral fraction of the same ablation plume. In the first mode, a nanosecond 266 nm laser pulse ablates the sample and the prompt ions are extracted and guided toward the time-of-flight mass analyzer. In the second mode, these prompt ions are rejected and a delayed second 266 nm laser pulse is introduced approximately parallel to the sample surface to ionize neutral species in the plume by multiphoton ionization. The newly generated ions are then guided through the same mass spectrometer and detected using the common readout chain. A laser configuration with a passive Q-switch for ablation with an active Q-switch for post-ionization was selected to combine reproducible plume generation with precise control of the relative laser-pulse timing. The laboratory implementation integrates the two laser subsystems with a CODEX TOF mass spectrometer for DIMPLE and dedicated trigger, control, and acquisition electronics. A validation strategy is defined to demonstrate prompt-ion rejection, identify ions originating from initially neutral plume species, and characterize the influence of post-ionization delay and laser-plume interaction geometry. By allowing the instrument to analyze either the prompt-ion fraction or, in a separate measurement, species originating from the initially neutral fraction of the ablation plume, the dual-mode approach is intended to support more comprehensive in situ composition analysis of planetary rocks and regolith.

Combined analysis of solid material and neutral gases provides complementary information on planetary composition, volatile release, and interactions between surfaces and their surrounding atmospheres or exospheres. This work presents a novel mass spectrometric instrument concept and investigates a pulsed electron source as a key enabling step toward its realization. The instrument combines laser ablation and electron ionization within a common ion-source architecture for time-of-flight (TOF) mass spectrometry of solid and gaseous samples. Three proposed operating modes share the same ion optics, ion mirror TOF analyzer, detector, and readout electronics. In laser -ablation mode, a nanosecond 266 nm laser pulse generates prompt ions from a solid sample for direct analysis. In electron ionization mode, a pulsed electron beam ionizes ambient neutral gases independently of laser operation. In the combined mode, prompt ablation ions are rejected, and a delayed electron pulse ionizes neutral species remaining in the ablation plume. Measurements without laser ablation would characterize the ambient gas background and help distinguish it from laser-associated signals. Realizing the electron ionization modes requires a controllable electron source whose emission pulse defines the ionization event and provides the TOF timing reference. To address this requirement, a carbon nanotube (CNT) cold-field emitter was characterized in a stand-alone laboratory setup using 30 µs extraction pulses at 10 kHz. The measurements yielded emission currents of several hundred microamperes, with pulse-to-pulse relative standard deviations approaching approximately 1% at sufficiently high currents. During more than 80 min of continuous pulsed operation at 900 V, the relative standard deviation remained between approximately 1.1% and 1.7%. Estimated electrical power associated with capacitive charging and electron emission was approximately 70–110 mW at emission currents of 200–310 µA, excluding pulser and control-electronics losses. These experimental results represent a key enabling step toward realizing this instrument by establishing a basis for further development and integration of its pulsed electron source. By combining solid-sample and neutral-gas analysis within a common mass-spectrometer architecture, the concept could provide complementary measurements of surface composition, volatile inventories, and surface-exosphere interactions with a single compact instrument. Such capability is particularly relevant to landed lunar exploration, where regolith composition, locally released volatiles, and the tenuous exosphere are closely coupled measurement targets.

The astrobiology community is in preparation for various life detection missions due to take place within the next few decades. These missions will target the inner solar system with the planets Venus and Mars as well as the outer one with the icy moons of Jupiter and Saturn. Novel and sensitive measurement techniques are of great interest for life detection, as the so far applied gold standards in this field, such as GC-MS, have encountered technical difficulties in providing conclusive results. In this contribution, we present the measurement capabilities of our laser-based mass spectrometer for the detection of organic molecules within Mars analog soil samples. The deployed mass spectrometer system consists of a compact time-of-flight mass analyzer (currently in preparation for a NASA CLPS mission to the lunar surface) which is connected to a passively Q-switched microchip laser operated at UV irradiation. As part of a NASA PSTAR project, named TATOOINE, the system has been deployed recently in the Mojave Desert, USA – a Mars analog environment. Organic extracts from Mojave soil have been generated through a solvent extraction unit (NASA Ames Research Center). The extracts together with internal standard molecules have been drop cast (1 µL) into different sample cavities on the sample holder (gold coated). Empty cavities were used as blanks. The holder was subsequently transferred into the vacuum chamber of the mass analyzer unit followed by chemical composition measurements using our laser mass spectrometer. The measurement performance of the system will be presented in more detail, with a focus on the environmental challenges, facing air temperatures at a shadow of up to 45 °C and high wind gusts. Operating flight-like systems and measurement protocols within analog environments represents the next logical step in preparing for future space exploration missions devoted to life detection.

CODEX (Chemistry, Organics, and Dating Experiment) must convert large numbers of laser ablation mass spectra into interpretable geochemical products during DIMPLE operations on the lunar surface. This work characterizes that measurement and data processing chain using the flight-representative ion optics of the CODEX Engineering Development Unit. DIMPLE (Dating an Irregular Mare Patch with a Lunar Explorer) will investigate the age and composition of rocks from the Ina irregular mare patch using CODEX, which combines laser ablation mass spectrometry (LAMS) with laser ablation resonance ionization mass spectrometry (LARIMS). Measurements of metallic NIST Standard Reference Materials were used to evaluate trace species response, spectrum accumulation, automated peak extraction, mass calibration stability, and spatial and ablation sequence variability. The m/z 208 feature was recovered at derived 208Pb mol fractions down to approximately 35 mol ppb, corresponding to certified elemental Pb concentrations well below the 2-ppm reference. Across selected species, higher-abundance responses follow species specific power-law relationships. Automated mass calibration remains stable across the measurement campaign, with residual metrics predominantly below the defined acceptance criterion and isolated outliers readily identifiable for quality assurance. Spatially and interval resolved analysis further shows that trace species response can vary between measurement locations and along the ablation sequence while remaining traceable through the processing chain. A measurement of a Madagascar flood basalt additionally demonstrates recovery of a rich LAMS spectrum from a natural matrix representative for DIMPLE science. Together with established quantitative laser ablation mass spectrometric calibration approaches, these results provide a practical pathway from individual laser shot spectra to traceable Level 2, Level 2A, and higher-level geochemical products suitable for DIMPLE science operations.

Future landed missions to the lunar south pole are expected to obtain in situ data to better inform the assessment of potential resources and provide reliable knowledge of the geotechnical state of the regolith surface. This information is vital for informing the design requirements for lunar infrastructure, such as the Artemis Moon Base. However, direct in situ measurements of lunar regolith density state, stiffness, and stress-dependent deformation behavior remain limited. This paper evaluates whether the deployed footpad of the Regolith and Ice Drill for Exploring New Terrains (TRIDENT) system, developed by Honeybee Robotics and planned for lunar deployment, can serve as an opportunistic geotechnical sensing instrument during surface drilling operations. Testing and evaluation of the TRIDENT footpad were conducted in the Mines Lunar Surface Simulator using the Colorado School of Mines-Lunar Highland Type-Testbed (CSM-LHT-T) simulant prepared at multiple density profiles. Following a standard ASTM procedure for plate load testing (ASTM D1195), both a circular plate and the TRIDENT footpad were subjected to controlled vertical loading to compare load-deflection behavior and determine whether footpad response could be correlated with established geotechnical plate-load testing methods. The work was designed to assess whether vertical reaction data from the TRIDENT system could be used to estimate geotechnical parameters, including surface deflection, the modulus of subgrade reaction (soil stiffness), and the strain modulus of in situ regolith. Results demonstrate that the TRIDENT footpad produces repeatable, density-dependent deflection trends that correlate with standard plate-load measurements. As expected, denser regolith simulant exhibited reduced vertical deflection and increased stiffness. The TRIDENT footpad generally produced larger measured deflections and lower calculated stiffness values than the standard plate, indicating that a geometry-specific correction is required before translating footpad response into standard geotechnical parameters. This difference is attributed to the honeycomb footpad geometry, which is needed to provide lateral restraint and stabilization during drilling operations for the TRIDENT system. The findings indicate that the TRIDENT system's footpad can provide more than mechanical stabilization during drilling. With appropriate calibration, the system may also generate in situ geotechnical data of the surface regolith during operations. This capability would allow future landed missions to obtain additional surface characterization information without requiring a dedicated geotechnical payload. The results support the development of analytical protocols to interpret drill-footpad load and deflection data as a practical approach to assessing lunar regolith stiffness and density state. This work also identifies the additional calibration needs for lunar applications, including expanded testing across simulant types, density states, environmental conditions, and flight-representative operational constraints.

Future planetary exploration requires rapid, high-resolution in situ chemical and isotopic analysis of surface materials with minimal sample preparation. The Laser Ablation Mass Analyzer (LAMA) addresses this need by combining pulsed laser ablation sampling with high-sensitivity time-of-flight mass spectrometry (TOF-MS) optimized for vacuum environments. LAMA pairs a Q-switched Nd:YAG laser (1064nm) with a heritage-derived TOF-MS design, introducing an innovative standoff analysis mode capable of measurements across adjustable target ranges from 5cm to 30cm. By utilizing dynamically tunable ion optics with shape-changing fluidic and polymer lenses, the system eliminates bulky mechanical focus components, resulting in a compact, low-power, few-kilogram payload optimized for rovers, landers, or astronaut-deployed platforms. Laboratory testing with planetary analog materials demonstrates ppmw-level detection limits for key rock-forming and trace elements, alongside semi-quantitative elemental ratio determination within +/-10% relative standard deviation. The instrument achieves micron- to millimeter-scale spatial resolution suitable for analyzing highly heterogeneous samples, while high-intensity laser operation enables the correlated detection of H+ and O+ ions for water ice prospecting. Additionally, LAMA features semi-autonomous targeting to locate high-priority phases across exploration areas and can operate within habitats for detailed sample screening. By providing direct, non-destructive surface analysis without sample handling, LAMA bridges the gap between orbital observations and ground-truth data, offering a transformative capability for upcoming planetary science and resource prospecting missions.

Mars surface missions, especially Curiosity and Perseverance, have demonstrated the power of applying multiple techniques to the study of common targeted and acquired samples, enabling direct inter-comparison of results. Looking forward, in situ analytical objectives at Mars and beyond will be increasingly detailed and stringent. For example, identifying, quantifying, and correlating dilute and/or spatially isolated chemical compounds, including potential biosignatures, may require investigation of multiple replicates (and larger quantities) of heterogeneous samples, or alternatively precision chemical mapping at fine spatial scales, by multiple coordinated techniques. Such analyses are optimized when examining common samples or subsamples that, in many cases, have similar “front-end” requirements, variously including physical manipulation, containment, phase separation, and chemical reaction/extraction. An approach where a group of instruments share a single sample pre-processing (SPP) capability may be a powerful component of integrated analytical payload designs to come. In support of NASA’s Mars Exploration Program (MEP) Future Plan Exploring Mars Together, Initiative I-3.4 “In Situ Sample Handling, Pre-Processing, and Analysis & Returned Sample Handling,” our team is helping develop SPP technologies to be available for instrument and payload development opportunities and collaborations for Mars, and potentially other bodies with similar objectives. In the present pilot phase, the focus is on assessing SPP needs of current and projected analytical techniques and producing demonstration prototypes that can interface with a variety of instruments while maintaining realistic path-to-flight engineering principles. As a mission scenario point of departure, we consider a mid-to-high latitude Mars lander or rover with a deep coring drill and a payload focused on astrobiology and geochemistry science as well as resource assessment, targeting analyses of water ice-rich samples that could be acquired from ice-cemented regolith, glacial ice, and/or intact rock with microfracture ice. Initially, SPP technologies to support such a payload may be categorized as applying common pre-processing steps to (i) bulk samples (ice-rich drill fines or separately crushed material) or (ii) intact cores or granules/chips (also ice rich at various spatial scales). We are prototyping and demonstrating hardware and protocols for both of these sampling scenarios, including containment, phase separation, digestion/extraction, manipulation, and instrument injection capabilities that are flexible to an array of Acquisition-SPP-Analysis combinations developed collaboratively with interested parties.

As more operational satellites are being placed in orbit, there is a greater need to know the location of objects orbiting the Earth to minimize the risk of collision. The location of operational satellites are well known because the operators of those satellites need to know where those satellites are located for operations. The location of space debris is not always as well known. In this paper, we design a constellation of satellites to optimally observe space debris. With increased observations of space debris, the knowledge of the location of the debris will be increased and will thus help to minimize the likelihood of a collision between an operational satellite and a piece of debris. A set of two-line element sets for space debris was obtained from Space Track to provide a realistic catalog of debris objects. For this work, the objective is to maximize the number of objects from this space debris catalog that will be viewed by the constellation. The first step in the process is a perigee/apogee filtering that removes all debris orbits from consideration that are not visible by a candidate constellation orbit. All remaining debris orbits are then evaluated to determine if the closest approach between the debris orbit and the constellation orbit are within viewing range. The research will use a metaheuristic algorithm, such as Particle Swarm Optimization, to identify the set of constellation orbits that maximizes the number of debris pieces observed. The decision variables for the optimization process are the orbital elements of the constellation orbits. Rather than using the tra- ditional Keplerian orbital elements, a slight modification is implemented. The semimajor axis (a) and eccentricity (e) are replaced with the perigee radius (rp) and difference between perigee and apogee radius (δ). This methodology has the advantage being able to easily enforce a constraint that the perigee radius is above a desired altitude.

Many vehicle trajectory design problems impose constraints that apply at points along the trajectory, not only at its endpoints. Such requirements arise when trading propellant against intermediate science objectives, maintaining safe re-entry and landing corridors, and identifying safe re-ignition zones for reusable spacecraft, and the binding constraints vary widely from one mission to the next, spanning initial and terminal conditions, fuel and control limits, and intermediate waypoints. The object that captures this requirement is the constrained reachable tube: the set of intermediate states that lie on at least one admissible trajectory joining a prescribed initial set to a prescribed target set while honoring the dynamics, the control bounds, and any imposed path constraints. This research combines two capabilities that existing methods provide separately. Set-based reach-avoid methods admit intermediate path constraints natively, as part of the set computation, but they yield outer approximations, are limited in state dimension, and do not return a control. Trajectory-based methods provide nonlinear dynamics, the realizing control, and trajectories that are admissible by construction, but formulating the constrained problem directly is difficult, and the common alternative of intersecting reachable and controllable sets cannot incorporate intermediate state constraints. This paper presents a method that unites these strengths. The constrained reachability problem is posed as an average deviation optimal control problem whose boundary-value structure is invariant to the placement, number, and type of intermediate state constraints, so that waypoints, keep-in zones, fuel budgets, and path floors enter each sweep direction as data rather than as a re-derivation. A successive convex programming algorithm solves this problem for arbitrary nonlinear dynamics with continuously differentiable Jacobians and any user-composed combination of convex state and control constraints, returning the realizing control with each converged boundary trajectory and an inner sampling of the tube. The approach is demonstrated on five problems spanning linear, two-body orbital, and coupled translational–attitude dynamics, exercising boundary-condition, waypoint, fuel-budget, and path-constraint regimes. Validation includes a double integrator with a known analytical solution and a comparison with a published proximity-operations example. The algorithm has been implemented, and all numerical examples have been computed and verified. The paper has been drafted and iterated on by the authors. The full paper will present the formulation, the algorithm, numerical results, and computational performance in detail. It will also describe how artificial intelligence tools were used to support the work, including organizing the literature, developing code, and drafting and editing the manuscript, and will discuss the limitations of those tools. All formulations, derivations, and numerical results were developed and verified by the authors.

The expansion of human space operations into the cislunar domain under Artemis demands communications infrastructure resilient to quantum computing threats. Quantum key distribution (QKD) provides information-theoretic security guarantees that classical encryption cannot match, making it a candidate foundation for cislunar command, control, and telemetry security. However, no systematic framework currently exists for identifying which orbital architectures can sustain QKD operations across realistic mission timelines and coverage requirements. This paper identifies the minimum relay architecture required to establish a quantum-secure communications backbone in the Earth–Moon system. We screen all 20 planar periodic orbit families in the JPL CR3BP catalog against a single physics-derived threshold: the key sustainability ratio ρ = K_gen/K_burn, where K_gen and K_burn denote the secret-key generation and consumption rates, respectively. The threshold ρ = 1 separates self-sustaining (ρ > 1) from key-depleting (ρ < 1) relay operation. The framework propagates QKD link performance across a global Earth-surface grid and the cislunar domain over a 30-day window using a first-principles optical link budget anchored to hardware parameters representative of space-qualified quantum terminals. Instead of heuristic orbit scores, categorical trust tiers, or arbitrary weighting functions, constellation performance is characterized by three quantities directly traceable to orbital dynamics: the key sustainability ratio ρ, the secure availability fraction T_crypto, and Security Reachability (SR), defined as the fraction of Earth-surface nodes receiving sustained QKD support. Independent Qiskit Aer simulations validate the underlying quantum-channel model and confirm operation below BB84 security thresholds. Of the 20 orbit families screened, only five produce non-negligible secure-key generation within the 65,000-km free-space QKD range budget. The Discrete Retrograde Orbit (DRO) achieves the highest sustainability ratio (ρ = 0.567) through rare but intense close-Earth periapsis passes generating burst rates up to 170 bps. The Resonant-2:1 family achieves ρ = 0.403 through more frequent, lower-peak contacts with broader temporal coverage. No individual family reaches ρ ≥ 1 at a 1-bps key-consumption rate, indicating that no single-orbit relay can sustain indefinite operation under current hardware assumptions. The central finding is the existence of a sharp infrastructure saturation point. A two-family DRO + Resonant-2:1 architecture achieves 78.9% Security Reachability, capturing nearly all attainable relay performance within the 20-family CR3BP catalog. Adding the three remaining active families improves reachability by only 0.8 percentage points while requiring three additional orbit families. A security-backbone analysis further shows that DRO and Resonant-2:1 jointly account for 56% of total network key-generation capacity, whereas Resonant-1:2 contributes only 0.03% and is therefore architecturally redundant. These results establish a quantitative orbit-selection framework for cislunar quantum communication infrastructure. A relay architecture built around DRO and Resonant-2:1 captures 78.9% of all achievable Security Reachability identified across the 20-family CR3BP catalog, while additional families provide sharply diminishing returns. The framework provides mission designers with a physics-grounded methodology for evaluating quantum-secure relay architectures for Artemis-era Earth–Moon communications.

Interplanetary trajectory design is a complex problem requiring the simultaneous optimization of both discrete variables, such as multi-flyby sequences, and continuous variables, including low-thrust control histories. Because optimizing both domains simultaneously under nonlinear dynamics is computationally prohibitive, conventional trajectory design relies on a two-phase framework: a broad search of the discrete design space using lightweight models such as Lambert calculations, to extract promising candidates, followed by the local optimization of continuous variables via nonlinear programming. However, lightweight models have limited capability to handle constrained multivariable low-thrust corrections and may therefore fail to capture feasible regions in complex low-thrust trajectory design problems. To address this limitation, previous work by the authors introduced a fast convex-programming (CP) based refinement into fixed-sequence broad search, enabling constrained multivariable trajectory-design problems beyond the capability of lightweight models to be evaluated within practical computational limits, thereby expanding the searchable design space. This paper extends the CP-augmented broad search to variable-sequence search by incorporating the Null Legs method. The proposed framework enables the comprehensive exploration of both discrete flyby sequences and constrained continuous low-thrust trajectories directly from the broad search phase. We evaluate the framework using a BepiColombo-inspired low-thrust Mercury transfer with nine flybys. With CP refinement, the variable-sequence broad search generates 547,760 candidates, compared with 74,588 without it, demonstrating a substantial expansion of the explorable design space.

The continued deployment of large satellite constellations has driven a rapid increase in the resident space object (RSO) population in low Earth orbit, making reliable tracking and collision avoidance increasingly critical operational concerns. As the RSO population grows, so does the risk of collisions and the onset of a cascading Kessler syndrome, making accurate RSO cataloging indispensable. Today, space surveillance is conducted primarily using ground-based optical and radar sensors, whose performance is constrained by weather, daylight, and geographic location. To overcome these limitations and enhance surveillance capability, space-based surveillance systems are emerging as a valuable complement to ground-based assets. Motivated by this, we present a mission analysis and feasibility study of a space-based surveillance system (SBSS) using commercial-off-the-shelf (COTS) dual-purpose optical star trackers, which are widely deployed for high-precision spacecraft attitude determination. The study investigates constellation design, sensor placement, attitude estimation accuracy, apparent visual magnitude, and orbit determination precision for RSO cataloging with dual-purpose sensors. To the best of our knowledge, no commercial or open-source tools support multi-observer mission analysis for space-based surveillance systems (SBSS). Therefore, a software simulation framework was developed. The framework models constellations of up to 80 observer satellites distributed across two ∼500 km orbital rings of 40 satellites each in near-equatorial and polar orbits. As an illustrative example, a population of 10,600 catalogued Starlink-class RSOs, each modelled as a 3-m diameter sphere, is propagated alongside the observer constellation, with the solar arrays neglected in the optical model because of their comparatively low reflectance. The framework then evaluates the effects of constellation size, satellite spacing, sensor placement, and the configuration of multiple optical heads. The detection pipeline first renders the field of view of each optical sensor, modelling RSO apparent visual magnitude from phase angle and reflectance while excluding eclipsed objects. It then performs dual-purpose processing for RSO detection and observer attitude estimation using a modelled COTS star tracker that incorporates magnitude limits and sensor noise. Our results conclusively show that coverage increases with constellation size before saturating at approximately 16 observer satellites with three optical heads per observer satellite in the cross track-ram plane, beyond which additional satellites primarily improve revisit cadence rather than overall coverage. Over the 24-hour propagation period, approximately 90% of the catalogued Starlink-class RSOs enter the field of view of the optimal 16-satellite constellation, while the remainder remain below the observer constellation because of orbital geometry. When the sensor apparent-magnitude limit is considered, the observable fraction is 88%. More than 50% of the detected RSOs are re-observed within seven hours, and approximately one-third within three hours, complementing ground-based observations and reducing orbit uncertainty by limiting covariance growth. In the full paper, we will present a comprehensive mission analysis and feasibility assessment of the proposed SBSS. These contributions are timely, addressing the need for SBSS that complement ground-based optical and radar assets by overcoming their limitations and enhancing overall surveillance capability. By demonstrating the feasibility of such a system, this work represents an important step toward next-generation space traffic management.

Constellation Monte Carlo studies depend on both a tasking policy and the orbital trajectories used to evaluate that policy. When reference trajectories are scarce, seed replay increases campaign count without expanding handoff geometry, while independent orbital-element jitter can produce dynamically inconsistent or Earth-intersecting orbits. We investigate whether constrained diffusion can expand a scarce orbital population for constellation tasking while preserving physical executability. A force-conditioned denoising diffusion model learns a 13-dimensional orbital prior in which semimajor axis is recovered analytically from perigee altitude and eccentricity. Each generated sample is propagated in Basilisk under one of five force-model tiers, separating learned population variability from simulated temporal evolution. We compare diffusion with jittered bootstrap, per-tier Gaussian mixtures, and a conditional variational autoencoder using an 800-trajectory corpus. Evaluation includes held-out and shifted-blind splits, a Lambert interior check, physics-residual estimation diagnostics, and 4,000 paired GoDSAT-compatible campaigns. On held-out trajectories, the largest diffusion model achieves maximum mean discrepancy of 0.0171 ± 0.0239, with similar held-out MMD to bootstrap (0.0170) and the mixture (0.0198), while generating samples farther from individual training priors (median nearest-training distance 2.7 versus 0.10 standardized units). All six generators remain readily distinguishable from the shifted-blind target population in every seed–tier block (classifier AUC 0.991–0.998). Endpoint-conditioned samples meet Lambert boundary conditions but show larger interior position error than the matched Lambert reference (29.2 versus 5.7 km mean RMSE). Residual diffusion improves selected sparse forecasts and catalog-mode recall but does not outperform classical estimators on custody ranking. In a common 16-satellite configuration, diffusion has similar mean custody (0.463) to seed replay (0.470) and bootstrap (0.461), while the force-tier mixture shifts custody by approximately 5.5 percentage points. Constrained diffusion is therefore useful for local, in-support Monte Carlo augmentation but cannot replace orbital dynamics or serve as an operational posterior. Trajectory-population construction should be reported as a consequential source of uncertainty in constellation analysis.

Cislunar operations are evolving from isolated missions toward a heterogeneous environment in which spacecraft operate near the Moon, along Earth–Moon transfer trajectories, and within the libration-point region. This creates a need for scalable computational methods for joint constellation design and planning over a large number of candidate observer locations drawn from heterogeneous orbit families. A representative application is cislunar space situational awareness (CSSA), where a small number of observer spacecraft are selected from candidate libration-point-orbit locations and tasked with time-dependent sensor-pointing directions to detect objects of interest. The underlying coverage demand can take two forms, either a static volume requiring persistent surveillance such as the Cone-of-Shame region, or a dynamic, time-varying demand such as tracking low-energy transfer (LET) trajectories as they pass through the Earth–Moon transit corridor. In both cases, the optimization problem jointly selects observer locations and time-indexed sensor-tasking allocations to maximize coverage of the relevant targets. However, direct mixed-integer linear programming (MILP) formulations of such problems become computationally intractable to solve within a meaningful time limit at realistic scale. In this work, we develop a Renormalization Group (RG) inspired framework to solve such large-scale design-and-planning problems. It is inspired by renormalization concepts in statistical physics where fine-scale degrees of freedom that are redundant at the mission-performance scale are eliminated, while representative orbit slots that preserve large-scale coverage structure are retained. We create a weighted graph with each orbital slot as a node, with edge weights derived from time-dependent mission coverage profiles that jointly capture redundancy and complementarity among candidates. Specifically, the weight of an edge connecting two slots depends on the extent of the similarity in the accesses of the two slots to the targets. Spectral analysis of the graph laplacian naturally yields clusters of slots that are indistinguishable at the mission scale and separates them from slots that provide genuinely distinct access geometry. Representative slots from each cluster are chosen to create a compressed candidate manifold as input to the joint design-and-planning solver. The proposed framework is applied to both the static Cone-of-Shame and dynamic LET transit-window scenarios described above. For each scenario, the RG-based formulation is benchmarked against an exact branch-and-bound (BnB) solver in Gurobi and a recently proposed custom Lagrangian-relaxation-based (LM) heuristic. Experiments on small-scale problem instances verify that the proposed method recovers the same optimal solutions obtained from BnB. However, on realistic, large-scale instances under a fixed meaningful time budget, it substantially outperforms both BnB and the Lagrangian heuristic, producing higher-coverage constellation designs together with their corresponding sensor-tasking allocations. The key insight is that heterogeneous orbit-family sets contain latent structure, where slots within the same orbit family are often mutually redundant while slots across different families provide complementary coverage of otherwise weakly observable regions and time intervals. Our approach exposes this structure directly, in contrast to conventional methods that rank and prune individual slots by coverage score alone.

In February 2022, 38 of 49 newly launched Starlink satellites reentered unplanned after a geomagnetic storm raised thermospheric density, an effect that pre-launch analyses based on NRLMSISE-00 had underestimated. When density is underpredicted, drag is higher than anticipated, and a satellite decays faster than planned. A mission that sizes its drag margin therefore requires an upper bound on density rather than only a best estimate. The bound is set to be exceeded a stated fraction of time, such as 10 percent, because a bound that is never exceeded would have to cover the strongest possible storm and carry that margin into every quiet day. Such bounds matter wherever drag estimation is critical, as in formation and constellation keeping and in space surveillance and tracking and space traffic management. Machine-learned density models that supply an uncertainty fix it when training ends and check it on data from the same span of years and altitudes as training. This paper presents a method for keeping the bound valid in flight. A transformer learns a correction to NRLMSISE-00 from space weather and orbit inputs and outputs a median and an upper bound, which is then recalibrated online by quantile tracking, a form of adaptive conformal inference, from the exceedances observed during flight. The method is trained on GRACE-B accelerometer-derived density for 2008 to 2015 and tested in 2016, when the decaying satellite's mean altitude was lower than any altitude in training. The network reduces the error of NRLMSISE-00 from 25.23 to 14.41 percent. Its upper bound, however, was designed to be exceeded 10 percent of the time and in the test year it is exceeded 18.35 percent of the time. Giving the network altitude makes this worse, because on a decaying orbit altitude tells the network how far into the mission the data are. Recalibrating the bound in flight brings the rate back to 10 percent. Over 72-hour windows, the trained bound left more than 1 km of along-track drift unaccounted for in 7.23 percent of windows, and the recalibrated bound in none. When the density measurements that drive the recalibration arrive a day late, the rate still averages 10 percent, but on the worst day the bound is exceeded more than 64 percent of the time. These results come from one satellite and one test year. They suggest that missions planning against a density bound should specify it by how often it is exceeded in service and how quickly its measurements arrive.

Satellite disposal is a major topic in the space industry. In Low Earth Orbit (LEO), satellites experience re-entry due to atmospheric drag. The risk of ground casualties induced by satellite debris surviving re-entry is even exacerbated for today’s Mega-constellations. Therefore, more sophisticated approaches to ensure safe demise appear as necessary. The THREAD (THermite REactions Assisting satellite Demise) project, funded by a HORIZON-EIC-2024-PATHFINDER grant, is maturating an innovating strategy to solve this challenge. Thermites are pyrotechnic materials that initiate a self-sustained exothermic reaction when heated to their ignition temperature. Thermites could be incorporated in the satellite structure, close to components that are challenging to demise, providing additional heat during re-entry and assisting their complete demise. Preliminary projects already validated the relevance of this Thermite-for-Demise (T4D) strategy, thus THREAD aims at paving the way for its industrial adoption. THREAD activities cover the development of suitable thermite formulations, their manufacturing into usable building blocks and the whole analysis of viable T4D strategies, including the understanding of thermite behavior under launch and on-orbit stresses. While the launch environment is dominated by important mechanical stress, such as vibrations, the constraints to which spacecrafts are exposed in orbital environment are diverse. Among other concerns such as extreme and rapidly cycling temperatures is the complex, mixed space radiation environment. The impact of these constraints on thermites has so far been scarcely investigated. Space radiation alone encompasses several types of particles whose relative contributions are highly dependent on orbit parameters and specific use cases. In this work, we focused on how radiation can affect the properties of the thermites. The first irradiation campaigns have been performed under X-rays and Gamma rays using the PETRA platform of Hubert Curien Laboratory and at the Calliope facility of ENEA, respectively. These are the first thermite irradiations ever reported in open literature. For the project, dedicated irradiation setups that are safe for both human operators and equipment have been developed. Dosimetry has also been studied in detail: doses expected during a typical mission profile are assessed depending on factors such as orbit type, duration and shielding. The spatial distribution of the deposited dose in thermites during irradiation in laboratory conditions is evaluated through Monte Carlo codes. Irradiation tests have been performed at various doses and doses rates to better understand the possible role of accelerated testing. Sensors such as thermocouples and cameras are used to detect and record in situ possible ignitions triggered during irradiation. Then, key figures of merit of thermites such as ignition and burn properties have been characterized after irradiation to detect any radiation-induced change. Data are still currently collected and the whole set of results will allow discussing the suitability of thermites to space environment applications. This is especially important to evaluate the minimum amount of shielding for thermite usage. A first analysis of the results of irradiation and other stress tests will allow identifying the most impacting stress conditions that will be discussed at the conference and used to define future test campaigns.

Sustained manned lunar exploration requires effective protection against a complex radiation environment from combined sources. In fact, a lunar habitat is exposed both to a primary flux, common to deep-space probes —comprising Galactic Cosmic Rays (GCRs), including protons, alpha particles, and HZE nuclei, as well as Solar Energetic Particles (SEPs) - and to an upward secondary flux of lunar neutron albedo. The latter is generated by GCR - induced nuclear spallation and inelastic-scattering processes within the upper meters of the lunar crust and can significantly increase the effective dose inside the habitat. This study presents a methodology for optimizing a multilayer radiation shield for a crewed lunar habitat, using the lunar surface environment as the primary design case. The other mission phases to be tackled by the lunar module are then investigated as cross-validation cases. Preliminary optimization trials indicate that the proposed methodology is capable of identifying multilayer configurations achieving high protection at the lunar surface. The shielding compact is bounded by two structural plates made of Al 2024 or Al 7075 alloy, a constraint introduced to maintain assembling similarity with traditional structures. The optimization is better focused on the internal core comprising hydrogenated polymers, neutron absorbers, and ultra-thin metallic interface foils. The design variables – including material selection, layer sequence, and individual thicknesses, are optimized to attenuate primary ions, secondary cascade products, and the neutron-albedo field. The software platform integrates a particle-transport solver with an advanced multidisciplinary optimization environment. Radiation transport is performed by the Cosmic Shield Simulator, a purposely coded C++ framework based on Geant4, using advanced physics models to simulate particle–matter interactions. The simulator evaluates the performance indicators required to compare candidate designs, including Protection Efficiency, Energy Efficiency, areal Weight, and Total Thickness. The identification of the optimal solution is performed by Cosmic Shield Optimization Engine, implemented in Python within a three stages architecture combining Optuna adaptive sampling, Pymoo evolutionary optimization, and Gaussian-process refinement. This optimization approach, detailed in the paper, efficiently addresses a highly nonlinear solution space by minimizing an objective function that balances shielding performance against constraints. Mission final verification also evaluates the robustness of the optimized architecture and estimates its protection margins across distinct radiation environments: (i) transit in LEO parking, accounting for trapped particles through the IRENE AE9/AP9 models; (ii) passage through the Van Allen radiation belts; (iii) Earth–Moon cruise, dominated by SEPs described by the ESP-PSYCHIC model and GCRs represented by ISO 15390; and (iv) the operations at the lunar surface recalled above.

The proliferation of space debris in low Earth orbit (LEO) threatens operational satellites and future space access. Current removal missions rely on ground-controlled capture of pre-characterized targets, an approach that does not scale to the diverse, uncooperative debris population. This paper presents the design and early validation of PAL-CleanX, an AI-driven robotic servicer that autonomously executes detection, classification, capture, and disposal with minimal human intervention. The key innovation is an integrated perception-to-action pipeline that combines a convolutional neural network (YOLOv7-based) with a reinforcement learning planner (PPO) to select between two capture tools in real time: a deployable net for small irregular objects and robotic arms with underactuated magnetic grippers for large metallic debris. A finite-state machine uses the real-time pose estimate and tumble rate to make the tool choice onboard, removing the need for pre-mission target programming. Validation is based on high-fidelity closed-loop simulations and component-level hardware testing. In simulation, the perception network achieved 92% mean average precision and orientation errors below 3°; the proximity policy reduced delta-v consumption by 18% relative to classical guidance while maintaining 100% collision avoidance in 1,000 Monte Carlo trials. As a component-level demonstration, a magnetic gripper on an air-bearing testbed successfully captured a tumbling representative target at spin rates up to 0.4 rad/s. A benchtop prototype integrating a camera-LiDAR sensor head and one robotic arm is under assembly to validate the perception-to-grasp sequence in the lab, with integrated testing and full system prototyping as near-term steps. NASA debris-evolution studies indicate that removing five large, high-risk objects per year could curb the growth of the LEO fragment population. PAL-CleanX is designed to contribute to that goal by enabling a modest constellation of autonomous servicers to target such objects at lower marginal cost. The paper also identifies the legal, policy, and verification steps required for an on-orbit technology demonstration, offering a pragmatic path from component validation to operational deployment.

This paper evaluates how effective neural networks are at modeling non linear orbital interactions in the Low Earth Orbit to predict the satellite debris collision risk and classify conjunction events into high and low risk categories. Using European Space Agency (ESA) Conjunction Data Messages (CDMs) spanning 2015 to 2019, a dual regression-classification framework was developed. The deep neural network achieved a 64% reduction in RMSE over a linear baseline and a test R2 of 0.8, confirming that the relationship between orbital parameters and collision risk is mostly non-linear. The binary classification model achieved 96% overall accuracy and an F1 score of 0.71 for high-risk events under a 32:1 class imbalance, without oversampling or synthetic data augmentation. These results support the use of AI driven methods for LEO traffic management and contribute to efforts aimed at ensuring long-term sustainability of Earth's orbital environment.

LEO has become a shared industrial environment. Commercial constellations, human spaceflight, space tourism, and in-orbit servicing now operate in the same congested shells, and the resulting collision risk and debris proliferation reinforce one another. Space traffic management remains fragmented across national jurisdictions and dependent on voluntary data sharing, and it lacks the interoperability, real-time responsiveness, and automated computation that long-term orbital sustainability requires. Building on a dual-entity model previously published by the author, comprising the Interstellar Grid Administration (IGA) and the Space Operations and Deconfliction Center (SODC), this paper develops a hybrid architecture that combines common coordination standards with decentralized, technology-enabled coordination. The IGA acts as the standards and trust root, publishing interface control documents, message schemas, and cryptographic trust boundaries. The SODC is its technical counterpart, a neutral, AI-augmented operations hub that fuses distributed sensor data, executes predictive collision analytics, and automates coordination through blockchain-secured protocols while preserving operator autonomy. The central contribution is the Adaptive Conjunction Management Protocol (ACMP), a coordination layer that scores conjunctions on a debris-weighted risk index, assigns maneuver responsibility deterministically through a Maneuver Responsibility Score computed identically by every node, and maintains a global orbital-capacity index from multi-operator data fusion. ACMP is supported by federated learning models trained on conjunction geometry, spacecraft health telemetry, and network-latency patterns, so that operators contribute to a shared risk picture without disclosing proprietary trajectory data. A discrete-event Monte Carlo simulation of a 4,380-spacecraft, 14-operator LEO population, calibrated against reported collision-avoidance maneuver rates and scored with the NASA Standard Breakup Model, compares ACMP against bilateral coordination and a centralized hub under degraded communications and partial compliance. ACMP retires 91% of the unmitigated collision risk, reduces expected new catalogable debris from 1,916 to 165 fragments per year, holds above 90% resolution of required maneuvers up to 20% inter-node packet loss, and eliminates duplicate avoidance maneuvers by construction. The results also expose a collective-action problem. Aggregate safety is set almost entirely by whether the largest operators participate, while the operators who gain most from federation are the smallest. The paper closes with a deployment roadmap linking the architecture to CCSDS and IADC frameworks, to the Traffic Coordination System for Space, and to emerging commercial-spaceport and cislunar infrastructures.

Safe operation of the geostationary (GEO) belt depends on catalogues that stay correlated through unreported maneuvers, yet short optical tracklets, dense satellite clusters, and sudden velocity changes routinely sever the link between a new detection and its parent object. Prior control-energy methods established that a minimum-energy metric can associate a maneuvered tracklet to a catalogue orbit; this work advances the same optimal-control foundation from a feasibility demonstration into a characterization framework that additionally probes maneuver-type inference and quantifies when the metric can be trusted. A short-arc tracklet is reduced to an attributable; admissible range/range-rate states are generated under GEO constraints and convex-sampled; and each candidate is scored by the minimum control energy ($G_2$) required to reconcile it with a propagated orbit, whose global minimum both ranks associations and reconstructs the post-maneuver state. A first novel direction is maneuver-type inference. For seven real east–west station-keeping (EWSK) maneuvers of Meteosat-9, evaluating the metric under the executed maneuver hypothesis returns an energy at least an order of magnitude below that of a modeled north–south hypothesis, indicating that the control energy carries recoverable maneuver-type information beyond association evidence alone. The comparison is deliberately one-sided: all seven tested maneuvers are real EWSK events with operator-reported ground truth, the north–south case is a modeled alternative rather than observed NSSK data, and no real north–south maneuvers are tested. Hence, this is a proof of concept, not a validated two-way classifier, pending balanced-class data. The reconstructed post-maneuver states nonetheless match truth to well within the observing sensor's field of view, and the range/range-rate recovery is corroborated by an independent pipeline on the same mission, giving agreement across two methods and two reference works rather than a single source. The second and more fully validated contribution is a quantified operating envelope for the metric. Across 1000-run Monte Carlo campaigns spanning impulsive and continuous maneuvers, orbit-to-tracklet gaps extended to 72 hours, and observation noise from survey- to high-uncertainty levels, the study maps how acceptable energy bounds scale: high noise inflates the metric by up to two orders of magnitude and multi-day gaps erode separability, turning qualitative feasibility into actionable association thresholds. On a real clustered-GEO scene the metric reliably eliminates impossible candidates, and a five-satellite Intelsat test driven purely by public predictive ephemeris delineates where the energy metric complements, and where it is overtaken by, a distance metric, charting the accessibility limits of predictive-data-only association. Relative to prior energy-metric association work, the distinguishing advances are: (i) a first, single-class demonstration that the control energy encodes maneuver-type information on real data; (ii) cross-pipeline, cross-mission validation beyond a single reference case; (iii) a quantified noise-and-epoch operating envelope covering continuous-thrust and multi-day scenarios; and (iv) an accessibility-focused assessment of predictive-data-only clustered association. Together these reposition the control-energy metric from a proof-of-concept association score into a practical initial-characterization tool that ranks candidates, eliminates impossible associations, signals likely maneuver type, and reports its own confidence — directly supporting follow-up tasking and catalogue maintenance for maneuvering GEO objects.

The orbital space debris problem continues to compound, as launch frequencies increase and existing debris objects increase the probability of collisions over time. A crucial component of solving the debris problem involves being able to track debris in the Lethal Non-Trackable (LNT) range, debris 1 to 10 cm in diameter, which is too small to reliably track from the ground and too large to be effectively shielded against. This paper builds on a previously developed high-performance C++ implementation of a Labeled Multi-Bernoulli Filter for space debris tracking and develops a sensor tasking strategy for space-based debris tracking constellations. The constellation studied is a ring of 100 sensors in an 800 km equatorial orbit, each able to detect debris only within 20 km, so that an object is seen for about three seconds per pass. To make the filter work at this scale, a fused proposal and a gating algorithm are integrated. The fused proposal combines the filter's particle cloud with the incoming measurement to re-acquire objects returning after an orbit, raising the share of repeat passes matched to their existing track from 0 of 17 to 17 of 17, while the gating algorithm and related optimizations cut the run time of a 30-orbit simulation of 1,000 objects from 1,299 s to 136 s. For sensor tasking, debris objects are found to be so scarce per sensor that the problem separates into two parts. Searching for new objects can be treated as a pointing schedule fixed before the run, which is tuned by gradient ascent in a simplified environment that does not need the filter. Following known objects is handled by a custody rule that points a sensor where the filter predicts an object will pass. For comparison, an expert policy is first constructed in a perfect knowledge environment. The tuned schedule, with and without the custody rule, is then compared in the full filter against the expert policy and a random baseline, by the share of objects found and the share still tracked after 30 orbits. The tuned schedule finds only two to four percentage points more objects than the random baseline, while the expert policy finds all of them, so search is limited by what the sensors know and not by how the schedule is tuned. With a 45 degree half-angle field of view the custody rule raises the share of objects tracked after 30 orbits from 19% to 44%. The integration of the Labeled Multi-Bernoulli filter and sensor tasking algorithm in the context of space-based sensing represents a crucial step towards real-world implementation and successfully tracking and cataloging debris in the LNT range at scale.

This paper presents a novel orbital debris remediation technology that enables precise, just-in-time modification of debris trajectories for collision avoidance and accelerated deorbiting of small objects. The core concept relies on the detonation or deflagration of metalized energetic mixtures to generate and inject clouds of nanometric particles into targeted orbital points. Acting as a transient momentum trap, the nanoparticle cloud leverages high particle count densities to transfer momentum to the targeted debris without requiring their capture or direct contact with a remediation vehicle. This approach delivers a controlled deceleration of orbital objects covering a wide range of sizes and mass-to-frontal area ratios while minimizing the risk of catastrophic fragmentation. An air-launched sounding rocket charge delivery architecture provides the operational flexibility required to rapidly achieve matching orbital inclinations, enabling localized interventions that can displace threatening objects out of predicted collision corridors within six hours by shortening their orbital period, a technique referred to as Just-in-Time Collision Avoidance (JCA). Modeling based on the detonation of a 200-kg ammonal (NH4NO3+Al) charge shows that, for an initial 500-km altitude circular orbit after a 6-hour orbital propagation time, an along-orbit miss distance of 5 kilometers requires a tangential velocity change of 77.2 mm/s. In addition, reduced-order scaling analysis shows that the optimal trigger distance and maximum cross-track offset tolerance both vary with the square root of a defined momentum-response parameter, χ = C_m A/m. When applied to 15,884 low-Earth-orbit intact objects in the ESA DISCOS catalog, the JCA criterion is satisfied by 68.2% of all cataloged objects for a momentum-transfer efficiency of C_m = 1, assuming a cross-track tolerance of ±200 m and a miss distance of 5 km. For smaller debris, whose state vectors are often less precisely known, remediation is more efficiently achieved through longer-lived nanoparticle clouds produced by deflagrating compositions. Thermodynamic calculations for Mg-S indicate that about 95% of the product mass can remain in the condensed phase after expansion. A converging-diverging nozzle followed by a long conical extension is then used to produce elongated, weakly diverging particle plumes. The proposed deployment method positions the cloud’s apogee near the target orbital altitude, providing a residence time of approximately 2.4 s within a ±20 m vertical window and an interception corridor of approximately 20 km along the orbit. This method is particularly well-suited for rapid debris remediation at post-collision orbital pinch points within hours of a major breakup event. For a 1000-m interaction path and C_m = 2, 73.37% of the combined Cosmos 2251 and Iridium 33 fragment population satisfies a 70% orbital-lifetime-reduction criterion. These results support the proof-of-principle physical feasibility of using transient nanoparticle clouds for targeted collision avoidance as well as for the collective deorbiting of space debris. The predicted performance remains, however, strongly dependent on the efficiency with which nanoparticle clouds transfer momentum, underscoring the need for further experimental characterization of hypervelocity nanoparticle impacts.

The increasing population of objects in low Earth orbit has created a need for efficient methods for debris detection, tracking, and removal. Optical observation provides a rela-tively low-cost approach for identifying debris targets, but detection alone does not provide sufficient information for selecting objects for active deorbiting. This study develops an integrated framework that combines morphological op-tical image processing, trajectory estimation, and electro-dynamic tether-based deorbit analysis. Candidate debris objects are extracted from optical images using image pre-processing, threshold-based segmentation, and morphologi-cal blob analysis. The detected centroid measurements are then used for trajectory estimation using a Kalman filtering approach. Based on the estimated orbital state, candidate objects are evaluated for their suitability for electrodynam-ic tether-assisted deorbiting. The tether model considers the induced electromotive force, current generation, and Lo-rentz force resulting from the interaction between the con-ducting tether and Earth's magnetic field. The proposed framework is evaluated through image-processing experi-ments and orbital simulations to investigate the relationship between detection accuracy, target selection, tether param-eters, and orbital decay. The study aims to provide a com-putationally lightweight link between ground-based optical space surveillance and future tether-assisted debris-removal missions.

We develop a preliminary mission concept for demonstrating a gravity tractor (GT) by changing the mutual orbit of the secondary in a binary near-Earth asteroid. The study combines target screening, interplanetary trajectory optimization, controlled N-body simulations, deflection-detectability analysis, and preliminary payload and operations design. We compiled 124 binary systems, applied a nondimensional GT-effectiveness metric to down-select to13 targets, and optimized chemical-propulsion rendezvous trajectories for 2035--2040 launches subject to a Falcon 9-class launch constraint and a 1000-day maximum cruise. Three systems emerged: (350751) 2002 AW, (162000) 1990 OS, and 2020 AZ2, with estimated maximum post-rendezvous spacecraft masses of approximately 1400 kg, 1200 kg, and 900 kg, respectively. We modified an N-body collisional code, pkdgrav, to propagate each binary with a controlled spacecraft that counters binary gravity, maintains a dead-banded position and velocity relative to the secondary, enforces plume-clearance geometry, and tracks propellant use. For 60-day tractoring cases, the simulations produce secondary semi-major axis changes of approximately 10-20 m and orbital period changes of approximately 300-1500 s. A 10\% mutual-orbit phase offset 30 days after tractoring requires period changes of 322, 222, and 248 s for 2002 AW, 1990 OS, and 2020 AZ2, respectively, and consumes approximately 40 kg of propellant. These results identify 2002 AW and 1990 OS as the leading targets and motivate a two-phase tractoring demonstration at a conservative and a closer operating distance. A camera, rangefinder, and radio-science payload and a proximity-operations campaign of less than one year provide the corresponding measurement and operations framework for demonstrating the GT.

Earth continues to be at risk from impacting asteroids and comets. Near-Earth Object surveys scan the sky each night with the principal goal to find larger objects that may pose some future danger. Although impact by larger objects is quite rare, smaller objects do hit occasionally, and there have been numerous recent cases of small (~1-meter diameter) objects being discovered hours before impact. While each survey uses their own search strategy, the appearance of an impacting object changes dramatically depending on when in the night it is observed, when it is close, and where it will impact. We examine some of the impacting objects to look at their Earth-approaching signatures to better understand why Pan-STARRS and other surveys may have trouble discovering these objects. We also extend our analysis to larger impacting objects that would be observable a few days before impact to assess their appearance.

A demonstration mission of rapid response asteroid characterization is the highest priority planetary defense mission following DART and NEO Surveyor. Near Earth Object (NEO) threats can appear with little warning, and rapid response missions must be capable of launch readiness within short time frames compared to typical missions. Use of commercial off the shelf (COTS) spacecraft and payloads can help reduce risk and cost, if their capability is proven in on-orbit demonstrations. Impulse Space conducted a study to see if Mira and commercially available optical payloads could meet the trajectory and measurement requirements of a rapid response planetary defense mission, and developed two design reference missions to inform a future demonstration mission. It was found that the Mira spacecraft, designed for LEO, GEO, and cislunar space, could be altered for operations in deep space. Measurement requirements for NEO characterization can be partially met by commercially available optical payloads, although more custom solutions may be needed for spectroscopy characterization. The ability to track a potential target through a fast flyby is dependent on both bus slewing capability and the technical specifications of the optics on board, and NEO surface characterization requirements are a driver. Once the capability of the bus and payloads was explored, trajectories to potential NEO targets of size 50–100 m were analyzed for a launch window of one year, with a maximum time of flight of 2.5 years. Two targets were selected, each for a different mission concept. 2024 YR4, an asteroid significant to planetary defense, was chosen for a flyby demonstration mission. 2015 YU1 was chosen for a NEO flyby – Earth flyby – NEO rendezvous mission, to compare the data collected between a fast, 5 km/sec relative velocity flyby to what could be collected during a long duration rendezvous. Analysis of NEO orbit uncertainty on flyby miss distance and required trajectory corrections also suggests a rapid response mission would benefit from a highly maneuverable spacecraft, as well as early measurements to help reduce target uncertainty, from either the reconnaissance spacecraft itself or other systems. These results highlight the utility of commercial, maneuverable spacecraft and COTS payloads for enabling rapid response planetary defense missions.

Underactuated robotic grippers present a novel solution for the manipulation of non-cooperative objects, particularly in the unstructured environment of space. Where rigid grippers fail due to reliance on precise positional data and fixed geometry, compliant grippers are capable of passively adapting to irregular geometries, requiring only approximate pose information to achieve a successful grasp. Despite their potential, adaptive grippers for the capture of non-cooperative objects remains an underrepresented area in literature. This work proposes the design, simulation, and validation of an underactuated robotic gripper for the non-destructive acquisition of non-cooperative objects in space. The design of this gripper follows a three-finger grasping modality, with spring-based flexure hinges that provide passive compliance. Underactuation of this gripper enables passive adaptation to irregular debris geometry, without fine-tuned actuator input. This approach facilitates successful grasping with no complex sensing or precise positioning. In this paper we detail the development, simulation, and evaluation of the adaptive gripper, mounted to a seven degree of freedom robotic arm for testing. We detail the methodology behind the tendon-driven design and the integration of prototype parameters into simulation. Further, we provide qualitative and quantitative analysis of grasping success rate, simulation accuracy, and estimated contact forces during grasping trials. The underactuation reduces system complexity without sacrificing reliability, and presents a viable alternative for debris capture where object geometry and position are uncertain.

Surface navigation on the Moon is subject to extreme illumination contrast and deployment-time perceptual disturbances. While modular navigation pipelines have long supported planetary rover navigation, robustness is compromised by measurement errors that propagate through hand engineered components. End-to-end learned policies are an alternative approach to navigation, which maps observations directly to control actions without intermediate components. Vision-language-action models, in particular, have driven significant breakthroughs in robotics and autonomous vehicle research. By leveraging internet-scale vision-language pretraining, these models can be exploited for their commonsense reasoning and visual understanding to provide robust generalization with limited task-specific data. However, their applicability and performance have yet to be characterized for lunar surface operations. In this study, we design a vision-language-action policy for autonomous lunar rover docking and characterize its performance relative to a modular navigation pipeline. We develop a scalable training framework that generates safety-aware expert demonstrations in a high-fidelity lunar simulation environment. In closed-loop evaluations, our policy performs competitively relative to the modular baseline under perceptual disturbances unseen during training, achieving 57–71% docking success rates in the most challenging conditions compared with 0–21% for the modular baseline.

Recent developments in modular satellites enable the vision of robotic servicing of satellites for life extension or to add novel capabilities. Challenging environmental conditions, however, render perception and control in outer space difficult and potentially lead to failure. We therefore suggest a different approach to robotic servicing: instead of trying to automate everything, we intend to keep the human in the loop throughout the whole manipulation. Leveraging transparent and stable teleoperation controllers and modern shared control methods, we create a holistic framework that provides force cues to a human operator from the remote environment and an augmentation using Virtual Fixtures. These supportive fixtures can be created from few perceived or demonstrated points, corrected and improved by the human operator and finally be used either in direct teleoperation or executed autonomously on the servicing robot. In our work, we consider the manipulation phase once the client satellite has been successfully docked. In this phase, as part of a life extension or servicing mission, an Orbital Replacement Unit (ORU) is to be assembled to the client satellite. Based on via points selected from current end effector location, ORU location and the placement location on the target satellite, a Virtual Fixture is created to help the human operator to perform the manipulation. The manipulation is first performed with simulated remote robot, where the operator can adapt the fixture to avoid potential collisions and optimize the manipulation. The optimized fixture is then sent to the real robot, where the full operation is performed with simulated time delay of 800ms as expected on a real satellite servicing mission. Our proposed framework is implemented using a torque-sensitive impedance-controlled servicer robot on the On-Orbit Servicing Simulator (OOS-Sim) testbed at the German Aerospace Center (DLR) coupled with a haptic input device for the human operator. We evaluate our approach on the use case of assembling an Orbital Replacement Unit (ORU) from the servicer satellite to the client satellite. Using the iBOSS iSSY interface, this ORU is mechanically and electrically coupled to either the servicer, the gripper or the satellite, allowing for the exchange of power and information. In the remainder of the paper, we describe our design choices, algorithms, implementation details and the evaluation on the OOS-Sim testbed.

As space becomes increasingly congested and contested, robust autonomous navigation is essential for servicing, debris removal, and situational awareness with respect to unknown resident space objects. This work presents a novel learning-based pipeline for pose estimation of uncooperative and unknown spacecraft using only monocular images from a single servicer. The approach combines a transformer-based neural network with a Multi-State Constraint Kalman Filter (MSCKF) to estimate the pose (i.e., position and orientation of the target spacecraft relative to the camera) throughout rendezvous and proximity operations. Unlike existing vision-based methods that require prior knowledge of the target shape or inertia properties, rely on additional sensing modalities such as depth, lidar, or stereo, or only recover translation up to scale, the proposed pipeline generalizes to previously unseen spacecraft using a single monocular camera. The transformer network estimates the odometry, the change in pose between images up to scale, from SuperPoint features matched by LightGlue. These odometry estimates serve as MSCKF pseudo-measurements. The MSCKF uses these pseudo-measurements along with an orbit and attitude kinematics model to estimate the pose of the target. In particular, the relative orbit elements, the target's attitude with respect to the servicer's camera (where the target's body is defined arbitrarily by the filter), and the associated angular velocity are estimated directly by the filter. Given the monocular approach and short distance to the target, the full observability of the range to the target is recovered via attitude maneuvers by the servicer. Feature tracks and loop closures are used to bound short- and long-term drift, respectively. The filter is initialized using coarse relative orbit information that could be available from upstream methods such as angles-only navigation, and no other prior such as attitude, shape, size or inertia properties. The method is trained and evaluated on a re-rendered high-resolution version of the SPE3R dataset, which includes synthetic images of 103 spacecraft. Eleven of these spacecraft are held out during training to evaluate the generalization to unseen targets. Monte Carlo simulations are then used to evaluate the navigation pipeline on rendered trajectories of the held out spacecraft. The odometry front-end is further evaluated on hardware-in-the-loop images from the SHIRT dataset, where the network retains a $4$--$6$ fold advantage in the median translation direction error over a classical baseline. The results demonstrate that learned vision pipelines as a front-end for Kalman filters provide median errors of $3.7^\circ$ in attitude and $2.2\%$ of range in ROE when navigating about unknown targets. The approach supports inspection, servicing, and debris removal missions involving targets without available shape models.

Large apertures for communications, radar, and astronomy increasingly exceed the launch fairings that carry them, motivating a shift from monolithic deployment toward on-orbit robotic assembly of modular elements. Architecture studies and laboratory demonstrations have established the feasibility of modular robotic assembly, but the demonstrated aperture systems fasten each element to a central structure or base plate and leave neighboring elements unjoined, so the assembled surface is not a single rigid body. This paper presents the hardware architecture of MOSAIC (Modular On-orbit Scalable Apertures from Interlocking Components), a parabolic antenna assembled by a single robotic arm, and demonstrates the robotic skills its assembly requires. The antenna is tessellated into hexagonal tiles, identical within each ring, that embed captive, non-back-drivable bolted fasteners joining each tile to the central hub and to its neighbors. A single end effector, a coupler with a screwdriver at a fixed offset, grasps a tile and engages one of its fastener inputs in the same motion, so that the arm places and fastens every tile with no tool change and no free parts, and a rotating central hub brings every placement into one sector of the arm's workspace. Assembly is decomposed into four skills, grasp, alignment, compliant insertion, and fastening, built from two interpretable primitives: fiducial localization with consensus checking, and admittance-controlled mating that ends each motion on a measured contact preload rather than at the estimated pose. On a laboratory testbed with a UR10e manipulator and prototype tiles, the grasp succeeded in 25 of 25 trials over a +/-50 mm variation in the presented tile position, and in separate trials the insertion and fastening of a tile to the central hub succeeded in 23 of 25, with the compliance of the insertion, rather than the accuracy of the estimate, absorbing the residual destination error. The timed bolt drive, rather than the motion of the arm, dominates the estimated per-tile cycle of about 145 s. A scalability analysis shows that stowed mass and volume per unit aperture are constant across the tile family, at 10.3 kg/m^2, and that the fastening effort grows linearly with tile count. The design is intended to be general, and the instance reported here was developed toward a sub-scale demonstration on NASA Goddard's Fly Foundational Robots (FFR) mission, for which we identify the failure modes and the remaining steps on the path to flight.

It has been known since the seminal work of Neustadt in the 1960s that the optimal control of spacecraft under fuel-consumption metrics is impulsive. This property is often implicitly known, yet not adequately exploited in theoretical or algorithmic frameworks of optimal control of spacecraft. This paper addresses an increasingly important gap in spacecraft literature for extending impulsive optimal control to pursuit-evasion differential games and demonstrates the relevance of such formulations for spacecraft relative motion scenarios. This includes studying the theoretical properties of several useful pursuit-evasion game formulations, which are shown to have important structure that not only inform improved numerical schemes for solving for Nash-equilibrium policies, but have important implications to the nature of spacecraft co-orbital games. This paper derives feedforward-optimal input profiles for terminal distance pursuit-evasion games. An important property of norm-based games arises, which is that the ordering of actions affects the optimal play. This feature is addressed with the introduction of a novel feedback policy which consists of solving a convex, semi-infinite program in feedback MPC form. It is conjectured that this policy is optimal, i.e., non-exploitable. Several numerical results of the newly proposed game formulations are demonstrated.

Autonomous multi-spacecraft docking is an increasingly important capability with the advent of on-orbit servicing and manufacture as well as distributed spacecraft operations. Proximity operations are safety-critical control problems that become more difficult as the number of spacecraft involved increases. Multiple chaser spacecraft must dock individually at assigned ports while maintaining collision-free trajectories under realistic inter-agent communication constraints. Multi-agent reinforcement learning (MARL) with centralised training and decentralised execution (CTDE) can synthesise decentralised policies feasible for on-board execution. However, MARL optimises expected return and does not provide hard safety guarantees against catastrophic events, such as collisions. We investigate whether control barrier functions (CBFs) can be used with communication-constrained MARL during training and deployment to enforce safety while preserving the performance of the learned policies. A six-degree-of-freedom (6-DOF) multi-agent docking environment is developed using the Clohessy–Wiltshire relative-orbit dynamics equations and quaternion attitude kinematics. A proof-of-concept formulation is considered, where nine 6U CubeSat-sized chasers must simultaneously dock at a planar-faced target whose docking ports are spaced 1.1 m apart. Each chaser has per-axis linear and angular acceleration limits of 0.005 m/s² and 0.02 rad/s², and controls its position and orientation to dock under realistic constraints, including bounds on relative velocity and angular rate, as well as keep-out limits for collisions. Safety is enforced by high-order control barrier functions (HOCBFs). The work extends single-agent Proximal Policy Optimisation (PPO) docking to the multi-agent setting using Multi-Agent PPO (MAPPO) with full attitude motion. Each agent's learned policy observes only its own state and the state of its two nearest neighbours. The lack of complete observability reflects the communication constraints of a spacecraft swarm. The learned policy is combined with a quadratic-programming safety filter that minimally modifies the commands to enforce inter-agent separation, a range-dependent approach-speed limit and a target keep-out volume. Over 5000 Monte Carlo episodes across five seeds, the system achieves a 99.54% all-agents-docked success rate at a 4 cm position and 2° attitude capture gate, with zero inter-agent and target contacts, at a per-agent Δv cost 127% higher than a collision-agnostic linear-quadratic regulator (LQR) reference trajectory, which includes the cost of collision-avoidance manoeuvres. When the same reference LQR is evaluated with the safety filter, the fleet deadlocks. It avoids all inter-agent collisions but achieves no successes, demonstrating that the filter improves safety but not performance. A policy trained without the CBF filter does not advance past the easiest curriculum gate of 0.15 m in 25 million frames, and achieves 0.10% success when the filter is implemented during evaluation. It is found that the filter must be present during training for the task to be learned at all in this configuration.

This work has investigated a Multi-Agent Reinforcement Learning framework for autonomous target conditioned along-track station acquisition and reconfiguration of a two-spacecraft Low Earth Orbit formation. Relative motion has been represented through quasi-nonsingular Relative Orbital Elements, while a Multi-Agent Proximal Policy Optimisation architecture based on Centralised Training and Decentralised Execution has combined a shared decentralised Actor with a centralised Critic. Each spacecraft has received a compact eight-dimensional target-conditioned observation containing Exponential Moving Average filtered in-plane ROE components, local orbital-state information, spacecraft mass, and inter-satellite geometric information, while an exponential tracking reward and terminal station-keeping terms have shaped the discrete prograde, coast and retrograde control policy. Rather than relying on a fixed offline dataset, training trajectories have been generated online over 20 million simulated environment steps, with episode-specific along-track targets sampled from the interval [-5000,+5000]m. The present study has considered two homogeneous spacecraft and an unperturbed nominal orbital model in which J2, atmospheric drag, third-body perturbations, and solar-radiation pressure have been disabled in order to isolate the underlying phasing and learning dynamics. Following training, the frozen deterministic policy has been evaluated over 1000 independently randomised target-conditioned station-acquisition episodes, achieving an 85.6% joint terminal success rate, with no simulator-defined collision or out-of-bounds termination observed in the evaluated population. A representative closed-loop scenario has additionally compared the learned policy with a Quantisation-Aware Linear MPC and a Phasing-Targeted Linear MPC under a common reduced-order plant and ternary actuator interface, allowing targeting accuracy, control expenditure, and online computational cost to be assessed under consistent conditions. The computationally intensive policy optimisation is performed entirely offline, whereas decentralised execution requires only local observation processing and a feed-forward Actor evaluation, thereby transferring the principal optimisation burden away from online operation. The demonstrated constellation size remains limited to two spacecraft; parameter sharing and decentralised execution provide an architectural basis for extension to larger homogeneous formations, but large-constellation scalability and execution on flight-qualified processors remain subjects for future validation.

The rapid development of deep learning has significantly transformed image analysis, enabling the extraction of both geometric and semantic information from visual data. However, its application in spacecraft onboard navigation, and especially in close-proximity operations, remains limited. This is primarily due to the lack of interpretability and an absence of reliable uncertainty estimates, which together make standard neural networks unsuitable for safety-critical scenarios. To address these limitations, this paper proposes an architecture that explicitly quantifies both aleatoric (data-dependent) and epistemic (model-dependent) uncertainty for a regression task by conditioning the latent space. The latent space is structured using an energy-based model rather than contrastive learning, avoiding the need for large negative-sample sets. The model's energy is defined by the performance of a pixel-wise geometric task, image segmentation, which acts as a self-evaluating failure-mode detector: when segmentation fails, the energy term pushes the out-of-distribution sample away from the in-distribution region. The architecture is designed such that predictive confidence decreases when reliable target geometry cannot be extracted, thereby improving prediction reliability. With this structured latent space, epistemic uncertainty is quantified by applying Gaussian Discriminant Analysis to the latent features, using Mahalanobis distance to identify degraded or target-absent images. Aleatoric uncertainty and the point estimate are instead handled by a dual-head probabilistic regression layer: by minimizing the Gaussian negative log-likelihood, this layer predicts the regression targets while capturing heteroscedastic uncertainty, allowing the predicted variance to scale dynamically with sensor noise and varying target conditions. Together, these components provide uncertainty-aware navigation without relying on ensemble-based Bayesian approximations or more computationally demanding probabilistic models such as Gaussian Processes or Deep Kernel Learning. This architecture is tested on a representative task for proximity operations: regression of the line-of-sight and range of a chaser with respect to a known target. Due to the lack of annotated space imagery, the network is trained on synthetic visible-spectrum images. Point estimates are evaluated on synthetic and laboratory imagery, while real space imagery is used only for epistemic uncertainty analysis, since it lacks pose annotation. A dataset of relative trajectories is additionally generated and rendered to test the effectiveness of the uncertainty estimates when coupled with a relative navigation filter. Both the relative trajectory and training datasets are released as an additional contribution of the work. The proposed framework demonstrates that energy-conditioned latent representations combined with lightweight uncertainty estimation can improve the reliability of monocular vision-based navigation for autonomous rendezvous and proximity operations.

One approach to active debris removal is to capture the debris object with a robotic arm mounted on a servicing spacecraft. Approaching and capturing the target requires precise, coordinated control of the free-floating platform and the manipulator, respecting their coupled dynamics. The spacecraft's position and attitude are commonly controlled with binary thrusters, which can only be switched on or off and are additionally subject to dwell-time constraints (minimum on time, cooldown period post usage), and reaction wheels. This discrete switching behavior poses a fundamental challenge for optimization-based real-time control. Conventional approaches approximate the discrete thruster behavior with continuous inputs, which are then mapped to valve commands via pulse modulation. This preserves numerical efficiency but sacrifices optimality and can violate timing constraints near the capture phase, where control precision matters most. The nonlinear dynamics of the robotic arm further increase problem complexity. This work formulates the coupled base–manipulator control problem as a binary-aware nonlinear program and employs nonlinear model predictive control (MPC) for closed-loop approach and capture. It builds up on a binary-aware formulation using Delta-Sigma-modulation which converts a continuous control signal to binary commands. To remain real-time feasible, a cascaded-fidelity, multi-phase MPC scheme is proposed. Over a short horizon, the binary-aware mixed-integer (MI) formulation with a full dynamic model including exact switching and minimum-time constraints is solved, while the long horizon is approximated with a single rigid body model and continuous relaxation of the constraints. The online optimization is warm started from previous solutions and runs for a fixed number of iterations to bound solve times. The end-effector is driven to a specified grasp pose on the target through terminal constraints such as end-effector velocity and look-at constraints, coupling the thruster scheduling directly to the capture objective. The approach is implemented in a simulation environment of a spacecraft with an attached 7-degree-of-freedom manipulator. We evaluate the optimality gap of the multi-phase scheme relative to a full MI-MPC over the entire horizon, its performance against the conventional relaxation-plus-pulse-modulation approach in terms of fuel consumption, constraint adherence, and capture-pose accuracy, and the trade-off between full-body dynamics short-horizon length and computational cost, characterizing the region in which the scheme remains real-time feasible. We quantify how much of the optimality the multi-phase formulation recovers of the exact MI solution at a fraction of its computational cost, enabling MI-based thruster control at rates suitable for on-board closed-loop capture operations.

This paper assesses the feasibility of a robotically assembled multi-aperture reflector exceeding 100 meters in diameter. As of yet, spaceborne reflectors have not been deployed at this scale due to the limited volume constraints of modern launch vehicles. However, in-space assembly could cross this threshold and enable greater observational capabilities in space. To this end, a robotic structural system is presented that makes use of cable actuated prism modules as building blocks for a reflective surface. The system can be assembled in a planar configuration and then reconfigured into any spherical surface by varying the ratio of top and bottom faces (the skew ratio) via cable actuation. The range of radius of curvature and reflector diameter are dictated by the prism’s skew ratio and height, respectively, yielding greater mission adaptivity than traditional ring truss designs without sacrificing structural stiffness. This work describes key governing relationships between overall surface diameter, structural mass and surface accuracy. The surface accuracy is dictated dynamically and geometrically. To quantify dynamic effects, a numerical model is developed to characterize fundamental frequency and mode shapes from cable tension and mass distribution. It is shown that cable tension can be varied across the reflective surface to optimize the prestress distribution. Geometric effects include density of points in the triangulated surface, manufacturing and tensioning errors, and robotic assembly errors. As such, a novel statistical method is developed to quantify the surface approximation variations as a function of number of modules and model parameter variations. Combining these results, several feasible reflector designs are presented and weighed against mission requirements.

Space mission concepts involving large scale structures, deep space habitats, and its related infrastructures often include lightweight, high-performance, and geometrically periodic primary structures. Recent developments show that these can be implemented with simple and low-cost assembly robots by leveraging the regularity of a mechanical metamaterial system. When composed of conductive materials such as carbon fiber reinforced polymer (CFRP) composites and aluminum, the lattice structure creates a network of electrical resistors. Measurement of electric resistance values between nodes of this structural grid may be utilized for detecting damage on structure, which is likely to change the resistance value on one of its struts. This method is suitable for automated structure evaluation, which enables the unmanned inspection of critical, large-scale structure and long-term maintenance activities. This will be significant to the reliability of the overall modular structural assembly system. In this paper, we investigate the feasibility of fracture detection by resistance measurement. We first collect resistance data on the structural building blocks (termed voxels) and their assemblies to analyze the distribution of resistance value. The equivalent computational model of the resistor network is compared with the collected data for verification. Then we investigate how the resistance measurement changes when there is a fracture in one of the struts, using both the computational model and experimental data. To determine the range of resistance measurements which could be interpreted as a broken voxel, we perform a statistical test using the computational model, as well as training a simple neural network model from the gathered electric resistance data. Furthermore, we explore the effects of neighboring voxels on resistance measurements, as well as the minimum number of measurements to determine the location of broken voxels. We also compare the methods to apply the health monitoring system – sensing by mobile robot and embedded sensing - in terms of mass, cost, and complexity.

Modular structural systems promise to enable autonomous In-Space Servicing, Assembly, and Manufacturing (ISAM) to realize lightweight, reconfigurable structures that can be assembled at scales far beyond the constraints of launch vehicle fairings. Recent work demonstrates systems based on a three-dimensional structural module, called a voxel as a volumetric pixel, with high specific stiffness, reversible connectivity, and a regular geometry that enables robotic agents to navigate and manipulate the structure using the lattice itself as a positioning reference. However, transporting complete voxels is volumetrically inefficient, making autonomous fabrication from flat-packed components a key challenge for scalable in-space assembly. To understand where this capability matters most, we first compare four archetypes of lunar construction at equal structural function: conventional construction performed by humanoid robots, an onsite prefabrication factory, a deployable Earth-prefabricated control case, and assembly of discrete modular lattice structures by relative robots. Because delivery from Earth amplifies the cost of onsite resources by two to three orders of magnitude, delivered mass and stowed volume become the governing objectives. A dimensional scaling analysis shows that the reach of a builder is limited by geometry rather than by mass budget, and that robots indexing to local lattice features avoid errors that grow with structure span. Discrete modular structures carry a mass penalty relative to monolithic ones, but this penalty diminishes with span, and reuse across builds can offset it. Together, these results show that the value of a modular, relative-robotic approach depends on delivering components in compact form and producing the modules themselves at the point of use. In-situ assembly of the modules themselves is therefore a high-leverage capability.

Mathematical proofs are available showing Euler angles developed with current methods for orientation parameterization cannot be mutually orthogonal, since single sequences of three rotations around moving coordinate axes inevitably results in non-orthogonal coordinate axes. A topological proof is embodied in the so–called hairy ball theorem that uses four–dimensional parameterizations (quaternions or angle–axis representations) to demonstrate inability to assign three independent, continuous angular parameters. Another proof illustrates mathematical breakdown when using three sequential elemental coordinate frame rotations where a rotation axis aligns one of the other two axes. At this point called gimbal lock, one degree of freedom is lost, and the mathematical matrix representation drops from rank three to rank two resulting in singularities. A new method eliminating gimbal lock singularities is proposed by parameterizing orientation using three (as opposed to one) sequences of three consecutive rotations leading to the respective elaboration of three mutually orthogonal Euler angles representing orientation angular rotations about body–fixed axes (not inertial or intermediate axes). Following heuristic and analytic development, validation is provided by space experiments flown in May 2026 using error plots of the difference between classical and proposed mutually orthogonal Euler angles. Inertial measurement unit drift sources are seen using classical Euler angles, but not with the proposed Euler angles.

Industry is beginning to recognize the economic and operational benefits of in-orbit servicing. However, the risks inherent to close-proximity operations require expensive in-flight validation, restricting the sector to large institutional actors. Small spacecraft such as CubeSats offer a low-cost demonstration platform, yet at this scale the capture envelope of a docking interface shrinks while sensor and acquisition noise grow. The feasibility of CubeSat docking therefore hinges on the tolerances of the interface itself, which can only be established with confidence through capture simulations validated against experiment. No CubeSat-scale interface has been characterized to this level, and docking and refilling have yet to be demonstrated at this scale. We present a docking and propellant-transfer interface for CubeSats using water electrolysis propulsion, designed to leverage the tuna can volume. The interface employs a three-arm mechanism that achieves both capture and propellant transfer, driven by a single linear motion. Its docking envelope is characterized using a combination of experiments and kinematic multi-body simulations. Surface contact parameters are determined experimentally through dedicated impact and sliding tests on representative material samples. Their influence on docking outcome is evaluated by reproducing a counterbalanced docking test bench. The validated simulation model is then used to characterize the misalignment and misplacement tolerance of the interface under representative microgravity conditions. A capture envelope of at least 12 mm is established at worst-case angular misalignments of 4 degrees. From this envelope, attitude determination and control requirements are derived. A redundant actuation and sensing architecture, comprising a cold-gas thruster array and a sensor suite built from commercial off-the-shelf components, is presented. Its minimum impulse bit, thrust, and sensor precision satisfy the derived requirements, with safe approach and capture demonstrated in a dynamic simulation. The presented methodology shows that the docking and propellant-transfer interface matches the capture envelope of larger, heavier alternatives at reduced size and mass. Beyond this specific design, it compares the relative importance of contact parameters governing docking behavior at the CubeSat scale and identifies low-cost means of validating them. The derived attitude control requirements inform the component selection for a demonstration mission using two 16U CubeSats. The mission would constitute the first flight of six-degree-of-freedom attitude control by water electrolysis, forming part of the Ice2Thrust in-orbit demonstration.

Limited onboard power and compute motivate separating spacecraft manipulation into high-level planning and lightweight local decisions. This separation could support edge execution, but its value depends on reliable command delivery and efficient physical execution, not response speed alone. Motivated by terminal sample transfer, we evaluate the local decision boundary in a terrestrial robotic-arm simulation. Claude Sonnet 4.6 supplies shared plans and stall-triggered recovery; local selection uses either its enum-constrained tool or TypeSafe Jev 1.13.0 with seven frozen reactive primitives. The evaluation contains 750 episodes across three Meta-World MT1 task banks and five controllers, including 150 matched pairs for the primary comparison. Decisions occur every 20 physics steps under a 500-step horizon. Jev-based selection raises completion from 140/150 to 150/150, a 6.67-percentage-point increase. All ten baseline failures lack the required tool output under a 120-token cap; incomplete logs prevent causal diagnosis. Among 140 jointly successful pairs, mean steps instead rise from 69.66 to 71.37. Paired accounting attributes the entire failurecapped step advantage to interface-failure pairs. Median validrequest latency falls from 1,241 to 364 ms, yet pacing increases mean episode time from 5.50 to 24.68 s. Deterministic rules match 150/150 success at 0.109 s per episode. Thus, the tested substitution improves observed completion, not motion or end-toend efficiency. These results favor retaining deterministic local control as the reference for resource-constrained designs.

Physical autonomous agents can encounter contact conditions that are poorly represented by the model used during training. This paper examines whether a biomechanical elasticity representation can be transformed into a physical prior for compliant capture of a tumbling, non-cooperative spacecraft. Bio-Dissipative Prior Transfer (BDPT) formalizes the transformation as a source representation, an engineering transfer operator, and a downstream physical model whose behavior is evaluated against a matched conventional prior. The present study is restricted to compliant spacecraft capture. The executed reduced-order pilot uses a four finger, eight-joint tendon-actuated gripper, 36-dimensional observations, a temperature-dependent synthetic Hill-type elasticity prior, PPO training with 64 parallel environments and 60 updates, and 1,500 held-out trials per controller. Bio-PPO lowered mean peak contact force from 1.299 N to 0.775 N, a 40.3% relative reduction, while capture success decreased from 38.3% to 28.7%. The archived artifact contains aggregate force summaries and rounded success rates rather than trial-level logs, so median and interquartile range cannot be reconstructed and the success-rate confidence intervals and two-proportion statistic are reported only as approximate, non-confirmatory quantities. The pilot therefore establishes a measurable force–retention trade-off rather than a causal advantage for the biological prior. A matched conventional-prior ablation with identical physics guidance, independent training-seed records, and recovery of the trial-level logs are required for a confirmatory causal result. The contribution is consequently an auditable spacecraft capture instantiation of BDPT with explicit physical equations, numerical verification, statistical boundaries, and falsifiable acceptance criteria.

This paper demonstrates that deterministic sampling clock offsets between RFSoC-based ADC channels can be compensated using array calibration and digital equalization, reducing reliance on tightly phase‑matched and temperature stable clock distribution. A dual‑RFSoC testbed with a controllable delay in one sampling clock path validated a model relating sampling clock phase offset to RF phase shift. Measured 501 MHz tone phase shifts agreed with predictions from 1.6 GHz clock offsets, and a 31‑tap FIR equalizer successfully aligned channels for coherent combining. These results support lower‑SWaP, more scalable digital beamforming and dispersed‑sensor architectures.

Wirelessly distributed low-Earth-orbit (LEO) satellite swarms flying in formation can act as a single large phased array. By coordinating thousands of small satellite elements at half-wavelength spacing, an aperture far larger than that of any single spacecraft is synthesized. For instance, 10,000 elements at S-band can form an aperture about 8 meters across, enabling direct-to-device (D2D) links to ordinary handsets. It is commonly expected that coherently combining N elements yields an array gain of 10log10(N) (40 dB for N=10,000). This paper quantifies the array gain actually achievable when the array forms multiple simultaneous beams and is fed by a wireless fronthaul rather than physical wiring. We consider a regenerative architecture: a parent satellite performs digital beamforming (DBF), precoding the K beam signals into M fronthaul signals, and distributes them over a millimeter-wave (24 GHz) wireless fronthaul to N (approx. 10,000) child satellites (K < M < N), each equipped solely with analog phase/amplitude controllers and frequency converters. The M fronthaul streams are multiplexed across the polarization, frequency, and spatial domains. We model the downlink as a cascade of the digital precoder, the wireless-fronthaul channel, the per-element analog weights, and the S-band steering matrix, assuming isotropic elements and ideal phase coherence with known element positions (element patterns, attitude, and calibration errors are deferred). Using zero forcing and minimum mean square error (MMSE) precoding with alternating optimization of the constant-modulus analog weights, we compute the realized per-beam array gain (the directivity of the resulting element excitation) for 600/1000 km altitudes, up to ~200 beams on a ground grid, and ~10,000 elements. We present three main results. First, for a single beam the full gain is recovered exactly (40.0 dB at N=10,000), confirming the model. Second, for many simultaneous beams the gain saturates well below the ideal — about 26 dB at 100 beams, even after optimization. Third, this saturation is set neither by the element count N nor by the fronthaul multiplexing order M: the gain is essentially independent of N (about 26 dB at both 2,000 and 10,000 elements), and increasing M improves the channel conditioning but fails to elevate the gain ceiling. This performance limitation originates from the multi-beam analog constraint: each child radiates a single signal that must serve all beams at once, so the key design parameter is the per-child analog degrees of freedom, not N or the number of parent antennas. In contrast, an idealized architecture employing an ideal fronthaul with per-beam linear superposition recovers nearly the full gain (39.9 dB at N=10,000), confirming that the loss comes from the analog, single-output, wirelessly-fed realization, not from multi-beam operation itself. These results provide a quantitative array-gain budget for swarm-based distributed phased arrays and reveal the limiting factors of multi-beam operation. As a next step, future work will convert this scale-invariant directivity into EIRP under per-element power constraints to construct a D2D link budget.

Direct-to-cell satellite communication requires large antenna apertures to overcome the severe link-budget constraints of handheld terminals. Formation Flying Phased Array Antennas (FFPAA) address this limitation by using a cluster of small satellites as a distributed large-aperture array. This paper studies spatio-temporal ground-coverage optimization for simultaneous multi-beam transmission by an FFPAA. As the formation moves along its orbit, the geometry to each ground cell varies over time, while co-frequency interference and fixed total transmit power limit the number of beams that can be formed simultaneously. We model link quality by the signal-to-interference-plus-noise ratio (SINR) and formulate a coverage problem over an entire satellite pass, where a cell is counted as served only if its SINR remains above a required threshold continuously for at least a minimum duration. With geometric steering and fixed beam amplitudes, received powers can be precomputed, reducing the problem to an integer linear program whose only binary decision variables are the beam assignments. We compare an integer programming solver, a greedy method, and a randomized multi-start construction on instances with 56 to 1422 cells. At 56 and 88 cells, the greedy method reaches an independently derived upper bound in under one second, whereas at 1422 cells interference and power dilution limit the achieved objective to about 35\% of the bound. The relative performance of randomized construction depends strongly on the minimum continuous illumination time: at 1422 cells, its objective relative to deterministic greedy allocation increases from 0.90 for a one-step (10 s) requirement to 2.92 for a twelve-step (120 s) requirement. These results identify the operating regimes in which solver-based optimization, deterministic greedy allocation, and randomized construction are effective.

A Formation-Flying Phased-Array Antenna (FFPAA) consists of many small satellites flying in formation, creating a single large aperture. Accurately simulating an FFPAA's radiation pattern is crucial in order to analyze the effects of various design parameters. The published analyses of similar satellite arrays that we reviewed compute that pattern with common array-theory methods and do not verify it against a full-wave solution, which may leave a gap between the simulation and the actual performance. This paper presents a detailed comparison of commonly used radiation-pattern simulation methods, run on sequentially rotated circular microstrip patch arrays on a half-wavelength square lattice at 1.8~GHz. Three simulators are run on identical arrays: an array-factor simulator, a Method of Moments with Macro Basis Functions (MBF-MoM) simulator, and the full-wave electromagnetic solver Ansys HFSS as the reference. The two reduced-order simulators differ in exactly one term, the excitation each radiates, which isolates mutual coupling in the excitation weights. From 4 to 256 elements the two agree with each other to a few hundredths of a decibel in peak gain. Both overestimate the full-wave broadside peak gain by a similar amount. On 14 further full-wave arrays with element position and pose errors, the errors move the full-wave pattern much further than the choice between the two reduced-order simulators does. The three simulation methods offer a trade-off between compute time and modeled accuracy. Simulating a 1,024-element full-wave reference took 80.7 hours on 12 cores with 144.0~GB of peak memory per process. We conclude that each simulation method belongs at a different design stage.

Renewed international and commercial interest in lunar exploration is rapidly increasing the number of spacecraft operating in cislunar space. The Artemis program and commercial lunar missions will further expand this population, increasing the importance of cislunar Space Domain Awareness (SDA). This paper investigates the feasibility of passive radio-frequency sensing using existing large-aperture radio telescopes for cislunar SDA applications. Observations of cislunar space present significant challenges due to the large distances involved and solar illumination. Cislunar trajectories can exhibit strong dynamical sensitivity, making long-term custody challenging. Cislunar sensing technologies include both active and passive electro-optical and radio-frequency methods. Active methods in electro-optical (laser ranging) and RF (radar) present a particular challenge due to the prohibitive power levels required. Passive electro-optical observations suffer from the issue that even though the object to be tracked is geometrically accessible, there may be poor illumination by the Sun.  Additionally, the high albedo of the Moon presents dynamic range challenges for passive optical observations. Passive RF sensing offers an alternative approach because the spacecraft itself provides the illuminating source through its communications transmissions, requiring only a receiving system on Earth. In preparation for the return to the Moon, the NASA CAPSTONE mission serves as a pathfinder to study the orbit of the future Lunar Gateway. We evaluate the feasibility of using the CSIRO Mopra 22 m Radio Telescope at Coonabarabran (NSW, Australia) for the detection of RF emissions from lunar-orbiting spacecraft at X-band and S-band. The results from an observational program are presented and demonstrate that RF emissions from CAPSTONE and other spacecraft including KPLO and LRO are detected during periods of contact with the NASA DSN stations at Tidbinbilla, Australia and Goldstone, California. RF signal analysis detailing pattern of life is presented. Doppler-shifted signals from spacecraft transmissions are analyzed and compared with reference spacecraft ephemerides to place bounds on range rate accuracy. Using only a single receiving antenna range rate is constrained within 10 m/s error. RF emissions from CAPSTONE, LRO and KPLO are presented with SNR ranging from 35 dB to 15 dB. Observations demonstrating simultaneous Doppler signatures from multiple spacecraft within a single antenna beam are also presented. Signal processing considerations along with timing issues and spacecraft modes of RF transmission are also discussed. Orbit Determination using NASA's General Mission Analysis Tool (GMAT) is presented along with a discussion of the challenges of the scheduling and effectiveness of passive RF collection. The key contribution of this paper is the demonstration of an end-to-end passive RF sensing pipeline encompassing observation scheduling, data collection, signal processing, Doppler analysis and ingestion into NASA's GMAT to support orbital determination updates. Receive link budget considerations and the implications for smaller-aperture instruments suitable for future cislunar SDA applications are also discussed.

This paper presents the design, fabrication, and testing of a lightweight, wideband, polarization-diverse UHF antenna array element produced using metal 3D printing. The element was developed to support UHF/VHF radar and AESA applications while addressing the limitations of traditional machining for complex, weight-optimized geometries. The design was optimized in HFSS for operation from 300 MHz to 1 GHz with a ±45° scan volume and polarization diversity. Additive manufacturing using AlSi10Mg enabled thin-walled, hollow structures and integrated weight-saving features that would be difficult to realize conventionally. Prototype measurements showed close agreement with simulations, including active VSWR below 3:1 across the operating band in an infinite array environment. Results demonstrated that the 3D-printed elements achieved the required electrical performance with no observed degradation from weight-reduction features, while providing approximately a threefold reduction in weight compared with traditionally manufactured elements of equivalent stiffness.

This paper presents the design and experimental investigation of an inflatable-rigidizable deployable dipole antenna concept for low-frequency space radio astronomy. The concept is being developed within the Low-Frequency Interferometric Radio Explorer (L-FIRE) roadmap and as a technology demonstrator for a 40 MHz pathfinder payload on a 3U CubeSat developed with Aster in Space. Each antenna arm consists of a thin truncated-cone shell in which a fiberglass-resin load-bearing layer is combined with a polymer outer layer and an LLDPE gas liner. During deployment, the folded shell is expelled, inflated to its operational geometry, and subsequently intended to rigidize by ultraviolet-initiated curing. This work focuses on the inflation phase of the deployment sequence. Four prototype shells were manufactured using silicone oils as resin analogues, spanning viscosities from 0.36 to 31.1 Pa·s. A purpose-built 2 m vacuum chamber was used to characterize inflation over differential pressures from approximately 40 Pa to 8 kPa. The experiments show that 2 kPa is sufficient to inflate the central portion of the shell, while approximately 4 kPa produces a satisfactory near-conical geometry. No discernible dependence on resin-analogue viscosity was observed within the investigated range. Instead, persistent creases introduced during manufacturing and stowage were found to dominate deployment behavior. At approximately 7.5–8 kPa, adhesive-joint failure occurred in two prototypes. The observed failures are attributed to manufacturing imperfections and contamination rather than failure of the load-bearing laminate, highlighting the importance of joint design and manufacturing quality in subsequent prototypes. The measured upper-bound gas leakage rate was 2.38×10⁻⁴ mol/s. Using this conservative value, 3.37 g of ethanol would be sufficient for a 300 s inflation/rigidization interval, while its vapor pressure of 5.83 kPa at 20 °C falls within the experimentally demonstrated inflation range. These results provide experimental support for the inflation stage of the inflatable-rigidizable antenna concept and establish quantitative requirements for improved prototypes and subsequent rigidization and deployment testing.

NASA’s Jet Propulsion Laboratory (JPL) has initiated a strategic project supporting the Artemis II moon flyby mission by employing the Green Bank Telescope (GBT) and Goldstone Solar System Radar (GSSR), currently recognized as the world’s most capable ground-based radar system, to characterize small spacecraft and objects in cis-lunar space. The primary objective of the Artemis II cis-lunar project is to demonstrate and optimize the GSSR/GBT cis-lunar configuration for the development of an operational system capable of real-time detection of cis-lunar objects in support of future lunar human missions. This paper presents an investigation and assessment of the impact of lunar surface clutter on system temperature variations affecting cis-lunar object tracking, utilizing data collected from seven radar experiments conducted as part of the Artemis II cis-lunar radar project in 2026. The findings indicate that lunar clutter significantly affects the signal-to-noise ratio (SNR) of cis-lunar target echoes. Additionally, the study evaluates variations in the Artemis II – Orion’s echo strength resulting from spacecraft dynamics and pointing, providing insights into the use of cis-lunar radar measurements for refining orbit and altitude estimations. These research outcomes are expected to enhance engineering understanding and inform technical constraints for optimizing cis-lunar radar configurations for the future detection and tracking of cis-lunar targets, including satellites, debris, asteroids, and lunar surface assets.

This work demonstrates a bistatic radar capability for cislunar spacecraft tracking and characterization using NASA Deep Space Network (DSN) 34-m antennas and the Green Bank Telescope (GBT). The capability was first validated through observations of the Lunar Reconnaissance Orbiter (LRO), which demonstrated successful spacecraft detection, Doppler estimation, and ranging under low signal-to-noise ratio (SNR) conditions consistent with link-budget predictions. The validated approach was then applied during a five-track observation campaign of the Artemis II spacecraft Integrity using DSS-13 as transmitter and the GBT as receiver. Continuous-wave (CW), binary phase-coded (BPC), and chirp waveforms were exercised across multiple cislunar geometries and spacecraft ranges between approximately 242,270 km and 399,310 km from the Earth. CW observations provided robust spacecraft detection and Doppler observables with sub-Hz frequency resolution, BPC waveforms enabled relative-delay and Doppler localization, and chirp waveforms achieved meter-scale delay-Doppler characterization with resolutions down to 1.87 m. The results demonstrate that distributed bistatic radar architectures combining DSN transmitters with large-aperture receiving assets can provide independent observables for cislunar spacecraft tracking, navigation support, and target characterization in support of future lunar exploration missions. Unlike previous cislunar radar demonstrations focused primarily on feasibility or post-event recovery analysis, this work demonstrates a repeatable multi-waveform measurement sequence suitable for planned lunar mission-support operations.

Accurate navigation and timing on Earth rely primarily on global navigation satellite systems (GNSS), such as GPS, GLONASS, Galileo and BeiDou. However, GNSS is unavailable in deep space, degraded in obstructed environments, and vulnerable to jamming and spoofing. Pulsar-based navigation offers a complementary approach, which uses rotating neutron stars as natural beacons. The time of arrival (TOA) of the pulses from several pulsars with known timing models yields the position, velocity and clock offset of the receiver, conceptually similar to GNSS, without any human-made infrastructure. Furthermore, millisecond pulsars exhibit a timing stability comparable to atomic clocks over timescales of years. In this paper, we revisit radio pulsar navigation, which was historically displaced by X-ray navigation due to the assumed need for large antennas, and present a system-level perspective for terrestrial and space platforms. We first review the history of pulsar navigation, and argue that the radio band offers a larger pulsar population, a mature timing infrastructure, coherent signal processing and ground testability. We then present a receiver architecture, from the antenna to the navigation filter, and relate each subsystem to the terms of the radiometer equation. In particular, we propose a pulsar-gated and whitened array processing approach, which estimates the steering vectors of multiple pulsars simultaneously in the presence of strong radio-frequency interference, and enables both beamforming and attitude estimation using a single array. Finally, we present an example configuration for a vehicle-mounted phased array, and quantify the range error as a function of the aperture and the integration time. We show that an aperture of 8~m$^2$ yields a range error of approximately 1--2~km per pulsar within one hour for the brightest pulsars, which can bound the drift of an inertial navigation system and discipline the on-board clock, whereas a standalone solution requires larger apertures.

The Artemis program and related lunar initiatives are driving a new era of space-based wireless connectivity in which Wi-Fi 6/6E, 3GPP/5G cellular, and Delay/Disruption Tolerant Networking (DTN) architectures will support communications between lunar surface systems, orbiters, and Earth (NASA, 2021; NASA, 2025a). These distributed networks will increasingly rely on artificial intelligence (AI) and machine learning (ML) to manage dynamic environments and defend against threats such as jamming, interference, spoofing, and environmental hazards (NASA, 2021). However, lunar–Earth communication latency and limited human availability challenge situational awareness and timely decision-making, raising critical questions about how humans operators and AI-enabled systems can effectively coexist to ensure secure and resilient operations. This research evaluates human–automation collaboration models, including human-in-the-loop (HITL) and human-on-the-loop (HOTL) frameworks, to assess their ability to preserve situational awareness, accountability, resilience, and performance in lunar and cis-lunar wireless networks. The analysis examines how these frameworks must evolve to address the latency, autonomy, and mission-critical constraints unique to space operations, and explores the influence of organizational culture, training, and governance on AI-enabled security processes. The project further evaluates the role of Explainable AI (XAI) in supporting operator trust and understanding of autonomous behavior, and the application of Zero Trust Architecture (ZTA) for continuous verification across distributed lunar networks. The findings highlight both the promise and limitations of automation in space cybersecurity and underscore the need for balanced human–AI collaboration to support secure, resilient, and accountable lunar wireless operations.

Recent work on cloud-native orchestration in the edge-cloud-space continuum shows that conventional cloud assumptions no longer hold when LEO satellites act as compute nodes. Developments and research on space-based compute nodes further drive technological progress so that characteristics of latest and novel non-terrestrial networks (NTN) will reach a comparable level to terrestrial networks (TNs). We show that NTN should not be treated as a degraded substrate that breaks terrestrial assumptions. Instead, we draft an architecture that enables NTNs delivering a major service extension in converged TN-NTN communication and compute infrastructure. LEO mega-constellations, inter-satellite links, distributed ground stations, and predictable orbital mechanics make NTNs capable extensions of TNs. The pivotal step is to engineer TN and NTN for convergence rather than coexistence, yielding a single TN-NTN converged communication-and-compute continuum in which connectivity and computation are jointly provisioned, abstracted, and orchestrated across the ground-space boundary. By following this approach, we inherently tackle by design the current limitations. Focus is set on leveraging synergies of the technological advantages of converged TN-NTN systems, so that industrial services can be properly hosted and increase overall service capabilities. We have two main contributions. First, we present an architecture for the TN-NTN converged continuum, layered so that communication convergence, workload mobility, and automated orchestration build on one another rather than being bolted onto a terrestrial centered baseline. Second, and going beyond purely conceptual treatments, we define a concrete validation setup and demonstrate the architecture along three case studies of industrial services grounded in Siemens demonstrators and integrated on our COSNO emulation platform, each isolating one capability and the evidence required to prove it: Case Study 1 - Transparent TN-NTN Communication Convergence. A mission-critical nomadic 5G system integrates terrestrial radio, device-to-device sidelink, and LEO satellite backhaul behind a zero-trust overlay, exposing a single transparent channel that hides whether traffic traverses terrestrial or satellite paths. We run one target-agnostic industrial service over the converged channel and collected relevant metrics which were found to be comparable to those of the baseline terrestrial system. Case Study 2 - Movable Workload across the Continuum. Building on transparent connectivity, a safety-critical railway control service (track-side element controller interlocking over RaSTA protocol) is placed and migrated across the TN-NTN continuum with terrestrial and satellite paths in active redundancy. We validate the solution across different dimensions, demonstrating that compute and not only connectivity moves seamlessly across the ground-space boundary. Case Study 3 - Automated Orchestration. Finally, a Function-as-a-Service layer automates placement and migration: a function is submitted with a service-level objective, and the converged continuum decides where and when it runs. We measure deployment latency, placement quality versus an oracle, availability under orbital handover, and the orchestration overhead of automation, demonstrating end-to-end self-management of the converged continuum. Together, the case studies cover communication, computation, and control, showing how a converged ground-space continuum can turn the space tier from a fragile add-on into a peer extension of terrestrial infrastructure. They also show how such a continuum can enable resilient industrial services and unlock new ground-space applications.

Deep space communications between Earth and a spacecraft in the vicinity of a planet are highly affected by the geometry between Sun, Earth, and the planet, particularly when the Sun-Earth-Probe (SEP) and Sun-Probe-Earth (SPE) angles are small. In this paper, we discuss a different way to visualize the Solar System that simplifies the geometric analysis, and we demonstrate the concept using the Mars scenario. The traditional way is that when viewed from above the Earth’s North Pole, Earth and Mars are revolving around the Sun in a counterclockwise direction with Earth at 1AU from the Sun and Mars at 1.5 AU from the Sun. But if we fix the Sun-Earth line, one can visualize Mars to be revolving around the Sun-Earth line in a clockwise direction with a period of 780.25 Earth-day. Using this simplified view, we derive some simple yet useful geometric relations between Sun, Earth, and Mars: a) duration of Mars in Earth’s daytime, b) relationship between SEP and SPE, and c) solar conjunction duration as a function of SEP angle requirement. For human Mars missions, it is highly desirable to ensure continuous communications between Earth and Mars. For short-stay mission, an operation workaround is to conduct the mission outside the solar conjunction duration. But for long-stay mission, it is unavoidable. We consider two deep space relay architectures that mitigate the Mars solar conjunction problems: 1. Earth-Sun heliocentric relay – relay orbiter(s) in Earth-leading and/or Earth-trailing orbits to ensure sufficient angular separation between relay and Sun 2. Mars-Sun heliocentric relay – relay orbiter(s) in Mars-leading and/or Mars-trailing orbits to ensure sufficient angular separation between relay and Sun Using simple geometry we derive the optimal placement(s) of relay orbiter(s) as a function of SEP and SPE requirements and estimate the Mars-relay and relay-Earth ranges. Note that solar conjunction duration is expected to be a small fraction of the Earth-Mars synodic period, so a dedicated relay architecture might not be cost-justified. We discuss an evolvable Earth-Sun relay concept that mitigates Mars solar conjunction and at the same time provides a platform for heliocentric science observations.

Atmospheric reentry at hypersonic speeds causes a "communications blackout," a phenomenon caused by intense bow shock heating that ionizes the area surrounding the vehicle. This creates a plasma sheath, which acts as a high-pass filter that blocks frequencies up to 100 GHz, impairing most conventional communications methods. Terahertz communications (THz, 100 GHz to 10 THz), being well above the cutoff, have the potential to penetrate the plasma sheath and provide a vital communications link. In prior work, we evaluated the theory with real reentry flight data and showed that sub-THz (100-300 GHz) communications can penetrate the sheath while providing high data rates (2 to 8 Gbps), servicing both ground-to-vehicle and vehicle-to-vehicle connections. Ground links, however, are limited in contact times, node availability along long trajectories or remote locations (i.e. open ocean), and pass through the densest portion of the sheath. Given these challenges, we consider the missing third link: satellites. Low Earth orbit (LEO) satellite constellations can serve as an alternative connection to hypersonic systems, offering global coverage, providing access in almost all environments, multiple satellite contacts at any moment, and continuous contact across long trajectories. Communication paths from space bypass most atmospheric absorption, concentrated below 100 km, and travel through the least dense portion of the sheath: the wake. In this paper, we extend our prior end-to-end framework to the satellite domain by building a full-body computational fluid dynamics (CFD) model validated by Orbital Reentry Experiment (OREX) vehicle reentry data and propagating the Starlink satellites over a real reentry case: the Artemis II splashdown. Combining the CFD-derived plasma sheath models, satellite viewing geometries, atmospheric absorption models, and state-of-the-art sub-THz radio parameters, we compare the link capacity of sub-THz (D-Band and 220 GHz) with current communications methods (X-band and Ka-band) in both satellite and ground connection scenarios. We observe that lower-frequency communications pass through the plasma sheath wake in the satellite case but are limited in capacity and unusable for ground links. Across domains, while ground links at sub-THz achieve the highest capacity of >30 Gbps at 220 GHz, satellite links are both highly capable (between 13 and 29 Gbps at 220 GHz), more robust to increased plasma density, and offer a higher aggregate data transfer, with a potential for over 20 satellite links available at each altitude.

NASA’s planned expansion from individual spacecraft operations to sustained lunar infrastructure requires a corresponding shift in space network management. Emerging lunar systems, including nuclear power assets, crewed surface bases, robotic systems, relay satellites, commercial service providers, and science payloads, will form complex, security-critical systems of systems that must be managed, updated, monitored, and protected across intermittent, high-latency links. The SCaN Program Operations Cloud (SPOC) already reflects this transition on the ground by using cloud-native infrastructure to unify scheduling, data transport, and network management across government and commercial providers. This paper proposes extending that cloud-native model beyond the ground segment by using Delay Tolerant Networking (DTN) to support Kubernetes-based orchestration over space links. We present an architecture in which spacecraft, lunar surface systems, and relay assets are treated not merely as remote endpoints, but as managed participants in a federated space network. Kubernetes provides a standard mechanism for deploying services, applying configuration, enforcing policy, supporting self-healing behavior, and managing software lifecycles, while DTN provides the disruption-tolerant transport needed for space environments. A central design principle is preservation of Kubernetes as an industry standard: rather than modifying Kubernetes itself, we use a TCP tunnel over DTN so that Kubernetes components communicate through ordinary TCP interfaces while DTN absorbs delay, disruption, and link variability underneath. This avoids de-standardizing Kubernetes and preserves compatibility with existing tools, implementations, and operational practices. We describe ongoing experimental work over NASA’s Laser Communications Relay Demonstration (LCRD), which provides an operational optical communications testbed with approximately four seconds of round-trip time. Baseline measurements using TCP iperf and HTTP file transfer characterize conventional transport behavior over the LCRD path, while Kubernetes-over-DTN experiments evaluate the feasibility of cloud-native orchestration across the same environment. These tests demonstrate a practical path for extending enterprise orchestration services over space links using High-rate DTN (HDTN), including support for future disconnected deployment approaches such as bundle-based application packaging. The resulting architecture is directly applicable to LunaNet, lunar surface infrastructure, and future Moon-to-Mars operations. By combining Kubernetes, DTN, and SPOC-style cloud operations, NASA can move toward a Solar System Internet in which space assets are secure, manageable, autonomous, and interoperable network constituents rather than isolated mission endpoints.

The rapid expansion of commercial space activities, lunar exploration, and distributed autonomous missions is creating demand for a resilient, high-capacity networking architecture that extends far beyond today's Earth-centric communications infrastructure. Chascii Inc. is developing an integrated optical networking platform designed to provide high-speed communications, precision navigation, positioning, distributed sensing, and edge computing services across Low Earth Orbit (LEO), cislunar space, and eventually deep space. The envisioned architecture consists of interconnected optical nodes deployed in LEO, Geosynchronous Earth Orbit (GEO), and the Earth-Moon L1 and L2 regions, forming a scalable Space Internet capable of supporting scientific, commercial, civil, and national security missions. Beyond communications, the network is intended to enable distributed orbital data centers that bring computation closer to space assets, reducing latency and increasing operational autonomy. At the core of this architecture is SpheroSat, a new class of Autonomous Optical Infrastructure Node (AOIN) that integrates omnidirectional optical communications, autonomous navigation, precision positioning, distributed sensing, artificial intelligence, and edge computing into a compact spacecraft platform. Unlike conventional communications satellites designed primarily as bent-pipe relays, SpheroSat functions as an intelligent network node capable of dynamically routing optical traffic, performing onboard AI inference, supporting autonomous swarm operations, and participating in distributed computing tasks. Its spherical optical architecture provides near-omnidirectional optical coverage through multiple laser terminals, enabling robust crosslinks, rapid network reconfiguration, and resilient operation within large constellations and cooperative spacecraft swarms. This paper presents the SpheroSat system architecture, operational concept, and technology roadmap developed during a nine-month National Science Foundation (NSF) Small Business Innovation Research (SBIR) Phase I effort. We describe the spacecraft's optical communications architecture, AI-enabled networking capabilities, distributed timing and positioning concepts, and its role as a foundational building block for future space infrastructure. We also discuss a phased deployment strategy beginning with LEO demonstrations and expanding to cislunar and deep-space operations. By combining optical communications, autonomous networking, distributed sensing, and edge computing within a single modular platform, SpheroSat represents a new paradigm for space infrastructure and provides a pathway toward a scalable, resilient, and intelligent interplanetary optical network and a network of solar system–wide distributed data centers that can serve as the backbone of future commercial and exploration missions.

We develop an algebraic framework for cross-domain temporospatial network routing using local algebraic network charts. A local algebraic network chart consists of a local graph view, its affine picture variety, an oriented finite \'{e}tale refinement, and routing sheaf data. The locality reflects the fact that in realistic networks, no node or domain has access to a single globally correct topology, clock, or routing sheaf. Each observer makes forwarding decisions from local routing data, and the global problem is to determine when these local data glue to coherent temporal behavior. The framework is also presented as a formal design methodology. Algebraic ideals specify local constraints; schemes organize the corresponding design spaces; \'{e}tale refinements expose routing-relevant alternatives; routing sheaves organize local decision data; and descent expresses compatibility of separately designed domains. In this sense algebraic geometry is not merely a verification tool, but an enabling design language for multidomain temporal routing architectures. Graph varieties provide algebraic models of network incidence, but their Zariski topology is too coarse to distinguish routing choices. We therefore use oriented \'{e}tale refinements as routing-sensitive geometric substrates on which routing sheaves may live. The \'{e}tale path sheaf is our first example, but the framework anticipates a broader collection of routing sheaves, including schedule-valued, cost-valued, capacity-valued, policy-valued, and certificate-valued sheaves. We emphasize that sheaves are passive structures: they organize local data and compatibility conditions. Routing is active: algorithms such as graph search, Dijkstra, Bellman-Ford, contact-graph methods, or delay-tolerant link-state procedures select sections, partial sections, or forwarding actions. We model temporal behavior as finite constructible diagrams of local algebraic network charts. Temporal transitions are represented by directed temporal contacts and persistence data rather than graph morphisms; custody is the same-node special case of a temporal contact. We implement this framework in Macaulay2 and demonstrate temporal journeys that exist even when no instantaneous graph contains an end-to-end path. We also show that static union graphs can fail in both directions: they may give false positives by ignoring temporal order and custody, and false negatives by omitting scheduled temporal contacts not represented as static edges. A larger scaling example shows that many path-level temporal sections collapse to a small number of inter-domain route classes, motivating pushforward to interface contracts and descent. These finite diagrams serve as computable fragments of a larger stack-theoretic program in which moduli describes possible local network states and descent describes compatibility of exposed routing behavior.

We construct picture varieties for arbitrary hypergraphs, extending Martin's picture varieties from ordinary graphs to incidence structures with edges of any cardinality. The motivation is that many systems, especially communication networks, are not naturally described only by pairwise links. A single communication event may involve several nodes at once, as in multicast, broadcast, anycast, shared-medium access, relay groups, scheduled contact opportunities, or domain-level constraints. Hypergraphs provide a natural language for these higher-order interactions. Given a hypergraph, we assign each vertex to a projective space whose dimension is determined by the largest hyperedge incident to that vertex, and we assign each hyperedge to a hyperplane in the corresponding dual projective space. Incidence between vertices and hyperedges is imposed by bilinear point-hyperplane equations. A leading-coordinates convention allows hyperedges of different dimensions to coexist in one construction: lower-dimensional edges read only the coordinates they need, and no missing intermediate faces are added. Thus the construction applies both to abstract simplicial complexes and to hypergraphs that are not closed under taking subfaces. We define a generic locus on which every hyperedge determines a unique hyperplane, and we take its closure to obtain the hypergraph picture variety. We prove that this generic locus is smooth and irreducible, compute its dimension, and show that its closure is an irreducible component of the full picture space. We illustrate the construction with five examples, including graphs, simplicial complexes, non-simplicial hypergraphs, and dimension-skipping hypergraphs. The examples show how projective saturation introduces additional compatibility relations and how hypergraphs with the same generic dimension can have different boundary geometry, degrees, and component structures. Complete Macaulay2 code is provided. This framework is intended as a geometric foundation for later temporospatial network models. In particular, hypergraph picture varieties provide incidence spaces over which one can add time, orientation, routing, and sheaf-theoretic data. This is expected to support natural models of anycast, multicast, and broadcast behavior in delay tolerant and intermittently connected networks.

Large-scale space-air-ground integrated networks (SAGIN), including satellite constellations and heterogeneous relay systems, evolve over time and require routing methods that scale efficiently while adapting to changing network topology. We present a geometric routing framework for temporal directed networks based on low-distortion embeddings into curvature-adaptive product manifolds. Given a time-indexed sequence of weighted directed graphs, we compute the discrete Ricci curvature on each network snapshot to characterize its local geometry. These curvature statistics are used to select a global mixed-curvature latent space comprising Euclidean, hyperbolic, and spherical factors, whose dimensions, weights, and curvature parameters adapt over time to the structural profile of the network. Node embeddings are then learned by minimizing a distortion objective augmented with a curvature-based regularization term, encouraging latent distances to preserve graph distances while aligning the embedding geometry with the graph’s discrete curvature structure. Routing is performed on the original network by a greedy policy that iteratively selects feasible outgoing neighbors minimizing the geodesic distance to the destination subject to operational constraints such as link availability, capacity, and degree limits. To exploit temporal coherence, embeddings are warm-started across successive snapshots and updated via incremental Riemannian optimization rather than recomputed from scratch. Extensive experiments on a variety of dynamic network scenarios show that the proposed method achieves competitive routing performance while substantially reducing solve time relative to non-temporal baselines and graph-based algorithms. These results demonstrate that curvature-adaptive product manifolds provide a mathematically principled and computationally efficient foundation for scalable routing on evolving aerospace communication networks. We conclude with a discussion of future work and potential research directions.

This paper presents NornTrace, a deterministic framework for evaluating and auditing routing decisions in cislunar Delay-Tolerant Networks (DTNs). The central problem addressed by NornTrace is not only whether a route succeeds, but whether the reasoning that produced the route can be reproduced, inspected, and compared after execution. Cislunar DTNs must operate across intermittent connectivity, scheduled contact opportunities, long propagation delays, constrained links, and disruption-prone environments. Existing routing evaluations commonly emphasize delivery ratio, latency, hop count, or forwarding efficiency. Although these metrics are essential, they do not preserve the intermediate decision process that determines why a candidate relay was selected, rejected, or deprioritized. As a result, policy differences may be difficult to distinguish from uncontrolled simulator behavior, and routing decisions may remain opaque to researchers and mission operators. NornTrace treats determinism, decision provenance, and auditability as first-class experimental requirements. The framework combines seed-controlled scenario execution, deterministic candidate evaluation, policy comparison, structured artifact serialization, and render-only trace replay. An explicit artifact boundary separates backend routing logic from the visualization layer. For each routing decision, the backend records the scenario context, available candidates, nominal routing observations, risk indicators, candidate classifications, selected next hop, policy divergence, and replay metadata. The frontend consumes this artifact without recomputing classifications or routing decisions, ensuring that the visual explanation remains consistent with the backend execution. This architecture allows identical routing and network conditions to be replayed and independently inspected, while also enabling different routing policies and contact-plan models to be evaluated through a common experimental workflow. NornTrace is therefore positioned as evaluation infrastructure rather than as a new production routing protocol. A prototype implementation demonstrates the framework using security-aware routing as a representative case study across four controlled scenarios: clean, degraded, jammed, and mixed-risk. All four scenarios generate replayable decision artifacts. Next-hop divergence occurs in one of the four scenarios, while unsafe-candidate filtering occurs in two, showing that a security-aware policy can alter the candidate set without necessarily changing the final route. Under fixed scenario inputs and random seeds, repeated executions produce stable candidate classifications, next-hop selections, divergence states, and replay instructions. In the jammed scenario, the artifact records why the legacy baseline selects a risky relay and why the security-aware policy selects an alternative candidate, allowing the decision process to be reconstructed after execution. The full paper will extend this evaluation with quantitative measurements of delivery ratio, end-to-end latency, path-risk exposure, and scaling behavior across larger contact graphs and contact-plan-driven scenarios. The framework is also designed to accommodate CGR-compatible routing baselines. These capabilities position NornTrace as a reproducible systems platform for studying resilient, auditable, and security-aware routing in future cislunar communication networks.

The LunaNet architecture, currently being specified by space agencies and starting to be implemented by industry, is to provide communications and navigation infrastructure to enable Artemis and Moon Base operations. This paper is focused on the networking protocols and service interfaces aspects of LunaNet, the objective being to (i) provide an overview of the main aspects of LunaNet architecture, data services, and protocols, (ii) expose immediate development challenges and fundamental research problems on the horizon that need to be addressed, (iii) overview projected infrastructure and data service needs of LunaNet, (iv) present key Artemis and Moon Base LunaNet use cases, and (v) discuss solution approaches to the development challenges and research problems. The paper begins by providing an overview of the current LunaNet Interoperability Specification (LNIS) and LunaNet Data Services Document (LDSD) which define LunaNet as a federation of Luna Network service providers (LNSP) that interoperate according to architecture, protocol, and interface specifications. These specify the networking among lunar and terrestrial LunaNet end-users at (i) the data link layer using frame services based on the CCSDS AOS, USLP, SLE, and CSTS protocols and (ii) the network layer using IP (i.e., IP Modes 0, 1, and 2, as defined in LDSD) or DTN bundle protocol (BP). The overview points out immediate development challenges and imminent research problems. These include the need for developing methods of associating LNSP interfaces with reachable destinations and updating this information over time as network topology changes occur, developing IP Mode 1 based on LNSP tunneling functionality, extending existing CCSDS SLE and CSTS protocols to work with LNSP proximity links, and research on routing protocols needed in IP Mode 2. The overview also points out more fundamental intrinsic problems that need to be dealt with in realizing end-to-end LunaNet frame, IP, and DTN BP data services - central problems being the time-varying LunaNet network topology and time-varying link latencies, caused by the relative orbits and rotations of the Moon, Earth, lunar-orbiting relays and lunar-orbiting users. The paper then provides a summary of the projected number of users, data service demands, and LunaNet infrastructure needs over upcoming Artemis missions and Moon Base phases of development. Key use cases are presented involving the Human Landing System (HLS), Lunar Terrain Vehicle (LTV), lunar data aggregator, cross-hauling of data between Moon bases via a lunar orbiting relay and transmitting Moon Base observations to a terrestrial Mission Operations Center (MOC). Following this, the paper presents possible solution approaches to the identified development challenges and research problems Finally, although transport layer protocols are not included in LunaNet specifications since they operate ‘over-the-top' of LunaNet, transport protocol mechanisms and parameters need to be optimized for LunaNet to meet the needs of LunaNet user applications that depend on transport protocols such as TCP and QUIC. The paper concludes with a discussion of how these transport protocols, and associated user applications, can be configured and optimized for performance in the face of large time-varying end-to-end bandwidth-delay products of links.

Global Navigation Satellite Systems (GNSS) are essential to next-generation space systems, supporting positioning, navigation, timing, synchronization, and mission coordination across terrestrial and non-terrestrial infrastructures. However, civilian GNSS signals are vulnerable to spoofing attacks, in which counterfeit GNSS-like signals can mislead receiver acquisition and tracking processes and corrupt the navigation solution. This vulnerability is especially relevant for future space architectures that increasingly rely on resilient, autonomous, and interconnected navigation services. This work proposes PULSAR-GNSS, a Physical-Layer Unknown-Signal Learning for Spoofing Authentication and Rejection in GNSS framework. The method is based on radio-frequency fingerprinting, which exploits subtle hardware- and channel-dependent impairments embedded in the received I/Q samples. These impairments, introduced by oscillators, mixers, amplifiers, antennas, propagation effects, and the transmission chain, can form a distinctive digital signature of the received signal source. PULSAR-GNSS learns this signature in a compact representation space and uses open-set recognition to reject observations that do not match the learned authentic signal behavior. The proposed framework uses a residual neural-network encoder to transform receiver-side I/Q observations into a low-dimensional fingerprint representation. Spoofing detection is then performed in this learned representation space through a distance-based rejection rule. Experimental results show that PULSAR-GNSS achieves a mean AUC of 94.55%, a mean accuracy of 92.87%, a mean macro-F1 of 92.83%, and a spoofing Detection Rate of 99.53%. These results are competitive with recent GNSS spoofing-detection studies and highlight the potential of open-set RF fingerprinting for resilient receiver-side spoofing detection under a different authentic-versus-spoofing evaluation protocol.

This work addresses the resilience of mission- and industry-critical services in dynamic Satellite Computing Networks by introducing a cross-layer service-level awareness architecture for converged terrestrial and non-terrestrial (TN-NTN) infrastructures, enabling predictive adaptation to ensure industrial service continuity. Resilience for industrial services over converged TN-NTN is widely framed as a question of mechanisms, e.g., incident detection, redundant paths, failover, store-and-forward. Existing resilience design frameworks only consider very limited inherently dynamic systems like edge computing, industrial control and state-of-the art NTN systems. Thus, resilient critical services in dynamic TN-NTN require system-state awareness to apply principles such as graceful degradation and dependability across subsystems. This paper addresses that gap: the awareness problem. A system cannot fail over intelligently via a controlled transition path until it knows, at the service level, that its requirements are at risk, which action would restore them, and whether applying that action will badly influence existing services. In highly dynamic, only partially observable Satellite Computing Networks (SCNs), creating this service-level awareness and acting on it without destabilizing the live system is the central unsolved challenge. We address it with a resilience-awareness architecture that operationalizes the situational-awareness and validation constituents of the resilience engineering framework for non-terrestrial settings. It organizes resilience around a closed awareness-to-action-to-validation loop and answers three questions that the failover-centric literature leaves open: (1) How is awareness created at the service level? Applications declare requirements (availability, latency budget, control-loop stability) and each subsystem translates these top-down into its own capabilities and reports bottom-up against them. Thus, end-to-end service health and not just link health is continuously assessed. (2) Where and what to be measured? We define measurement points and resilience-awareness parameters across the SCN (e.g., orbit ephemeris, link quality, topology, contact windows), develop scalable information gathering procedures, and map them to service-level indicators. (3) What insights can be derived from system observations for resilient operation of critical services? We assess observed system information against hosted services and their service-level objectives (SLO), identifying NDT characteristics, predicted SLO margins, contact-window countdowns, behavior-pattern anomalies, and what-if impact estimates that translate awareness into justified ahead-of-time decision options. This work addresses the aforementioned questions by defining: (i) a resilience framework that uses awareness creation as its organizing principle, formalizing the requirement-to-capability translation and the SCN-state-to-service-awareness mapping; (ii) interfaces that transparently derive and apply resilience actions from awareness indicators, so services and applications benefit and can adapt; and (iii) an evaluation of service-level awareness for a representative industrial satellite service (e.g., industrial remote-control over NTN) via a Network Digital Twin, implemented on our satellite constellation emulation platform. The Network Digital Twin synchronizes with a real-world LEO constellation and link parameters, analyzes service-level awareness, predicts degradation, and assesses derived reconfigurations.” Finally, we discuss the main findings, not only in terms of what works but also where the awareness problem remains open: the limits of predictability under partial observability, the latency of awareness versus the speed of degradation, and the derived actions.

While terrestrial 5G systems provide flexible core-network programmability and service-aware traffic handling, satellite backhaul introduces propagation delay, jitter, bandwidth constraints, and link variability that may significantly affect delay-sensitive aerospace-based IoT applications. To study these effects in a controlled and reproducible manner, this work develops an experimental 5G–satellite emulation framework that integrates OpenAirInterface 5G core/RAN functions with OpenSAND-based satellite link emulation and Linux NetEm impairment injection. The proposed testbed focuses on the user-plane behavior of 5G traffic transported through an emulated non-terrestrial backhaul. The implementation validates the complete end-to-end data path, including UE attachment, PDU session establishment, UPF-based forwarding, satellite-path insertion, and controlled impairment injection. Rather than treating the satellite link as a generic delay element, the framework enables systematic evaluation of how satellite-induced latency, jitter, bandwidth limitation, and packet impairment propagate through the 5G user plane and affect application-level quality of service. The current implementation already demonstrates that UE-generated traffic can be forced through the emulated satellite path and monitored under configurable link conditions. The main research direction proposed in our paper is to transform the testbed from a connectivity validation platform into an application-aware QoS evaluation framework. The study defines representative T traffic classes, including low-rate telemetry, delay-sensitive command-and-control traffic, and high-volume payload transfer involved in monitoring applications, such as multimedia content, logs, or UAV/CubeSat data processing. Each traffic class has distinct service requirements for latency tolerance, jitter sensitivity, throughput demand, reliability, and deadline compliance. By evaluating these traffic classes over the same controlled 5G–satellite user-plane path, our paper aims to quantify how different mission workloads degrade under satellite-backhaul constraints and identify which services require prioritization, edge buffering, compression, traffic shaping, or delay-tolerant operation. The expected contribution is not merely the integration of OpenAirInterface and OpenSAND, but the development of a reproducible experimental methodology for mapping satellite-link impairments to mission-level QoS behavior in 5G-enabled aerospace IoT systems. This enables answering a practical research question: under what satellite-backhaul conditions can 5G-supported mission-critical IoT services remain operational, and when are additional adaptation mechanisms required? The resulting framework can support future studies on service-aware scheduling, edge-assisted processing, QoS-aware UPF routing, and resilient communication strategies for UAV, CubeSat, and remote sensing applications. The proposed testbed will be validated by simulating in-lab realistic emergency scenarios (e.g., wildfires), where effective and real-time monitoring of the crisis environment imposes stringent QoS requirements.

Non-Terrestrial Networks (NTN) Narrowband IoT (NB- IoT), with its ubiquitous coverage, presents an efficient and budget-friendly solution for low-rate communication systems. Potential performance instabilities induced by environmental and network architecture parameters of the NTN may raise the question: “how to ensure correct and timely data transmission?”. Formal methods enable more advanced probabilistic reasoning for diagnostics, making them a suitable approach for assessing communication reliability under constrained NTN conditions. In this work, we investigate real-time error detection of the communication system through runtime verification (RV) by designing specifications to detect communications system faults and enable subsequent mitigation. The RV solution candidate must satisfy the conditions of low overhead and onboard ease of integration into the NB-IoT. Therefore, we select the Realizable, Reconfigurable, Unobtrusive Unit (R2U2) that not only complies with these conditions but also has proven its practicality as an engineering solution in embedded systems. This study implements the R2U2 framework onboard a UE emulating a drone telemetry system with an NTN NB-IoT modem. Operating over \gls{geo}-based satellite communication links, the proposed architecture dynamically monitors low-data-rate safety-critical drone telemetry by evaluating temporal packet-level specifications against runtime constraints such as network data congestion. The experimental evaluation demonstrates that the R2U2 framework can be integrated with limited memory overhead and can detect communication degradation and congestion-related violations during operation.

Modern space missions increasingly operate as highly interconnected systems-of-systems composed of distributed spacecraft, ground systems, commercial providers and international partners, autonomous capabilities, and complex operational environments. While existing cybersecurity frameworks and Systems Security Engineering (SSE) provide essential guidance for engineering secure systems, they primarily evaluate whether prescribed security capabilities have been implemented. They do not fully address whether those capabilities collectively enable mission success within the context of mission objectives, operational concepts, trust relationships, and mission architectures. This paper proposes mission-centric security engineering as a new engineering paradigm that evaluates security according to its contribution to mission success rather than solely through the implementation of individual security controls (compliance-based security assessments). Building upon this paradigm, the paper introduces Mission Security Engineering (MSE) as a systems engineering discipline that extends established Systems Security Engineering by evaluating how engineered security capabilities collectively influence outcomes and success across the mission. The paper establishes the engineering foundation for Mission Security Engineering by defining its central thesis, engineering principles, distinguishing characteristics, relationship to existing systems engineering disciplines, and engineering focus areas. Rather than replacing existing cybersecurity frameworks, safety engineering, reliability engineering, mission assurance, or Systems Security Engineering, Mission Security Engineering integrates these complementary disciplines within a mission-level engineering perspective that evaluates security as an emergent property of mission architecture, operational behavior, trust relationships, and supporting capabilities. As mission architectures continue to evolve toward increasingly autonomous, software-defined, and interconnected operational systems-of-systems, engineering methodologies must evolve accordingly. Mission Security Engineering provides a conceptual foundation for evaluating mission security beyond individual systems and establishes a framework for future research in mission-centric security assessment, mission resilience, engineering analysis, and operational assurance.

Regenerative satellite payloads, which host the 5G Next Generation Node B (gNB) onboard instead of transparently relaying the radio signal, are a key building block of the Non-Terrestrial Network (NTN) integration that 3GPP started in Release 17. Combining them with the Open Radio Access Network (O-RAN) architecture adds software-defined control through RAN Intelligent Controllers (RICs), enabling in-orbit reconfiguration, multi-tenancy, and data-driven radio resource management. Open-source implementations make this vision accessible to research institutions and New Space missions, yet there is little quantitative evidence on how they behave on the embedded processors that a satellite can actually carry. This paper presents an out-of-the-box experimental benchmark of two open-source gNB stacks (i.e., OpenAirInterface (OAI)/Duranta and srsRAN/OCUDU), and two Near-Real-Time RICs (i.e., the ORAN-SC Near-RT RIC and FlexRIC). All components are deployed on an ARMv8 platform representative of the processing units used in small-satellite payloads, driving a USRP B210 software-defined radio. We compare them in terms of functional compliance (3GPP NTN features, O-RAN interfaces, multi-tenancy), resource efficiency, real-time behavior, and engineering suitability. Results show that these platforms can host a joint gNB + near-RT RIC deployment using any of the considered components, although further optimizations are still required for carrier-grade performance. However, the performance and standards compliance vary significantly among them. Whereas srsRAN has a smaller memory footprint, OAI shows a lower CPU utilization. And while both have significant NTN feature support, they still differ in some relevant functionalities. On the controller side, FlexRIC requires far fewer resources than the ORAN-SC RIC, while the latter offers broader O-RAN compliance. Building on these results, we derive a decision framework that maps mission profiles and Size, Weight, and Power (SWaP) budgets to gNB/near-RT RIC combinations. Finally, we discuss the open gaps remaining before these stacks can be flown in NTN regenerative payloads.

This paper presents a comprehensive bibliographical review on the evolution, miniaturization, and development panorama of Interferometric Fiber Optic Gyroscopes (IFOGs), highlighting the global research landscape and Brazil's strategic positioning. Based on the Sagnac Effect, IFOGs have established themselves as indispensable sensors for high-performance inertial navigation systems due to their solid-state architecture, high reliability, and vibration resistance. Currently, the field faces a paradigm shift focused on reducing size, weight, power, and cost (SWaP-C), moving from isolated sensitivity improvements to highly integrated, compact chip-scale solutions. Analysis of international literature reveals distinct strategic approaches among key technological hubs. China leads in photonic integration and the use of novel materials such as Thin-Film Lithium Niobate (TFLN) , whereas the United States stands at the forefront of chip-integrated waveguide coils aimed at eliminating conventional fiber coils. France excels in industrial maturity, delivering robust commercial Inertial Measurement Units (IMUs). Additionally, the Euro-Asiatic bloc focuses on stress modeling under severe environmental conditions , and India emerges in the design of internal components on new material platforms. To overcome thermal stability limits and geometric constraints, the global frontier adopts innovations ranging from air-core antiresonant fibers to three-axis architectures utilizing time-division multiplexing (TDM). In the Brazilian scenario, continuous interinstitutional research among ITA, UNICAMP, and IEAv has built sovereign expertise in digital signal processing techniques and closed-loop control architectures to mitigate stochastic noise. Facing an industrial bottleneck due to the absence of local integrated photonics foundries, the national ecosystem has directed recent efforts toward applied systems engineering, implementing robust control strategies (such as sliding-mode observers) and qualifying tactical-grade prototypes for space environments and launch vehicles. It is concluded that IFOG miniaturization has surpassed the conceptual stage, and mastery over control loops ensures technical autonomy and innovation absorption capabilities essential for consolidating the sovereignty of Brazil's aerospace program.

Navigation in cislunar space relying only on GNSS measurements results in degraded performance due to extremely poor geometric dilution of precision (GDOP) with all measurements coming from Earth. Collaborative Positioning, Navigation, and Timing (CPNT) enables a network of nodes to share measurements to improve the navigational state of every asset via cross-link. Sharing measurements via cross-link improves the observation geometry of the constellation, as well as providing addition measurement types to be processed. This enables improved estimation of position, velocity, time, and biases for each individual node in the constellation. The "DEcentralized Collaborative Estimator of Navigation and Time" (DECENT) application, is a Kalman-filter based software application for enabling autonomous spacecraft navigation in a multi-vehicle constellation which implements this CPNT architecture. It is comprised of the cPNT software library, a database for managing measurement and restart information history, and a messaging interface to support communication with other nodes. cPNT is a generalized library that contains all the functions required to perform collaborative PNT. The DECENT software stack was put through a series of lab demonstrations, where simulated GNSS and cross-link measurements were fed to a collection of single-board-computers (SBCs) representing the nodes of a satellite constellation. The SBCs were networked together on a local area network (LAN) and communicated with each other over the network using JPL's ION DTN reference implementation of delay tolerant networking to mimic the limited availability for communication. In this paper, we present the design and implementation of CPNT using the DECENT application for a system of 5 satellites in orbit at the moon. We also describe the technical setup of the lab demonstration and present both performance and timing results.

A surface station near the lunar south pole can work alongside the planned LunaNet infrastructure to significantly improve the navigation performance of lunar surface users. This surface station-enhanced (SSE) position, navigation, and timing (PNT) service provides differential corrections, Joint Doppler and Ranging corrections, and an additional pseudorange observable to nearby users to supplement a lunar relay network’s PNT services during the initial phases of the constellation’s deployment. A lunar station providing these services is essential during early missions to the south pole that require real-time, highly accurate PNT (<10 m) but only have access to 2 or 3 navigation orbiters. We propose the LITEHOUSE mission to demonstrate a surface station-enhanced PNT architecture on the Moon from a single payload on a commercial lunar payload service (CLPS) lander. LITEHOUSE (Lunar In siTu EnHancement Of User navigation with a SurfacE station) is a compact technology demonstration that utilizes a UST-Lite radio configured to act as both the “station” and a “user” receiver in a single package with antennas separated by ≥2 m on the lander. Both receivers acquire LunaNet’s Augmented Forward Signal (AFS) from a constellation of likely only 3 total orbiters. The station receiver utilizes its known ground truth position to estimate differential and JDR corrections and transfers the corrections to the user receiver, which deploys two real-time navigation filters: one filter utilizing SSE services and the other using traditional range and Doppler-based positioning with only AFS measurements. Although typical baselines for a SSE PNT service would be on the order of kilometers, this mission validates the full SSE chain entirely on one payload: station measurement reception, correction generation, transfer to user, and improved real-time user navigation. Navigation simulations estimate that LITEHOUSE can achieve around 1 m positioning accuracy at 3σ with only a 3-satellite LunaNet constellation, an order of magnitude improvement on traditional positioning with LunaNet’s AFS (16 m @3σ). When implemented, LITEHOUSE would be the first asset on the moon that self-positions to a real-time accuracy of <10 m @3σ using a sparse LunaNet constellation of three orbiters. LITEHOUSE directly advances NASA’s Moon to Mars objectives by proving SSE as a scalable architecture for lunar PNT: SSE and JDR corrections can serve unlimited users and can be relayed beyond line-of-sight. Even with a large navigation constellation capable of broad PNT coverage, an SSE station still provides long-term benefits by performing in-situ orbit determination, reducing DSN dependency, and enabling lunar coordinated time. LITEHOUSE also prepares SSE for an independent Martian PNT architecture, which will face the same constraints. In summary, LITEHOUSE is a low-cost, high-impact mission that demonstrates the performance gains of an SSE PNT service, increments its technology readiness level, and prepares the SSE PNT concept for implementation as a durable, scalable PNT architecture extending from the Moon to Mars.

Artificial Potential Field (APF) planners are attractive for small planetary rovers because they are lightweight, reactive, and computationally efficient. More recently, Robust Artificial Potential Field (RAPF) has demonstrated significantly improved target reachability compared with conventional potential field methods. However, RAPF remains sensitive to parameter selection and may generate invalid waypoint transitions or repeatedly replan the entire trajectory following local-minimum recovery or the discovery of previously unknown obstacles. This paper introduces Repair-based Robust Artificial Potential Field (R2APF), a novel extension of RAPF designed specifically for reliable and efficient single-rover navigation in partially observed planetary environments. The key contribution of R2APF is a repair-based planning mechanism that preserves collision-free portions of an existing trajectory and selectively replans only the affected segment when newly sensed hazards invalidate the current path. The proposed framework further integrates offline hyperparameter optimization to improve planner robustness and midpoint collision checking to detect invalid waypoint transitions that may occur between discrete planning points. These mechanisms collectively enable R2APF to maintain previously validated path information while adapting locally to newly revealed terrain hazards, thereby reducing redundant computation and unnecessary trajectory changes. This capability is particularly relevant to lunar surface exploration, where rovers encounter craters, rocks, slopes, and other hazards under limited sensing, computational, and operational constraints. In sensing-based lunar-like benchmarks, where obstacles are discovered incrementally during execution, R2APF improves the overall success rate from 97.96% to 99.84% compared with tuned RAPF. It additionally reduces planner calls by approximately 40%, lowers cumulative planning effort by more than 35%, and produces shorter executed paths across all evaluated scenarios. For a comprehensive study, in addition to APF, we also show comparison with other state of the art path planning paradigms such as A*, D*, RRT to highlight the benefit of the proposed R2APF. The results demonstrate that R2APF provides a more reliable, computationally efficient, and adaptive reactive planning framework for autonomous lunar rover navigation in cluttered and partially observed terrain.

Mars surface missions rely on orbiting spacecraft to relay scientific data and telemetry to Earth, with communication volumes directly impacting mission planning, resource allocation, and science return. Accurate prediction of relay communication data volumes is essential for operations teams scheduling rover activities and managing orbital assets. We present Sky Arc, a statistical data volume estimation system developed for NASA's Mars Science Laboratory (Curiosity) and Mars 2020 (Perseverance) missions that leverages years of historical overflight data to predict future communication performance. Sky Arc addresses a critical operational challenge: predicting relay communication volumes without requiring detailed link budget calculations for each overflight. The system employs a two-phase architecture that separates reference profile generation from operational predictions. During profile generation, historical overflight data from three Mars orbiters: Mars Reconnaissance Orbiter (MRO), Trace Gas Orbiter (TGO), and Odyssey (ODY) are processed to create statistical models organized by orbital geometry—specifically azimuth at maximum elevation and peak elevation angle. These profiles capture second-by-second geometric relationships and communication performance based on actual telemetry. The operational estimation engine requires only basic overflight parameters (maximum elevation angle, azimuth of maximum elevation angle, and rise time) to retrieve appropriate statistical profiles and generate volume predictions. This lightweight approach enables standalone deployment, making the system suitable for containerized services, continuous integration workflows, and distributed operational environments. The system supports terrain masking to account for local topography blocking line-of-sight communications and applies time constraints to model partial communication windows. The approach requires sufficient historical overflight data and repeating ground track geometries. Insufficient data for Mars Express and the Molniya orbit of MAVEN limit applicability, making the method most suitable for orbiters with consistent relay geometries and plentiful historical data (MRO, ODY, TGO). Accuracy validation against 10+ years of actual lander volumes shows Sky Arc achieving mean prediction errors within ±10 Mbits of traditional link budget prediction errors for all three orbiters. The standalone architecture, containerized deployment options, and simplified Python API are designed to enable adoption beyond traditional operations centers, with potential applications in mission planning, science team activity scheduling, and relay orbiter coordination. Sky Arc demonstrates that statistical approaches leveraging historical mission data can provide accurate, operationally useful predictions without the complexity of traditional link analysis tools. As Mars exploration expands with additional surface assets and relay orbiters, scalable prediction systems that learn from operational experience will become increasingly important for managing the Mars relay communication network.

The NASA Lunar Endurance rover mission concept was recently identified by the Planetary Science Decadal Survey as the highest priority for the Lunar Discovery and Exploration Program. Scheduled for delivery to the Moon’s South Pole-Aitken (SPA) basin via a Commercial Lunar Payload Services (CLPS) lander no earlier than 2031, Endurance will execute a 2,700-km traverse to collect samples across 20 distinct sites. The rover's ultimate destination is the lunar south pole, where the collected samples will be delivered to Artemis astronauts to return to Earth. Because this traverse occurs primarily on the lunar Farside, preventing direct-to-Earth communication, the mission must rely on NASA’s Lunar Communications Relay and Navigation Systems (LCRNS). This paper evaluates various telecommunications design options between Endurance and the LCRNS. Specifically, we analyze the CCSDS Proximity-1 handshake and Augmented Forward Signal (AFS) links using an S-band Low-Gain Antenna (LGA), alongside a high-throughput Return link utilizing a Ka-band High-Gain Antenna (HGA). To assess these link architectures, we utilize the JPL-developed Telecom Orbit Analysis & Simulation Tool (TOAST). Finally, we present trade studies on the optimal Ka-band antenna size given constraints on RF transmit power, pointing accuracy, and multipath losses. Additionally, we evaluate performance enhancements achieved by augmenting the telecom subsystem with an S-band high-gain antenna for link redundancy, as well as alternative LGA options to optimize the reception of Proximity-1 handshake and AFS waveforms.

This paper describes MAVEN UHF relay operations from conception through end of mission from an historical perspective. The MAVEN spacecraft was not designed for relay beyond accommodating a single-string Electra radio, and yet it became one of the Mars Relay Network’s most effective relay orbiters. The paper discusses mission and relay design, and relay performance with emphasis on constraints and the measures that were taken to overcome them to provision relay service. These include in-flight updates to improve monitoring of electromagnetic interference (EMI), to enhance service coverage through split-pass relay, to improve data throughput by aerobraking and by the implementation of low-density parity-check channel coding, and to mitigate power limitations during eclipse seasons through new relay attitudes.

Today’s Mars Relay Network (MRN) represents a highly successful international collaboration between NASA and ESA. As of this report’s writing, the MRN consists of five heterogenous science orbiters, launched over the past few decades. In the years ahead, the current MRN topology is expected to change significantly. Robotic missions to Mars are expected to increase in number, followed by dramatic increases in the data rates/volumes that will be needed in the era of Mars human exploration. This paper provides recommendations for a next Mars relay orbiter to begin the process of replenishing the existing network, providing relay support to both surface and orbital relay service users in the future. It suggests capabilities that could be implemented in a single future orbiter, rather than representing a set of capabilities that could be supported by a constellation of orbiters. However, the idea of follow-on orbiters was considered. An summary of the recommendation in several areas is below. Choice of Orbit - In the choice of orbit for a Mars relay orbiter, the drive for high data volume return is stacked against user access (access frequency, duration, and gaps) and requirements for providing position, navigation, and timing (PNT) services over the proximity link to relay service users. An altitude between 500 km and 1000 km is recommended for the orbiter with an inclination that best supports the expected user base if there are no requirements for PNT. If supporting PNT capability is a needed, then an orbit with an altitude between 1500 and 2000 km is recommended. Telecommunications Capability - The use of Ultra High Frequency (UHF) over the proximity link is recommended in the next Mars orbiter to maintain backwards compatibility with existing Mars assets. In addition to UHF, the use of K/Ka-band is recommended to provide much more bandwidth compared to UHF (over 130x). It is recommended to use Ka-band for the direct-to-Earth link with the option of either X or Ka-band for the direct-from-Earth link. Position, Navigation, and Timing - An ideal PNT orbit solution is one that provides two-way and one-way range and Doppler tracking with the greatest geometric diversity in the shortest amount of time. For surface localization, this naturally pushes orbit selection to higher altitudes and with multi-satellite constellations that can provide multi-fold simultaneous coverage. In this study, with priority being given to performing data relay with PNT playing a supporting role, a single satellite orbit near 2000 km in a circular, 54˚-degree inclination orbit provides the best overall surface positioning performance across all latitudes and at the lowest altitude. Science Opportunities - a spacecraft placed in either of the recommended low-Mars, circular orbits offers the opportunity to conduct a range of high-priority science investigations that address key knowledge and data gaps in our understanding of Mars, such as atmospheric profiling, winds, trace gas mapping, surface imaging, and space weather. These investigations, as prioritized by the Mars science community, would further help reduce EDL and surface operations risks for both future missions at the Martian surface.

This paper describes MAVEN’s in-flight record of monitoring spacecraft electromagnetic interference (EMI) levels using the bit error rate (BER) loopback and open loop recording (OLR or Sniff) capabilities of the Electra UHF transceiver (EUT). MAVEN established a regular cadence of these in-flight tests starting in cruise in 2014 to recent operations in 2024. The tests were helpful in verifying and maintaining the electromagnetic cleanliness of the spacecraft, which was an important factor in relaying large amounts of data from NASA Mars landers, including the Curiosity and Perseverance rovers. The paper describes the flight record of this testing and provides estimates of the accuracy of the tests. Analysis is included to quantify the impacts of antenna pointing and receiver temperature on BER tests. The paper also introduces a novel technique to radiometrically calibrate Sniff spectra and concludes with a summary of lessons learned from this testing.

The future of navigation on the lunar surface will be driven by the augmented forward signal (AFS) broadcasted by a set of interoperable navigation and communication constellations following the LunaNet interoperability specifications. AFS is a one-way ranging signal, similar to signals implemented by the global navigation satellite system (GNSS), that enables real-time positioning on the lunar surface to a large, scalable number of users. Our team is developing a surface station enhanced (SSE) lunar navigation architecture to improve real-time position, velocity, and timing performance for lunar surface users even with a minimal navigation infrastructure. An SSE payload on the lunar surface provides differential corrections, our novel Joint Doppler and Ranging corrections, and an additional pseudorange signal to users to enable accurate, <10 m (3σ) navigation. To develop a hardware prototype of the SSE payload, we built a low-cost, AFS testbed utilizing software defined radios (SDR) to emulate the entire LunaNet AFS range link in hardware. We designed a configurable AFS transmitter pipeline in GNU radio with inputs of the satellite ranging code, orbit model, and navigation data message. The transmitted signal follows the specification in LunaNet: a short, 2 millisecond code at 1.023 Mchip/s on the I-arm with data modulated at 500 bps, and a long, multi-tiered code at 5.115 Mchip/s on the Q-arm. A custom AFS receiver written in Python (also run in GNU radio) performs initial acquisition and tracking of the signal, decodes the navigation message, and estimates a pseudorange measurement in real-time. This receiver can perform initial acquisition and tracking in under 5 seconds and completes a full code tracking loop update in ~0.5 ms. For our own testing, the receiver can be configured to perform automatic calibration of both the transmitter and receiver clocks. The transmit and receive pipelines were built in software for quick iteration; future work will implement these in the FPGA for improved speed and responsiveness. Ultimately, this low-cost AFS testbed enables hardware emulation of the LunaNet navigation environment on the lunar surface. We will utilize this testbed to generate the signals transmitted by 3 - 5 LunaNet orbiters and received by two separate AFS receivers: one on an SSE payload and the other on a moving user. These hardware-in-the-loop tests will estimate the real-time navigation performance with and without SSE and increase the technology readiness level of an SSE payload.

High bandwidth free space optical communications require high-speed and high-performance optical modulators. This is particularly interesting for lunar surface to surface, surface to orbit communications. Previous work has demonstrated that a novel class of “stepped quantum well” (SQW) modulators can demonstrate higher performance and efficiency than other quantum well designs. These SQW modulators show promise for use in spaceflight to improve the optical communication efficiency, but their performance in deep space temperatures, especially the challenging lunar surface temperature has not been studied. To address this gap, we constructed a cryogenic vacuum chamber to simulate deep space conditions extending down to 25 K. Preliminary validation demonstrates reliable temperature control and stability, and this system can be automated to simulate repeated temperature cycling for lunar surface scenario. This test chamber will enable us to observe how the modulator’s behavior changes at differing temperatures, and what changes to substrate material may be required to expand the device’s active temperature range. The full paper will show experimental SQW performance in the chamber emulating lunar surface environment, and assess their suitability for spaceflight free-space optical communications.

Planetary robots increasingly generate image, stereo, LiDAR, multispectral, and three-dimensional mapping data at rates that exceed practical deep-space return capacity. This paper develops Shared Generative Scene Coding (SGSC), a utility-scalable data-coding architecture in which a synchronized geometric and semantic library is installed at both the planetary element and the Earth terminal before launch. Instead of transmitting a complete sampled scene, the flight system transmits a compact scene program containing model identifiers, geometry parameters, transforms, material classes, uncertainty, provenance, and residual information. The approach shifts the communications problem from transmitting the ‘world’ to transmitting selective ‘windows’ of the world. Three design requirements distinguish the architecture: engineering products decode conservatively so uncertainty cannot be mistaken for clearance; provenance tags structurally distinguish measured, inferred, and interpolated content; and reconstruction residuals are retained as a novelty signal that prioritizes raw measurements for science return. We develop a rate–utility formulation, progressive L0–L6 product hierarchy, and implementation path using analytic primitives, spherical-harmonic shape models, learned residual representations, and CCSDS-compatible packetization. We then perform matched single-image feasibility studies on a Mars scene from the Pathfinder mission and a lunar scene from the Yutu rover (Chang’e 3). Both experiments encode 120 compact hazard candidates using lightweight onboard-style segmentation and serialize them into scene programs. The Mars scene program compresses to 2.10 kB from a 678.6 kB source JPEG; the lunar scene program compresses to 2.00 kB from a 669.8 kB source JPEG. The resulting reductions are 323× and 334× relative to the supplied JPEGs, respectively, while preserving the selected detector output by construction. These are task-level representation savings, not equal-fidelity image-compression ratios. The cross-domain result shows that SGSC-style operational scene abstractions can remain in the same kilobyte-scale regime across substantially different planetary visual conditions. We ask a decisive open question: whether a shared generative library provides additional rate reduction beyond a conventional parametric encoder, and a preliminary held-out 2-D shared-library experiment yielded approximately 13–14% reduction in object-description rate at matched descriptor fidelity. We further evaluate a science-rich selective-abstraction mode that preserves object-complete measured imagelets for major rock targets while representing the remainder of the scene compactly. In a held-out Mars image, 12 mutually exclusive object-complete science regions covering 44.3% of the image produced up to 50.4% payload reduction relative to the source JPEG, while preserving the full visible morphology and local context of the selected rocks.

The proliferation of mega-constellation satellite networks has fundamentally transformed the landscape of global communications, necessitating a paradigm shift from traditional radio frequency (RF) ground stations to advanced optical ground stations (OGS). This paper presents a comprehensive analysis of recent developments in optical ground station technology and infrastructure deployed worldwide to support next generation satellite constellations including Starlink, OneWeb, and Amazon’s Project Kuiper. This research examines the innovative trends driving the adoption of free-space optical (FSO) communication systems, which offer unprecedented data throughput capabilities exceeding 10 Gbps per link compared to conventional RF systems limited to hundreds of Mbps. The analysis encompasses a global survey of 47 operational and planned optical ground stations across six continents, revealing strategic deployment patterns optimized for orbital mechanics, atmospheric conditions, and geopolitical considerations. Key innovative trends identified include the implementation of adaptive optics systems for atmospheric turbulence mitigation, development of multi-aperture ground stations for enhanced link availability, and integration of artificial intelligence algorithms for predictive beam steering and handover optimization. The emergence of commercial optical ground station networks as a service (OGSaaS) represents a significant shift from government-controlled to commercially accessible space communication infrastructure. However, several critical challenges persist in widespread OGS deployment. Atmospheric scintillation and cloud coverage introduce link availability constraints, with measured uptimes ranging from 65% in equatorial regions to 85% in arid climates. The paper quantifies weather-related outage patterns and presents statistical models for link budget calculations under various meteorological conditions. Technical challenges include precise pointing and tracking requirements with sub-microradian accuracy, thermal management of high power laser systems, and standardization of optical communication protocols across diverse satellite platforms. Security considerations present both opportunities and vulnerabilities, with optical links offering inherent beam confinement advantages against interception while introducing new attack vectors through laser jamming and spoofing techniques. The paper evaluates quantum key distribution (QKD) integration possibilities for ultra-secure satellite communications. Future outlook indicates convergence toward hybrid RF/optical ground stations providing redundancy and weather-independent connectivity, development of portable and rapidly deployable OGS systems for emergency communications, and integration with terrestrial fiber networks for seamless global connectivity. The analysis concludes that optical ground stations represent a critical enabling technology for the next generation of satellite-based internet services, Earth observation missions, and deep space exploration, despite ongoing technical and economic challenges requiring continued innovation and international coordination.

Earth observation satellites are increasingly expected to support time-critical disaster response tasks such as wildfire monitoring, flood damage assessment, post-earthquake damage assessment, and landslide monitoring. However, conventional Earth observation operations remain largely ground-centric: observation requests are mainly planned on the ground, large-volume image data are downlinked, and additional tasking is determined after ground-based analysis. When the acquired data are insufficient for task completion due to cloud cover, smoke, observation conditions, or visibility constraints, this process leads to inefficient use of limited downlink resources and delays in disaster situation assessment and additional observation decisions. This paper proposes a task-oriented observation request control framework over constrained inter-satellite links. We consider an upper-level integrating satellite that receives high-level disaster monitoring tasks from the ground and cooperatively uses surrounding Earth observation satellites as observation resources. The surrounding satellites are assumed to be low-autonomy satellites, similar to many currently operated observation satellites: they can observe specified areas of interest and return observation data and basic metadata, but they do not have advanced onboard reasoning or task-understanding capabilities. The upper-level satellite evaluates observation results, inter-satellite communication opportunities, and task requirements, and generates additional observation requests to neighboring satellites when the currently available information is insufficient for disaster situation assessment. To support reuse across different Earth-observation tasks, a vision-language model (VLM) is used to interpret observation imagery. The proposed framework does not aim to replace global ground-based mission planning. Instead, it focuses on on-orbit local planning and retasking. In the proposed approach, the ground provides a target monitoring area and high-level task requirements, while the upper-level satellite selects the next available observation satellite and issues an additional observation request when re-observation is required under constrained inter-satellite communication opportunities. By making such decisions before all raw data are downlinked to the ground, the framework reduces unnecessary ground downlink volume and shortens the time required to obtain disaster-relevant information. The effectiveness of the proposed framework is evaluated through task-oriented simulations based on a wildfire monitoring scenario. VIIRS-derived thermal imagery from TS-SatFire is divided into 32 × 32 pixel tiles, and a VLM classifies each usable tile as FIRE or NO_FIRE. Observation states that are unusable or insufficient because of conditions such as cloud cover or data loss are represented in a simplified manner as a full-tile blackout that triggers re-observation. The proposed architecture is compared with a ground-centric baseline in a simplified low-Earth-orbit scenario with four observation satellites, one integrating AI satellite, and one ground station. Across ten wildfire dates, the VLM achieves a pooled recall of 77.6%, precision of 71.4%, and F2 score of 76.3%. In Monte Carlo experiments with blackout probabilities from 0% to 50%, the 50% blackout case reduces mean operational completion time from 930.8 min to 468.6 min (50.2%) and reduces mean ground downlink volume by 98.8%. The contribution of this work is a communication-efficient cooperative observation architecture that enables existing or low-autonomy observation satellites to be effectively used under constrained inter-satellite links for time-critical disaster response.

In-flight connectivity is approaching a critical inflection point as passenger expectations for streaming-class bandwidth and low-latency applications are rising, while Wifi only cabin networks remain constrained by finite spectrum, interference, satellite backhaul variability, and peak-demand congestion. These technical limitations directly translate into business impacts customer dissatisfaction, premium brand erosion, uneven monetization of connectivity tiers and operational risk when mission-critical crew communications compete with passenger traffic. Existing solutions are largely reactive, offer limited determinism in handoffs, and insufficiently incorporate flight-phase rules, environmental constraints (turbulence, illumination, altitude), and predictable passenger behavior patterns that drive transient but recurring demand spikes. We present a deterministic, safety-aware hybrid Wifi–LiFi cabin networking architecture designed to improve service reliability, maximize usable spectrum, and enable differentiated, monetizable connectivity experiences without compromising regulatory compliance or operational safety. The system couples RF connectivity (2.4/5 GHz WiFi with cellular backup and Ku/Ka satellite backhaul) with a LiFi subsystem integrated into reading lamps and cabin lighting, using full-spectrum wavelength division multiplexing across seven visible bands (380–700 nm) to deliver high-capacity, localized links and reduce RF contention. Novel contributions include: (i) A Central Decision Engine that performs multi-factor weighted, deterministic switching between Wifi and LiFi using flight phase, seatbelt compliance and passenger movement, flight duration class, crew precedence, turbulence and lighting feasibility, and spectrum-zone optimization. (ii) A predictive demand layer that leverages deterministic passenger-behavior models (e.g., post–meal movement and bathroom-queue forecasting, long-haul sleep cycles) to pre-position capacity before congestion occurs. (iii) A zonal wavelength orchestration framework that assigns color bands by cabin section and dynamically rebalances wavelengths across zones to maintain fairness and service-level objectives under varying occupancy and illumination. The architecture also introduces a four-tier priority controller that guarantees flight deck performance, ensures resilient cabin crew communications (including short-range infrared LiFi), and enables differentiated premium and economy passenger services with intelligent queue management. Implementation provides commercial aviation with superior passenger experience through seamless adaptive connectivity, operational efficiency through intelligent bandwidth management, crew communication enhancement through priority-based resource allocation, and regulatory compliance framework ensuring safety-critical operation throughout all flight phases. From a business perspective, this approach increases effective cabin throughput by offloading traffic from Wifi to optical spectrum, improves customer experience consistency, and enables clearer product segmentation. It also reduces operational disruptions via deterministic policies and safety-aligned controls, while leveraging existing cabin lighting infrastructure to lower retrofit complexity and improve total cost of ownership. The resulting system provides a scalable pathway to next-generation aircraft connectivity with measurable gains in reliability, monetization, and operational assurance.

The proliferation of mega-constellations of satellites in Non-Geostationary Orbit (NGSO) constitutes a growing threat to the orbit-spectrum resources used by Geostationary (GEO) Fixed Satellite Service (FSS) systems. The increasing number of potential interference sources makes it necessary to develop efficient methods for estimating interference parameters, while guaranteeing statistically sound results for regulatory compliance analysis. This work presents a Monte Carlo simulation methodology for the probabilistic evaluation of aggregate NGSO interference on GEO-FSS downlinks, in Ku-band (11–14 GHz) and Ka-band (20–30 GHz), designed to significantly reduce computation time compared to exhaustive ITU methods. The method estimates two regulatory quantities: the Equivalent Power Flux Density (EPFD, ITU-R S.1503-3, RR Article 22) and the Interference-to-Noise ratio (I/N), as well as the Carrier-to-Interference ratio (C/I), used as an internal diagnostic indicator. The I/N criterion for NGSO systems is not yet subject to a fixed limit in the Radio Regulations and constitutes a central point of discussion within Working Party 4A ahead of the 2027 World Radiocommunication Conference (WRC-27). An essential contribution lies in the generation of satellite positions. NGSO satellites are randomly positioned within an orbital corridor centered on their nominal orbit, reflecting the actual station-keeping tolerances observed in commercial constellations. Several generation laws are implemented and compared, ranging from reference distributions to physically motivated models prescribed by ITU-R S.1503-4. This choice is not neutral: the law selected directly conditions the simulated EPFD distribution and, consequently, the conclusions regarding regulatory compliance. The methodology covers single orbital plane configurations as well as full Walker Delta and Walker Star constellations (T/P/F parametrization), with sampling bounds derived from Theorem 1 of Choi & Baccelli (2025). The strength of this approach lies in its rigorous statistical analysis framework. Since ITU regulatory criteria are expressed in terms of exceedance probability, the complete distribution of the simulated quantities is estimated through the empirical cumulative distribution function (ECDF). DKW confidence bounds guarantee the validity of this estimation on a finite sample, while Clopper-Pearson exact intervals provide an exact bound on the probability of exceeding a given threshold, thereby offering a quantified guarantee absent from deterministic ITU methods. Kolmogorov-Smirnov and Ljung-Box tests complete the framework, validating respectively the distribution and the independence of the realizations. As a future direction, identifying an analytical law specific to each simulated quantity (EPFD, I/N, C/I) constitutes a long-term avenue of investigation, potentially leading to parametric formulas directly usable for spectral coordination. The simulation is validated against Visualyse Professional v7.10.10.0 (Transfinite Systems), an ITU-recognized reference tool, across three levels of comparison: single snapshot, temporal sequences over percentiles, and full Monte Carlo distributional comparison via Kolmogorov-Smirnov test at α = 5%. This approach constitutes a computationally efficient alternative to exhaustive ITU methods, with direct applicability to aggregate interference limit studies in the framework of WRC-27.

In this work we present and demonstrate an open-source, terrain- and regolith-aware propagation framework for planning cellular network infrastructure on the lunar surface. Scalable surface communications are identified by NASA's Moon to Mars Architecture as a foundational capability for sustained robotic and human operations, and a first step was achieved in March 2025, when the Nokia Bell Labs Lunar Surface Communications System aboard the Intuitive Machines IM-2 mission delivered the first cellular network to the lunar south pole. As a lunar base expands beyond its landing site, planners must determine the number, location, height, and frequency band of access points connecting habitats, extravehicular activity crews, rovers, and autonomous robots across cratered terrain. However, prior lunar-surface propagation studies face limitations, including 1) reliance on flat-Earth geometry that fails beyond the roughly 10 km horizon, 2) omission of terrain diffraction, surface roughness, and spatially varying regolith dielectric properties, and 3) generation of dense statistical channels untied to the underlying physics and unsuited to hardware emulation. The proposed framework overcomes these challenges and consists of four core modules leveraging novel algorithms: Terrain Modeling, which processes Lunar Orbiter Laser Altimeter (LOLA) elevation data at 5 m per pixel and applies a spherical-Moon line-of-sight test; Propagation Physics, which adds coherent ground reflection with lunar Fresnel coefficients, an Ament/Beckmann rough-surface factor, multi-edge Deygout diffraction, and per-ray weighting by the standardized cellular antenna element; Regolith Characterization, which incorporates published loss-tangent maps and evaluates radar sounder stratigraphy at 12, 24, and 38 m depth; and Channel Synthesis, which reduces each link to one to three physically determined tap-delay-line coefficients that a hardware channel emulator can replay directly. The framework is applied to links at 442 MHz, 2.5 GHz, and 27 GHz, extended to 28 GHz and 140 GHz, for users with distinct power, antenna, mobility, and traffic profiles. The analysis shows that subsurface echoes lie roughly 75 dB below the direct ray and are negligible at all frequencies considered, while the direct and ground-reflected rays collapse into a single tap carrying the interference null structure. Coverage is found to be horizon- and terrain-bound: siting, band, and the spherical horizon set the footprint, while antenna pattern, roughness, polarization, and regolith govern link quality within it. The ground reflection remains coherent through S-band but becomes largely diffuse at Ka-band and above, and the regolith loss tangent, varying over threefold between candidate sites, moves the grazing link by less than 1 dB. Each propagation effect is anchored to an established baseline model, and the exported channel is independently reproduced in a standard cellular system simulator to within 0.001 dB. The study delivers georeferenced line-of-sight and link-margin maps for the Connecting Ridge and Shackleton Artemis candidate sites at all bands, together with a Colosseum-compatible rover trajectory quantized onto the emulator's four-tap, 10 ns grid, for hardware-in-the-loop evaluation. These results extend the IM-2 cellular demonstration into access-point spacing, mast-height, and band-selection guidelines for the progressive deployment of a sustained lunar-base surface network.

We present an empirical study of distributed multi-hop relay selection in a simulated UAV mesh, where link quality depends on distance and a stochastic line-of-sight obstruction process, each node independently chooses its next hop, and every route change incurs a per-node switching blackout. Because a packet must succeed at every hop, fewer hops confer a structural delivery advantage, but only if a policy can recognize when a longer skip-ahead hop is actually better. We compare a fixed-next-hop baseline against greedy, EWMA-smoothed failure-aware, and ground-truth greedy policies, using 30 paired seeds; ± throughout is a 95% confidence-interval half-width on the mean. At moderate obstruction persistence, all three adaptive policies improve PDR over static with non-overlapping intervals (static 0.277 ± 0.029 vs. ~0.36 ± 0.02); failure-aware and greedy are not distinguishable from each other. A sweep over persistence yields the central finding: at low persistence, every adaptive policy — including ground-truth greedy — falls to roughly one-fifth of static's PDR (0.053 vs. 0.275), because the switching blackout outlasts the condition that triggered the switch; ground-truth knowledge of link state does not help a policy that ignores the cost of acting on it. At high persistence, adaptation keeps its advantage (0.445 vs. 0.380). Greedy and ground-truth greedy pick the same relay on every seed tested, a limitation of this topology's two-candidate-per-hop structure rather than a general property, and a wider candidate set is needed to actually separate them. These results characterize conditions under which reacting to link observations helps or actively harms multi-hop delivery, motivating a switching-cost-aware selection rule as future work.

Encoding and decoding of discrete-time signals are foundational in aerospace applications to compress telemetry data, secure communications against noise, and ensure reliable error-free communications and navigation. Processing discrete-signals at different sampling rates, which is called multirate digital signal processing (DSP), is an important issue in DSP and has a variety of applications. Multirate systems are commonly used for audio and video processing, design of digital frequency converters, digital communication systems, discrete wavelet transform (DWT), digital filters design, to name but a few. This paper proposes a structure for encoding and decoding of discrete-time signals based on downsampling (decimation) and upsampling (interpolation), which are the two main operations used in multirate digital signal processing. We address the problem of decomposing a signal into low and high frequency components and its use in encoding and decoding processes by performing decimation and interpolation in the frequency domain. A two-channel quadrature mirror filter, LPFs, HPFs, decimators, and interpolators are used in implementing the proposed structure. The performance of the proposed structure is evaluated using simulated examples and the results demonstrate its effectiveness.

For low-voltage, high-current applications in radio frequency (RF) communication systems and renewable-energy interfaces, this paper develops a large-signal averaged model of a zero-voltage-switched (ZVS), phase-shifted full-bridge (PSFB) converter with a current-doubler output (CDO). The derivation explicitly incorporates the large-signal effects of the isolation-transformer’s leakage inductance during continuous-conduction-mode (CCM) operation. From the resulting averaged power-stage model, the fundamental small-signal frequency-response characteristics are extracted and formulated for straightforward implementation in circuit-analysis programs such as PSpice and LTspice. To enable bidirectional power flow between two isolated voltage buses, two current-mode regulation approaches are discussed and mathematically integrated with the CCM averaged PSFB-CDO power-stage model. A complete closed-loop model is then derived, including both the outer voltage-regulation loop and the inner current-mode control loop with active ramp compensation. The ramp-compensated current-mode formulation suppresses subharmonic oscillation in CCM while maintaining transformer flux balance and avoiding magnetic saturation. The proposed closed-loop averaged model is validated against an as-implemented pulse-by-pulse switching model of the same dual-loop controlled converter.

This paper proposes an action-oriented semantic item reporting framework for ground-satellite artificial intelligence (AI) agent communication over constrained links. With the progress of onboard AI, Earth-observation satellites are increasingly able to extract not only object candidates from sensor data but also semantic information useful for subsequent operational decisions. However, ground-satellite links are often limited by bandwidth, latency, visibility windows, and intermittent communication opportunities. Therefore, transmitting raw observation data or all onboard AI outputs to the ground is not always feasible, especially for time-sensitive monitoring tasks. Existing semantic and goal-oriented communication studies have investigated how to transmit information useful for inference or logical decision making at the receiver. In ground-satellite AI-agent collaboration, however, the ground agent often needs to select an operational action rather than merely determine whether an object is detected. In maritime monitoring, for example, a vessel candidate observed from satellite imagery may require additional confirmation, reobservation, or no further action depending on semantic information such as estimated vessel length, automatic identification system (AIS) association, and false-target assessment. The semantic items required for such action selection do not necessarily coincide with the features useful for the original detection task. To address this problem, we introduce semantic item reporting, in which an observing satellite AI agent preferentially acquires and reports semantic items useful for next-action selection by a ground decision-making agent. In the proposed framework, a natural-language Ground Request is interpreted into a structured Task Specification, and a common large language model (LLM)-based onboard agent selects the semantic information and expert functions required for each candidate. A deterministic Tool Dependency Resolver enforces implementation prerequisites, while a three-valued Sufficiency Guard determines whether the acquired information is sufficient for action selection. When the semantic state cannot be resolved onboard, a region of interest (ROI) is reported for ground-side inspection. The selected semantic or ROI payload is then passed to a separate communication-intent stage and a deterministic scheduler for finite-capacity downlink allocation. We evaluate the proposed framework in a proxy-based vessel-monitoring scenario using Sentinel-2-derived vessel candidates and three controlled Ground Requests. Across 1,000 candidates per request, changing only the Ground Request substantially changes the selected expert functions while the same function catalog remains available; for example, AIS usage decreases from 100% to 3.8% when AIS is removed from the decision condition, and length-estimation usage decreases from 54.7% to 2.7% when vessel length is removed. No Agent- or Guard-induced decision failures are observed in the 3,000 candidate runs. A separate communication-capacity evaluation further shows that, under the adopted payload model, the proposed method requires more downlink capacity than a task-specific deterministic rule but substantially less than transmitting all ROIs or full-scene imagery. The main contributions of this work are the formulation of semantic item reporting for ground-satellite AI-agent communication, a Ground-Request-driven common-agent framework that links onboard semantic acquisition with ground-side action selection, and a constrained-link reporting architecture that combines adaptive semantic acquisition with deterministic sufficiency control and finite-capacity scheduling.

Low-Earth-orbit (LEO) constellations now acquire far more data than they can downlink, motivating on-board machine learning. Federated learning (FL) is the natural fit--it keeps raw data on each satellite and exchanges only model parameters--but the inter-satellite links it must run over are scheduled and intermittent: optical pointing and acquisition takes tens of seconds, contact windows last minutes, and eclipse periodically removes power. Synchronous FL stalls under such connectivity, and most prior satellite-FL designs depend on a ground station as the parameter server, reintroducing the very downlink bottleneck that on-board learning aims to remove. This paper formulates the constellation as a delay-tolerant, data-parallel training cluster whose collective-communication fabric is a Delay/Disruption-Tolerant Network (DTN) running Bundle Protocol version 7 (BPv7). Each satellite performs real on-board stochastic gradient descent on its own non-IID data and exchanges full-parameter model updates as bundles routed by contact-graph routing with custody and store-carry-forward. Training is satellite-only--no ground station and no central server in the loop--using asynchronous decentralized gossip with a proximal anchor and staleness weighting. We build the system end-to-end: a memory-safe Rust BPv7 node (with demonstrated routing parity against NASA ION and bundle-wire interoperability with uD3TN) and JAX/Flax learners, evaluated on a software-in-the-loop testbed with results reported as mean+/-std over many seeds, and additionally validated hardware-in-the-loop across three physical NVIDIA Jetson AGX Orin boards (one satellite per board, model bundles gossiped over real Ethernet through the identical node). Our central, deliberately narrow finding concerns what the DTN actually contributes. Pinning the compute-to-communication ratio makes convergence machine-independent: with frequent gossip, satellite-only FL matches the centralized upper bound (anomaly-detection AUC 1.00 versus 0.68 isolated), degrading gracefully along a communication/accuracy tradeoff as on-orbit contact thins (0.80 at one exchange per 300 local steps). A multi-seed store-carry-forward ablation then isolates the network's role: it delivers every model update while a delay-intolerant transport drops up to 88%, yet final accuracy is unchanged within seed noise--and this holds across stationary regimes, non-stationary drift, and a 237x larger model. Decentralized gossip is robust to heavy update loss--so the DTN's decisive value is to make satellite-only asynchronous federation feasible at all, not to raise the converged model (overturning an earlier single-run impression that seeding does not support). The substrate is task-agnostic (a control policy closes 0.99+/-0.01 of the gap), scales to a multi-plane constellation, and tolerates Byzantine satellites (centered clipping and the geometric median hold to a 33% attacker fraction, where distance-based rules collapse). At deployed scale on real Sentinel-2 imagery (EuroSAT, 2.3M-parameter model, 9 MB bundles), federation closes roughly half the gap--the real-data non-IID benefit being milder than synthetic. To our knowledge this is the first end-to-end demonstration of federated training carried over a real BPv7 DTN, and the multi-board run reproduces the full gap closure (0.69->1.00) with zero-loss delivery of 600 bundles. The memory-safe node targets a rad-tolerant on-board computer, positioning satellite-only federated training for on-orbit deployment.

Digital twins for aerospace and other high complexity engineered systems require the continuous synchronization of system requirements, system state, simulation models, and verification evidence. However, most large scale engineering environments still depend on document centric workflows that fragment requirements, design intent, configuration history, and operational assumptions across disconnected tools and organizational boundaries. This paper presents an AI enabled digital engineering and systems integration framework—derived from mission proven JPL aerospace infrastructure—that unifies requirements traceability, state traceability, interface management, configuration evolution tracking, and simulation ready model structures into an executable Digital Twin backbone. The framework was evaluated through a detailed application to a representative offshore industrial platform architecture featuring multi year lifecycles, multiple contractors, and high consequence integration constraints analogous to those encountered in aerospace missions. The study analyzed approximately 4,000 pages of engineering documentation, including platform level design bases, ICS specifications, and more than 50 engineering standards, and validated findings through six structured multidisciplinary working sessions. Five systemic gaps identified in these industrial workflows parallel well known aerospace challenges: (1) cross discipline coordination performed informally without a dedicated systems integrator function; (2) contractor boundary misalignment, leading to divergent assumptions across design organizations; (3) underutilized Minimum Functional Objective (MFO) frameworks lacking digital traceability into equipment level requirements; (4) standards driven requirement volume exceeding 100,000 discrete items per platform, obscuring high impact engineering decisions; and (5) configuration drift between evolving engineering standards and static project design documents, with no automated mechanism for alignment or impact propagation. These conditions mirror the complexity and coupling patterns seen in multi contractor space programs, making them ideal test cases for aerospace grade digital engineering solutions. Using JPL’s Requirements Development and Qualification Suite (RDQS), we demonstrated an integrated pipeline including AI assisted ingestion of engineering standards, automated extraction and classification of requirements, hierarchical MFO → system → subsystem → component decomposition, verification evidence chains, and bidirectional change propagation. Critically, the architecture supports both requirements traceability and state traceability—linking system behavior, assumptions, and configuration state to their originating requirements and enabling continuous synchronization between physical assets, design intent, and digital twin simulation models. Automated standards currency tracking flags when design bases no longer reflect current technical references; interface control structures and suspect change propagation reveal impacts across discipline and contractor boundaries before fabrication or commissioning. A prototype digital twin slice was developed for a representative heat exchanger subsystem, demonstrating how model linked requirements, digital verification chains, and state aware configuration tracking provide real time impact visibility, accelerate engineering decision cycles, and reduce late stage integration conflicts. The results show that applying aerospace grade systems engineering—augmented with AI and simulation coupling—provides a scalable, lifecycle consistent architecture for Digital Twin development, suitable for both space missions and complex terrestrial industrial systems. A flare system pilot is proposed as the next step to institutionalize these capabilities and establish a reusable template for full mission or full facility digital twin implementations.

The current traffic management system is struggling to integrate unmanned aircraft systems (UAS) into national airspace, particularly in scenarios expected to become congested due to large-scale deployment of UAS for applications such as package delivery, food delivery, and other services. Currently, several industry operators working on UAS delivery uses cloud-based information exchange between the UAS and ground station. However, as more suppliers adopt drone delivery and operations scale up, managing low-altitude traffic through cloud will become increasingly challenging. To address this challenge, this paper proposes a universal and scalable technological solution to ensure safe and scalable operation of UAS at low altitudes beyond visual line of sight (BVLOS) when involving different third-party services, including UAS Traffic Management (UTM). To achieve this goal, a universal decentralized situational awareness and traffic management framework is developed using multi-modal communication (Bluetooth, non-cellular 5-G, etc.) for various range profiles, as well as collision risk advisory techniques for trajectory prediction and collision risk assessment. A Digital Enhanced Cordless Telecommunications New Radio Plus (DECT NR+), the first non-cellular 5G network, is employed along Bluetooth to enables ultra-reliable, low-latency communication among UAS and between UAS and ground stations, particularly during unmitigated collision risks, where each agent has access to shared traffic data in a nearby environment without dependency on a central server. A deterministic and resource constrained model-based trajectory prediction and risk assessment is designed and implemented on embedded framework using the current position and speed of the UAS from on-board sensors such as Global Positioning System (GPS) and Inertial Measurement Unit (IMU), thereby preserving the privacy of third-party customers while increasing interoperability across heterogeneous UAS fleets. To demonstrate the effectiveness of the proposed framework, a quantitative assessment across various metrics, including communication range, data payload, and Received Signal Strength Indicator (RSSI), has been performed. The proposed framework supports Federal Aviation Administration (FAA) priorities of advancing air mobility, enhancing safety, and enabling equitable access to emerging transportation technologies by enabling a traffic management that could handle the dense UAS BVLOS operation from dozens to hundreds per square mile, while reducing communication latency from 100 ms to 10 ms, and extend operational range from ~2 miles to over 20 miles through DECT NR+ mesh networking. The contributions of this work are summarized as follows: (1) Develop decentralized architecture, multi-modal communication strategy, and safety mechanisms for collision advisory to support safe, scalable BVLOS operations. Our system integrates Bluetooth for short-range, DECT NR+ for long-range, leveraging its mesh-based, non-cellular 5G framework, and Wi-Fi for ground station interactions, enabling dynamic adaptation to communication conditions. (2) Design a collision risk advisory for UAS operations in congested airspace as part of traffic management for small fleets to network-level BVLOS coordination across potentially hundreds of UAS in dense urban corridors. The deterministic and resource constrained model-based trajectory prediction and risk assessment is designed and implemented on embedded framework. (3) Perform a quantitative assessment for efficiency of proposed framework across various metrics including range of communication, data payload, and RSSI.

Detection of weak signals in space operations has multiple applications, from space awareness, to scientific discoveries, to search and rescue operations. This paper quantifies the limits to signal detectability using power spectral density estimation and/or spectrograms, a well-known and widely applied method both in amateur settings (e.g., using software defined radios), research activities, and in space operations. In particular, we derive expressions that relate detectability with metrics that are quantifiable a priori such as power to spectral noise density or residual Doppler rate, thus providing a meaningful to determine whether a signal is detectable or not, and which configuration parameters should be used to achieve detection. To validate the derived equations, we use both simulations and real life data from the Voyager 1 spacecraft to show that detectability is a meaningful metric for deciding if a signal will be visible or not in a spectrogram. We also show basic rules of thumb for how to select the parameters that drive the spectrogram estimation, thus providing guidelines for antenna and mission operators for how to configure the system during search and rescue activities.

This paper develops and evaluates a fuzzy-logic handover strategy, an interpretable artificial-intelligence technique, for low Earth orbit (LEO) satellite networks applied to aeronautical connectivity, using a purpose-built simulation environment. Non-terrestrial networks are becoming a core component of next-generation mobile communications, and LEO constellations are especially attractive for in-flight connectivity because they offer lower propagation delay, higher throughput, and reduced attenuation compared with geostationary systems. Their high orbital velocity, however, continuously changes the network topology and forces frequent link handovers, even for a stationary terminal. Short visibility windows and multiple simultaneous candidate satellites make the handover decision difficult and sensitive to degradations such as increased signaling, latency, and ping-pong events. Classical strategies based on a single deterministic criterion, and more recent graph-based or deep-learning approaches, either deliver partial performance or introduce training cost and limited interpretability. This work investigates whether a low-complexity, interpretable fuzzy policy can improve link quality without destabilizing mobility. To answer this, a modular simulator was developed in Python, comprising synthetic scenario generation, handover decision, temporal execution, and metric analysis. Each visible satellite is represented by a SINR profile combining a quality peak, temporal curvature, Gaussian noise, and random shadowing, reproducing the approach, dwell, and departure of satellites relative to an aircraft. Two decision policies are compared under identical input scenarios to isolate the effect of the policy: a baseline using hysteresis and time-to-trigger, and a Mamdani fuzzy inference system. The fuzzy policy combines three inputs, the current SINR of the serving link, the temporal SINR trend, and the SINR difference between the candidate and serving links, and produces a handover urgency that is compared against a decision threshold. Evaluation was carried out at two levels: single-run temporal analysis in a fixed scenario, and statistical validation over one thousand independent random seeds, with the same scenario applied to both policies in each seed to guarantee fair comparison. Across the one thousand seeds, the fuzzy policy consistently outperformed the baseline on the main link-quality metrics. Mean SINR increased from 13.24 to 13.43 dB, and fifth-percentile SINR, which captures the worst moments of the link, increased from 8.46 to 8.81 dB, a larger relative gain that is particularly relevant for service continuity in flight. The average number of handovers rose only slightly, from 2.89 to 2.97, while ping-pong events remained negligible and time in outage decreased, indicating that the higher context sensitivity of the fuzzy policy does not destabilize mobility. Both improvements were statistically significant under paired testing, with p below 0.001 and confidence intervals excluding zero, and the fuzzy policy was superior in the large majority of the one thousand seeds. By coupling a simulation-based evaluation with an interpretable artificial-intelligence decision method, this study shows that a fuzzy strategy of low structural complexity achieves consistent, statistically significant gains over a classical baseline in LEO aeronautical scenarios, offering a favorable balance between link quality, service continuity, and mobility cost, and a transparent, adjustable alternative to deep-learning handover methods.

Space-to-ground (S2G) free-space optical communication links experience significant performance degradation due to atmospheric turbulence-induced fading, with fade penalties ranging from 20 dB to over 50 dB depending on link configuration and atmospheric conditions. This paper presents a comprehensive performance analysis of digital layer fade mitigation techniques for S2G optical links operating at data rates from 2.5 Gbps (SDA OCT 4.0 standard) to 100 Gbps (OpenZR+), focusing on automatic repeat request (ARQ) protocols and channel interleaving. Our analysis reveals that ARQ with selective repeat protocol can reduce fade penalties from 21 dB to 10.7 dB for uplink with 4-beam spatial diversity, though effectiveness diminishes in severe single-beam scenarios where penalties remain around 30 dB. We characterize the fundamental tradeoffs between link margin reduction, throughput, latency, and delay jitter for various ARQ configurations. Channel interleaving with sufficient diversity can achieve near-optimal performance approaching the infinite interleaver limit, reducing fade penalties to approximately 5 dB, but requires fixed latencies of 50-100 ms. Implementation considerations are addressed for both ARQ and channel interleaving, including memory requirements, latency constraints, and hardware complexity at different data rates. The paper concludes with specific recommendations for digital layer fade mitigation strategies tailored to link data rate, atmospheric severity, and application requirements, providing a framework for optimizing S2G optical link design in the presence of atmospheric turbulence.

Frequency mismatch between the transmit and the receive oscillators as well as Doppler effect introduces carrier frequency offset in the received signal. The impact can be severe in highly dynamic LEO satellite communications. In this paper, we proposed a feedforward carrier frequency synchronization algorithm that is based on Maximum Likelihood Estimation (MLE). We compare our MLE method with other feedforward methods in a numerical example to demonstrate the rapid acquisition time and small tracking errors.

Aerospace communication systems operate across highly dynamic and heterogeneous radio-frequency (RF) environments in which communication reliability is jointly influenced by platform motion, propagation geometry, obstruction, multipath, interference, antenna characteristics, and other electromagnetic and physical conditions. A fundamental challenge is that distinct environmental mechanisms can produce similar observable communication failures. For example, Doppler induced by platform dynamics, line-of-sight (LOS) blockage, and external interference may all manifest as increased bit errors, packet loss, or reduced link quality, while requiring fundamentally different mitigation strategies. Conventional link adaptation approaches largely respond to aggregate indicators such as signal-to-noise ratio (SNR) or packet error rate and therefore provide limited ability to determine why a link is degrading. This limitation becomes particularly important for aerospace systems operating across changing environments, where communication strategies designed for one physical condition may be ineffective or unnecessarily conservative in another. This paper introduces a causal reasoning framework for environment-aware aerospace communications that explicitly models how physical and electromagnetic conditions propagate through measurable RF signatures to communication-layer outcomes. Rather than treating the wireless channel as a scalar link-quality measurement, the proposed framework represents the physical-to-communication dependency chain using a Bayesian network. The network relates candidate physical mechanisms—including Doppler, LOS/NLOS blockage, and interference—to modulation-agnostic RF measurements derived directly from raw in-phase/quadrature (I/Q) samples, including SNR, received signal strength, channel noise, frequency offset, Doppler, and timing jitter. These observations are probabilistically connected to bit error rate and packet success, enabling posterior inference in the reverse direction: given observed RF and communication degradation, the system estimates the most probable underlying physical cause. This causal structure provides both uncertainty-aware diagnosis and an interpretable reasoning path from environmental condition to communication consequence. The framework is implemented and evaluated using an over-the-air software-defined-radio testbed integrating GNU Radio, BladeRF transceivers, and ArduPilot software-in-the-loop flight dynamics. Controlled experiments reproduce distinct dynamic RF conditions through time-varying Doppler injection, physical LOS/NLOS obstruction, and channel interference. The experiments evaluate whether the Bayesian model can distinguish degradation mechanisms that may produce overlapping link-level symptoms and whether its posterior attribution remains coherent across changing operating conditions. The use of modulation-agnostic RF evidence further separates environmental reasoning from a particular waveform or communication control policy. The resulting framework establishes a foundation for environment-aware aerospace communication systems in which adaptation is driven not only by observed link degradation, but also by an understanding of its probable physical cause. Such causal knowledge can support downstream decisions including adaptive modulation and coding, frequency and link selection, routing, interference mitigation, and intelligent communication control. More broadly, the framework provides an extensible approach for reasoning about heterogeneous aerospace RF environments: additional environmental mechanisms and corresponding RF signatures can be incorporated as the operational domain changes. By connecting physical environment, RF behavior, and communication reliability within a unified probabilistic model, this work takes a step toward aerospace communication networks that can understand their operating environment and adapt communication strategies accordingly.

Hierarchical Deep Reinforcement Learning (HDRL) is an area of Reinforcement Learning that that can decompose complex tasks into smaller subtasks using a hierarchy of Deep Reinforcement Learning (DRL) algorithms. This allows HDRL to solve hard problems when complex tasks can be decomposed into subtasks such that individual subtasks are solved using DRL agents. Such an approach to solving complex tasks allows DRL agents to focus on narrower tasks, and gives HDRL a very broad application space. This paper considers techniques to decompose complex tasks for DRL implementation. Furthermore, since an HDRL system may include many DRL agents combined in a multiple layer hierarchy, this paper considers the specific challenges associated with achieving overall system performance. Specifically, HDRL challenges are a superset of those associated with DRL system design. Finally, a multi‐FPGA based implementation of an HDRL system can provide a high performance, high throughput, and low latency AI solution. Design data for high performance HDRL systems to support AI applications using FPGAs is provided herein. The implementation discussion and details in this paper focus on high performance FPGA design.

A launch or mission console already has the radios. What it does not have is a way to hear ten to twenty encrypted voice nets at once without taking the operator off the board. Time is lost in Observe and Orient, the front of the OODA loop. Echo C2 AI attaches after decryption, at the speaker line. It does not sit inside the crypto. The recovered audio is the line in, so the protocol under the waveform does not matter: digital nets, NATO-standard encrypted waveforms, Army or Navy stacks. Mixed nets can share one pipeline. Speech-to-text is the first stage, not the system. Encrypted radio leaves artifacts that ordinary transcription does not see, including static on the recovered audio. A cleaning stage treats those artifacts so the transcript used downstream is as accurate as the channel allows. Mission recordings can train customer-specific models that improve those artifacts over time and lock transcription to unit terminology and mission context. Ten to twenty nets run at once. Each net has a dedicated agent that summarizes that channel. Those summaries are shared across baseline users. Each operator can also run a personal agent on each net that attends to what that operator cares about. Separate agents scan transcripts for operator-defined keywords and raise flags when those topics appear. A mission agent ingests all transcripts and summaries, weights the nets that operator has prioritized, and produces a summary of summaries. The scarce resource is attention. The console gets one glance at the mission without scanning every channel. Transcripts and summaries can be viewed inside ATAK so the operator does not leave the C2 client. A 2026 maritime interdiction ran on a dedicated edge server for fully offline transcription and summarization. At Tough Stump in June 2026, Echo ran a TAK server at the edge and integrated with encrypted radios. At PACOM’s Cobra Gold 2026 in Thailand, Echo connected to a Stryker, transcribed Thai communications into English for Stryker battalion operators, and processed multiple channels. Launch-ops situational awareness is the aerospace instance of the same stack. This paper describes the pipeline as a communications and AI system: where it attaches to the waveform, why encrypted-radio artifacts need their own cleaning stage, how concurrent agents are tasked per net and per operator, and how the mission agent compresses the Observe and Orient stages of the OODA loop without replacing Decide. Unclassified. No concept-of-operations details under public-release review are claimed.

NR cell search within NTNs must tolerate combinations of residual frequency offset, Doppler rate, and channel evolution such that repeated observations build up independent phases between synchronization signals and search spaces. This work characterizes a fixed Hybrid 2+2 receiver that combines four observations and defines successful cell search as detection of timing, PSS, SSS, and PCI. We analyze stage-resolved reliabilities, relative first-loss boundaries, normalized impairment coordinates, and joint stresses within reference TDL-C and NTN-referenced channel profiles. The joint synchronization reliability first dips below 90% nominal at 6.19 39 Hz residual frequency and 170.52 24 Hz/s Doppler rate with 20-ms observation spacing. Sweeps across Doppler rate are nonmonotonic at low retention levels, showing level re-entry after the first loss at both R75 and R50. At 544 Hz/ s deterministic Doppler-rate stress, timing and PSS success are 0.811 and 0.584, respectively, when SSS, PCI, and joint synchronization fall to 0.263, 0.220, and 0.220 within NTN- TDLC5 LOS. Matched normalized channel-evolution tests at 20- and 160-ms observation spacing align closely within the tested TDL-C configuration, while factorial testing reveals profile-dependent deviations from the selected multiplicative-independence baseline. Within the evaluated domain, impairments due to residual frequency are largely monotonic, while the quadratic phase evolution induced by Doppler rate can cause loss and recovery. This stage-resolved analysis reveals which downstream acquisition decisions fail even when PSS detection appears comparatively robust. Overall, these results show that PSS-only detection does not necessarily characterize the reliability of repeated-observation cell search and that impairment interactions can depend on channel profile. Reported first-loss coordinates are empirical values describing sensitivity within the tested receiver and should not be interpreted as Universal NR/NTN synchronization limits or 3GPP conformance boundaries.

Constant-Envelope (CE) multicarrier waveforms (CE-OFDM and CE-SC-OFDM) demonstrated immunity to nonlinear distortion, thanks to their 0dB PAPR, and a noticeable resilience to multipath fading, thanks to their augmented diversity. The tradeoff in terms of increased bandwidth occupation could be acceptable in non-terrestrial transmission applications where the link budget is constrained by the atmospheric attenuation, and the requirements in terms of spectral efficiency are less stringent than those in terms of robustness. The fundamental detection scheme of CE-MC waveforms is based on the arctangent demodulation. Such a nonlinear operation inverts the phase modulation performed at the transmitter side. The contributions published in the recent literature evidenced the critical issues of the arctangent demodulation in terms of nonlinear effects (threshold effect and phase wrapping), which impose a careful parameterization to the transmission system in terms of modulation index and oversampling factor. However, other issues should be mentioned in terms of the complexity of the hardware implementation. Indeed, the computation of the arctangent function may require up to 12 CORDIC iterations to attain 0.1-degree accuracy. Alternative solutions rely on look-up tables, which are quite expensive in terms of storage space. In our paper, low-complexity solutions for the detection of CE-MC signals are evaluated in satellite transmission scenarios of technical interest for 5G and beyond applications. A linear receiver (BLR) will be first proposed as a baseline characterized by negligible implementation complexity. Then, a semi-linear enhanced receiver (ER) will be considered as a valuable alternative. An extensive campaign of simulations will provide the necessary set of results to assess the performance of the two low complexity receiver schemes, considering the presence of multipath fading and phase noise typical of satellite transmission operating in dense urban scenarios.

Free space optical communication is a technology used in satellite communications and data transmission applications due to its high data rate potential and available bandwidth. Its compatibility with CubeSat swarms makes it a promising technology for further development and research, as CubeSat swarms allow for simultaneous global measurements at lower costs. Using modulating retroreflectors (MRRs) alongside a wide-angle frequency comb beam from a ground station or a large spacecraft can communicate with the swarm all at once to achieve multi-channel high speed communications. The MRRs allow for lower CubeSat power requirements, as the CubeSats do not generate and steer their own laser beams. However, the single wide-angle beam configuration lacks power efficiency, as much of the beam’s light is lost into free space during interrogation. This project proposes the architectural design of a near-infrared 100-mode optical frequency comb with independently steerable beams, controlled by an optical phased array (OPA) with true time delay (TTD) architecture, and paired with previously developed MRR designs for CubeSat swarm interrogation. Using an OPA allows for rapid beam steering rates compared to mechanical steering methods, which minimizes slew time between CubeSats. Using a frequency comb beam keeps the benefit of encoding data simultaneously across all comb frequencies and decoding at the base station through wavelength division multiplexing (WDM). The OPA uses time delay lines and optical switches rather than standard phase-shifters to prevent the introduction of beam squint in a multi-frequency beam such as a frequency comb. Thermo-optic or electro-optic phase shifters can be used after the time delay to correct any time delay quantization errors. An aperiodic array layout is determined using particle swarm optimization (PSO) or similar optimization algorithms to minimize side lobe power level and maximize the main beam power level. Dwell times for each CubeSat for a given test orbital configuration can then be determined through optimization algorithms such as PSO.

Intermittent ground-station visibility limits the timeliness of data return from low-Earth-orbit satellites. A compact terminal communicating through a geostationary-orbit relay satellite provides a practical method for extending contact availability, but its feasibility depends on the transponder operating point, terminal antenna performance, achievable data rate, and link margin. This paper presents an experimental characterization of a compact C-band GEO relay link intended to support the engineering design of small-satellite communication terminals. Two test campaigns were conducted through an operational C-band GEO satellite transponder. The experiments covered both the C-band uplink and downlink and involved a terminal test site and a commercial earth station. The first campaign evaluated antenna pointing, fixed-gain-mode transponder characteristics, and small-signal end-to-end connectivity. The measured receive levels of two horn antennas were -113.2 dBm and -117.1 dBm, compared with calculated values of -110.6 dBm and -116.0 dBm, respectively. The deviations were within approximately 3 dB and were attributed primarily to feeder, atmospheric, and measurement losses. In contrast, the compact microstrip antenna produced a maximum receive level of -129.1 dBm, approximately 8.5 dB below the link-budget prediction, indicating the need for antenna-gain and insertion-loss recalibration. The satellite transponder was characterized over fixed-gain settings from 44 to 94. Measurements showed an average gain increment of approximately 0.5 dB per setting over most of the guaranteed operating range from settings 44 to 88. Linear fitting of the measured input-output data was used to estimate the transponder operating curves and saturation points for different gain settings. At setting 94, a carrier level of -130.48 dBm was observed above a measurement noise floor of approximately -134.5 dBm, confirming small-signal end-to-end connectivity. A follow-up campaign evaluated forward and return links using U-band and L-band modems connected through C-band frequency converters, a power amplifier, and horn or microstrip antennas. With the compact microstrip antenna, the U-band modem achieved a link margin exceeding 10 dB at 300 bit/s, while the L-band modem achieved a margin exceeding 10 dB at rates up to 2.4 kbit/s. The results indicate that an integrated terminal transmit power of approximately 27-30 dBm can support further reductions in terminal size and power consumption. The measurements provide practical guidance for transponder gain selection, antenna design, data-rate configuration, and link-budget allocation in compact satellite relay terminals.

Wideband spectral sensing is essential for detecting interference, tracking spectrum occupancy, and upholding wireless security. Accurate signal detection and labeling provides the information needed to distinguish signal types, sources, and behaviors. However, software-based processing struggles to keep pace with high bandwidths, requiring excessive latency and raw data storage. Real-time, hardware-accelerated signal labeling is therefore needed to identify and characterize signals as they occur, enabling efficient downstream signal analysis and classification. In this paper, we propose HI-WHIRLS, a FPGA-based real-time wideband signal detection and labeling algorithm based on the WHIRLS algorithm, with improvement in both time complexity and low-SNR performance. HI-WHIRLS is implemented on a Xilinx RFSoC ZCU216 evaluation board, utilizing the built in data converters to provide an end-to-end pipeline from wideband RF acquisition to signal labeling. The algorithm uses parallel, overlapped spectral processing to provide independently configurable time and frequency resolutions, making it adaptable to various spectrum environments and signal patterns. The system performs spectral anomaly detection and groups detections across time and frequency to produce labels describing timing, center frequency, and bandwidth. To improve sensitivity, HI-WHIRLS can adaptively configure the amount of temporal averaging based on the characteristics of an initial narrowband or wideband detection. By doing so, HI-WHIRLS can reliably detect and label signals as low as -10 dB, which is a 7 dB improvement from the original WHIRLS algorithm. Finally, we evaluate Hi-WHIRLS quantitatively on simulated datasets and also qualitatively on real over-the air RF signals. Our experiments show that the proposed HI-WHIRLS can reliably perform real-time RF signal detection and labeling with independent time-frequency resolutions and thus provide instant insights into spectrum activity.

The Lunanet standard requires 1 ns accuracy for satellite ranging using pseudo-noise (PN) sequences. A testbed developed to verify this requirement must be 10x better – resulting in the extremely tight accuracy of 100 ps. This paper describes the implementation of the software defined radio for the Lunanet communications and ranging. The software centric architecture imposes constraints unlike those faced by traditional hardware (FPGA) based solutions. Achieving and validating repeatable measurements with 100 ps accuracy is challenging. In this paper we start by describing the techniques used to measure sub-nanosecond delays. Initial measurements on the baseline hardware, USRPs, showed the feasibility of achieving the required timing accuracy. However, tests with the complete system, that includes the signal processing required for modulation and demodulation showed dramatically worse performance. The ranging waveform is phase modulated on the carrier (vs. direct carrier modulation) with a residual carrier. The residual carrier is tracked by a PLL, followed by phase demodulation. The combination of these two steps introduced a multipath-like effect, which in turn resulted in the smearing of the ranging correlation peak. The effect depends on the chip rate of the ranging signal. The unexpected interplay between the spectrum of the PLL phase error and the overall delay measured by the ranging system resulted in a difference of multiple nanoseconds for different chip rates. The initial set of measurements were done using a common clock between the transmitter and the receiver. Subsequent tests using different clocks resulted in both a large (over 100 ns) initial error as well as large (1-5 ns) variance. This behavior was traced to an interplay between the large (100 ms) fixed delay in the software implementation of the ranging target and the stability of the clock. A small (1 ppm) frequency offset is scaled by the fixed delay and results in a significant fixed delay offset. The random-walk of the relatively low-quality clock used for this measurement was also scaled by the fixed delay, resulting in high variance. The paper shows that measurements and analysis of the clock Allan deviation fully explain the magnitude of delay variance. The final, bonus, section of the paper describes one-way delay measurements. We measure the absolute delay from/to the RF port of the testbed to/from the corresponding digital sample generated/recorded by the SDR. The 100ps resolution requires careful attention to both test equipment and cabling.

High data throughput optical communication links across low earth orbit (LEO) and LEO-to-ground presently achieve 100/200 Gbps+. An economical approach is to leverage telecom digital coherent optics (DCO) as demonstrated by SpaceX Starlink, Amazon Kuiper, and NASA TBIRD, etc.. Telecom transceivers, however, are not designed for millisecond-scale fades due to pointing jitter and atmospheric turbulence in up/downlinks. On downlinks adaptive optics systems can mitigate beam scintillation, but on uplinks the beam is more likely to miss the aperture because the atmospheric layer is much closer to the ground station than to the satellite. Fades translate to SNR below the FEC threshold resulting in thousands of frames per fade being dropped. Since the physical layer cannot be altered, higher layer automatic repeat request (ARQ) or packet erasure coding (PEC) for fixed latency can be used. For millisecond-scale fades blocks of hundreds of milliseconds are needed to approach capacity. This can be achieved by using gigabyte-scale PEC blocks or by a combination of shorter PEC blocks and interleaving. For this work we have implemented the CCSDS Flex-IRA LDPC PEC recommendation on a 2017 AMD Epyc CPU using AVX2. The implementation was tested on representative atmospheric channel models of varying severity with code rates from ½ to 31/32. The PEC codec was written in C++, runs on Linux Ubuntu, and leverages the DPDK library to stream Ethernet traffic from an encoder PC to a decoder PC through a QSFP28 transceiver pair using NVIDIA Mellanox ConnectX-5 network interface cards. LDPC computational cost scales linearly with block size, and therefore, large blocks can be used to accommodate our fading channel without interleaving. The codeword is comprised of N=217 packets of tunable size T, which determines the block size. At 100 Gbps a 60kB packet results in 8GB PEC blocks of duration of 62.5ms. User Ethernet frames are used to construct K info packets. Adding N-K parity packets complete the codeword, which is fragmented into Ethernet payloads prior to transmission. The fragments are prepended with headers indexing the source block, packet and fragment so that the decoder defragmenter can reconstruct the partial block prior to decoding. The LDPC encoder works on separate blocks on parallel cores. The main bottleneck was the fragmentation which presently copies data from memory to the DPDK buffer. Our solution is to parallelize the fragmentation/defragmentation by transmitting/receiving separate codewords on dual streams at half rate (50Gbps). The latency can be kept the same by transmitting blocks of half the original size. On the decoder, DPDK Flow Director is used to sort fragments from each stream using a hash. We achieved 100 Gbps coded throughput for code rates of ½ to 31/32 by using dual fragmenters and up to 11 encoders for codewords above 13.5kb. The paper discusses low-level optimization techniques used to achieve 100Gbps, some specific to the AMD EPYC, such as locking memory to NUMA nodes, mapping network processing to CPU cores, use of DPDK Flow Director, and many others.

This paper presents the design and validation of an L/S dual-band DVB-S2 communication payload for low Earth orbit (LEO) CubeSats, developed to meet high-data-rate downlink requirements under strict size, weight, and power (SWaP) constraints. The link uses an L-band (870–900 MHz) uplink and an S-band (2150–2200 MHz) downlink with QPSK modulation, a code rate of 1/2, a frame length of 64,800 bits, a data rate of 1 Mbps in both directions, and a total power budget of 15 W. In terms of algorithm design, the receiver implements BCH decoding with the Berlekamp–Massey algorithm and a layered offset min-sum (OMS) LDPC decoder with offset β = 1 to reduce hardware complexity and accelerate convergence. Symbol timing recovery is realized as an all-digital closed loop combining a Gardner timing error detector, a polyphase interpolator, and a proportional-integral loop filter. Carrier synchronization adopts a two-stage coarse/fine architecture, in which a coarse frequency estimator based on multi-lag autocorrelation of the 90-symbol PLHEADER is followed by pilot-aided fine frequency tracking, in order to cope with the S-shaped Doppler profile with a total excursion of up to 112 kHz during a single pass. For hardware implementation, the system uses a Xilinx Zynq-7035 with an ADRV9361 RF module as the engineering model and targets the Zynq UltraScale+ XCZU4EV as the flight model. The BCH and LDPC codec IP cores are implemented using high-level synthesis (HLS) and deployed in the programmable logic (PL). Because the LDPC decoder alone occupies 78.91% of the BRAM on the XCZU4EV, the AGC, frequency rotator, symbol synchronizer, and physical-layer synchronizer are implemented in software on the ARM processing system (PS) in C++, a partitioning that balances resource limits against feedback-path latency. For testing and validation, hardware/software loopback tests show that the bit error rate (BER) falls below 10⁻⁶ at an Eb/N0 of approximately 6.6 dB, meeting the DVB-S2 quasi-error-free (QEF) requirement. A cross-band regenerative relay test between two SDR units sustains 1 Mbps throughput without frame drops and maintains constellation convergence under an injected 240 kHz carrier frequency offset. Sensitivity measurements in a conducted test setup yield a decoding threshold of Es/N0 = 7.68 dB. Finally, using the Doppler profile generated by a Systems Tool Kit (STK) simulation driven by the actual two-line element (TLE) set of the PARUS-6U1 CubeSat, the carrier tracking loop is shown to compensate for dynamic Doppler shifts of ±55.929 kHz.

This paper presents a measurement-level simulation framework for timing-induced navigation error in GPS-denied orbital and deep-space regimes. The framework maps oscillator stability, synchronization cadence, measurement geometry, environmental forcing, hybrid clock steering, clock-state filtering, and relativistic correction fidelity to range-equivalent and position-equivalent timing error, and separates the timing-limited, synchronization-limited, and model-limited regimes that govern architecture-level performance. Its central contribution is an explicit pseudorange-level validation in which simulated residual clock errors are injected into pseudorange measurements and propagated through nonlinear weighted least squares (WLS). To remove ambiguity about the estimated clock state, the implementation reports both a primary residual-clock position-only WLS mode and a diagnostic position-plus-clock-bias mode, with the dilution-of-precision (DOP) metric and the WLS solve sharing the same visible-source set. Across representative LEO and MEO cases, the 95th-percentile (p95) position-only WLS errors stay within the same order of magnitude as the architecture-level PDOP·c|Δt| predictions: 7.80 km predicted versus 4.69 km solved for the LEO OCXO 6-hour case, and 123.0 m versus 109.9 m for the MEO model-limited atomic case. The diagnostic clock-estimated WLS collapses the common-bias contribution toward the measurement-noise and geometry floor, confirming that the primary experiment intentionally tests unestimated residual timing error. The same framework evaluates multi-day deep-space holdover stress cases, in which weak oscillators accumulate mission-scale range error while atomic and projected optical classes remain at the kilometer level, together with a bounded quantum-aided relativistic correction feasibility envelope. Hybrid steering provides the strongest practical reduction in timing-induced degradation, while quantum-aided correction yields conditional benefit only after classical clock noise has been sufficiently suppressed in the model-limited regime. The central conclusion is that resilient GPS-denied timing performance is governed by the coupled interaction of clock physics, synchronization policy, measurement geometry, estimator structure, and model fidelity rather than by oscillator quality alone.

NASA's Artemis program and the broader LunaNet initiative are driving demand for lunar surface positioning, navigation, and timing (PNT) and communications relay, particularly at the lunar south pole, where users see Earth only a few degrees above the horizon. This geometric study evaluates compact seven-petal tulip-shaped orbit constellations, periodic circular restricted three-body orbits that concentrate sub-satellite passes over the south-polar cap, as candidates for combined PNT and direct-to-Earth relay. Two configurations are assessed: a six-satellite baseline and a minimum-cost five-satellite variant, both occupying the same shared three-body orbit and differing only in satellite count and along-track phasing. Performance is scored against three Initial Operating Capability C (IOC-C) geometry metrics drawn from the NASA LCRNS requirements: line-of-sight link availability, Lunar Augmented Navigation System geometric dilution of precision (GDOP < 6), and daily extravehicular activity (EVA) usable-PNT windows. All metrics are benchmarked against re-optimized elliptical lunar frozen orbit (ELFO) constellations over Service Volume II (latitudes south of −75°). Both tulip constellations satisfy all three IOC-C metrics across SV2. The six-satellite baseline passes with wide margin (75% worst-point daily GDOP < 6 availability; 18 h daily EVA support), and the five-satellite variant passes with thinner margin (44%; 10 h). The tulip constellations also shorten the worst-case gap between usable EVA windows to 5.9 h, versus 14.3 h and longer for the ELFO references, and roughly halve total inter-satellite occultation time, supporting more continuous crosslink ranging and time transfer. Distinct from the ELFO references, both tulip architectures maintain zero lunar occultation to Earth for every relay spacecraft, so any surface-to-satellite link can also serve as a direct single-hop Earth relay. Results indicate tulip-shaped orbits are credible candidates for compact, resilient south-polar PNT and relay architectures.

Disaggregated spacecraft constellations and satellite-connected aerospace edge networks need publicly verifiable randomness for coordination, scheduling, audit, and recovery. They also need evidence of ordering and elapsed work when global navigation satellite system (GNSS) signals are denied, spoofed, or unavailable. We present a GNSS-independent, hash-based verifiable randomness beacon for satellite-connected constrained operation. The design extends a quantum random number generator (QNRG)-seeded commit–reveal beacon previously deployed on a real FIT-IoT-LAB (FIT-IoT) testbed. It replaces its algebraic sharing and delay components with hash-based constructions, derives the delay challenge from a transcript frozen before reveal, and uses a constrained Nordic nRF9151 anchor to audit the resulting sequential-work trace. This removes the need for a shared clock while keeping the beacon record auditable from its transcript. We implement the protocol on 38 asynchronous nodes and evaluate committee sizes, reveal-withholding loads, communication cost, phase latency, and attestation independence. The evaluation includes both all-to-all and committee-based recovery, exposing the trade-off between communication overhead and recovery capacity. A single anchor attests the delay after the beacon output is derived, while the remaining nodes continue to reach the same output without attestation evidence of their own. Finally, we replay 50 beacon records through a metered satellite-internet of things (IoT) link as a non-terrestrial publication proof of concept. The result is an implemented beacon that combines verifiable randomness with independently auditable elapsed sequential work and can be integrated with constrained, intermittently connected systems.

This paper presents an orbit determination approach for distributed spacecraft systems integrating inter-satellite link (ISL) range measurements with star-tracker-aided line-of-sight bearing information. Future large-scale satellite constellations require accurate onboard orbit determination while minimizing dependence on ground-based tracking and continuous GNSS availability. This work investigates how ISL range measurements and star-tracker-aided line-of-sight bearing can be combined for local onboard orbit determination of a satellite in the constellation. The proposed framework integrates two-way ISL range measurements from four neighboring satellites with line-of-sight bearing measurements derived from star tracker's inertial attitude information. In the considered scenario, bearing measurements provide angular information that mitigates observability limitations of scalar ISL range measurements. ISL is assumed to have its own tracking function independently of satellite attitude and star tracker data so that star tracker provides an independent inertial attitude reference for the line-of-sight bearing. Therefore, added angular measurement model enables orientation transformation from line-of-sight direction of the ISL terminal to inertial reference frame. In the simulation study, true orbital states are generated using Orekit’s high fidelity numerical propagators. An extended Kalman filter (EKF) is implemented to estimate orbital parameters of the local satellite every 60 seconds assuming remote satellites’ orbital parameters are known as remote satellites broadcast their own orbital parameters. For orbit determination framework to be applicable onboard, estimator is designed using Keplerian propagator. ISL range measurements are modeled with calibrated bias and centimeter-level random noise. Angular measurement is modeled by calibrated bias and 5 arcsec random noise. The implemented EKF achieves decimeter-level positioning accuracy. Fisher information analysis with Schur complement shows constellation-wide common-mode shifts are observable by the proposed framework and measurement models. With ISL range and star-tracker-aided bearing, local satellite estimated its orbital parameters in absolute inertial reference frame. Whereas ISL range-only case provides decimeter-level relative orbit determination, common-mode states remain weakly observable. Adding star-tracker-aided bearing provides absolute reference frame recovery, with meter-level positioning accuracy

This paper presents a Ground Holding-Marking Stopping System for uncrewed aircraft, allowing them to recognize and obey runway holding-position markings and enhanced taxiway centerlines autonomously. During taxiing, an aircraft repeatedly approaches positions at which it must hold short of a runway; these are signaled only by painted markings on the surface, and crossing one without clearance is a runway incursion, among the most safety-critical failures of ground operation. An autonomous solution to the problem means perceiving the markings, deciding whether a stop is mandatory, and commanding a longitudinal speed controller accordingly. General-purpose perception systems do not transfer adequately to the airport surface, and no public imagery exists that contains annotations of these markings from the aircraft's point of view. The main contributions are twofold. First, a robust runway holding-position markings and enhanced taxiway centerlines recognition system that relies only on a forward-facing camera is developed. Since no suitable public dataset exists, the training imagery is generated synthetically in a photorealistic simulation environment and automatically labeled: the three-dimensional positions of the painted markings are projected into the camera under a calibrated model to produce exact bounding boxes at no annotation cost, and the data is partitioned by a geographic clustering that prevents any physical marking from appearing in both training and test sets. The recognition system then runs two neural networks in parallel. A YOLO detector, fine-tuned on this synthetic data, localizes the enhanced centerlines and holding markings with an mAP50-95 score of 0.675 on the custom synthetic test set, while a lightweight custom-made tracker links each marking across frames. In addition, a ResNet-50 classifier, specially trained for the task, resolves the solid-versus-dashed orientation of each holding marking with an accuracy of 0.996 on the same test set. The second contribution consists of a stop-handling module that converts the output of the recognition system into a speed reference command: it projects each marking to the ground to recover the remaining distance and gates the decision on the orientation verdict. Moreover, distant or fleeting look-alike lines are ignored, and the system decelerates the aircraft to a halt short of a mandatory marking and releases it once it may proceed. The Ground Holding-Marking Stopping System has been evaluated in a high-fidelity simulation environment, with aircraft dynamics represented by a MATLAB/Simulink multibody model validated against real flight data. The simulator is coupled with the X-Plane graphical engine for airport layout rendering. In closed loop, the system reliably brings the aircraft to a stop up to 0.15 meters short of every marking that mandated a halt and proceeds through those that did not, and it rejects every spurious look-alike detection raised by the dense apron surroundings without once triggering a false stop. The work contributes an annotation-free route to training airport-marking perception and a closed-loop validation that transfers to an unseen airport, advancing safe, camera-only taxiing through the autonomous handling of the markings that govern runway access.

As the low-altitude economy accelerates, integrating UAVs into non-segregated airspace demands more than ’best-effort’ wireless connectivity. Moreover, reliability and resilience within the digital communication domain is now an urgent priority due to the unmanned and autonomous nature of UAVs. The Command and Control (C2) communication system, an essential safety tether for Beyond Visual Line of Sight (BVLOS) operations requires a rigorous framework to meet strict aviation specifications, deterministically reliable and resilient. The existing Required Communication Performance (RCP) defines reliability by setting strict static bounds on parameters like Transaction Time, Continuity, Availability, and Integrity. Beyond the scope of security resilience, C2 system requires a dedicated focus on communication and connectivity: verifying whether the C2 link is active or impending degradation in C2 message traffic. Therefore, in this work, we address the gap from C2 reliability to system-level resilience by focusing only on communication and connectivity paradigm. To achieve this, we introduce a behavioral connectivity resilience model, in order to adapt system behavior to the operationally functional boundary. This is built upon two key contributions. First, we develop a phase-based C2 system architecture spanning proactive, reactive, and post-corrective operational windows to dynamically manage the RCP under potential C2 degradation condition. The first, Proactive Phase, anticipates and preemptively adjust transmission mechanisms on the C2 system. The second, Reactive Phase, actively mitigates real-time network delays to maintain vehicle control during active link stress, while the third, Post-Corrective Phase safely governs system resynchronization and state restoration. Second, we formulate a dynamic Operational Margin representation that envelopes these three phases specifically for Very Low Level (VLL) airspace. This time-variant behavioral framework is evaluated using a UAV equipped with a heterogeneous, multi-link communication suite (WiFi, 5G, and LEO Satcom) compliant with Specific Assurance and Integrity Level III (SAIL III) and above. By analyzing performance metrics across each operational window, we present the derived operational margins for the evaluated system, demonstrating that the UAV can safely navigate communication anomalies through adaptive, behavioral control of the C2 system.

Reliable tracking of unmanned aircraft (UA) is essential for surveillance and flight monitoring, as well as for future Unmanned Aircraft System (UAS) traffic management (UTM) operations. However, position reports obtained from different systems may vary in spatial accuracy, update rate, reporting delay, data availability, and altitude reference. This study presents a comparative flight evaluation of four position-reporting and logging approaches operated on a single UA: Automatic Dependent Surveillance–Broadcast (ADS-B), onboard Pixhawk logging with Real-Time Kinematic (RTK) positioning, onboard smartphone Global Navigation Satellite System (GNSS) logging, and a Remote ID system developed for UTM applications. Several initial flights have been completed, and additional tests will be conducted using repeatable waypoint-guided missions as well as manually controlled flights. The Pixhawk RTK solution is used as the reference trajectory when a fixed solution is available. Preliminary results from one of the higher-quality flight datasets show close agreement among the Pixhawk RTK, ADS-B, and smartphone trajectories during both long straight segments and repeated maneuvering patterns. Local differences become more noticeable during turns and dense maneuvering, indicating that part of the observed separation may be associated with temporal misalignment as well as differences in update rate. Post-flight log files will therefore be time-aligned to estimate relative reporting delay and to distinguish position differences caused by temporal misalignment from residual spatial errors. The comparison considers horizontal trajectory accuracy and update characteristics, together with reporting continuity, data availability, and trajectory reconstruction during straight flight, turns, climbs, and descents. Particular attention is given to altitude because the systems may report mean-sea-level altitude, barometric altitude, or altitude above ground level. The altitude data are therefore examined before and after conversion to a common vertical reference. In addition to characterizing the performance of each individual source, the study will examine the implications of inconsistent timing, position, and altitude information for surveillance-data integration. The results are expected to identify discrepancies that can be corrected through time alignment and reference conversion, as well as those that must be treated as source-specific uncertainty. The study will also examine how heterogeneous reports can be combined without introducing misleading trajectory or separation information.

Flying Ad-hoc Networks (FANETs) often include low-cost Unmanned Air Systems (UAS) engaged in coordinated team missions such as reconnaissance, search and rescue, disaster relief, etc. These UAS platforms are often inexpensive and low-power with communications carried over low-cost and limited-range tactical data links such as 5G SideLink cellular radios. This capability must support vehicle-to-everything (V2X) peer-to-peer communications when cellular base stations and/or satellites are unavailable, often necessitating "multi-hop" relaying to maintain FANET connectivity. FANETs therefore require a routing protocol to discover paths between communicating peers even if the paths include multiple relay hops beginning from a source then forwarded through any intermediate nodes and finally arriving at the target. Due to the highly dynamic nature of UAS mobility, these multi-hop forwarding paths often change dynamically over time as neighboring nodes move in and out of range in relation to one another. The routing protocol therefore needs to maintain the time-varying view of the FANET topology with the shortest possible paths used between all pairs of communicating peers. The study of multi-hop routing protocols for FANETs is not new; an IEEE Aerospace 2002 publication “Tactical Information Operations for Autonomous Teams of Unmanned Aerial Vehicles (UAVs)” introduced an early analysis of the TBRPF protocol running over actual UAS using IEEE 802.11 radios. Several more optimized multi-hop routing protocols in robust implementations have since become available. However, the current literature lacks an evaluation of the performance of emerging link capabilities (e.g., 5G NR sidelink and Layer 2 forwarding in Wi-Fi mesh) in conjunction with the recent routing developments (e.g., BABEL). Using low-cost and low-power tactical data links such as 5G cellular with SideLink, UAS neighbor-to-neighbor communications range is often limited from a few hundred meters to a few kilometers even when FANET vehicles are airborne and have an unobstructed line of sight. This means that FANETs spanning areas with larger diameters are obliged to engage an efficient and effective routing protocol running on each UAS which must provide multi-hop forwarding services as necessary. In this paper, we provide an analysis of candidate FANET routing protocols suitable for affordable tactical data links. The protocols include public domain standards originating from the traditional Mobile Ad-hoc Network (MANET) framework as well as proprietary services offered by commercial suppliers. We specifically examine the OSPF-MDR, OLSRv2, B.A.T.M.A.N and Babel proactive protocols with an analysis also of the AODV reactive protocol. An analysis of proprietary services based on available literature is also included. A critical aspect of our study includes a performance analysis from anticipated FANET UAS mission profiles using the CORE/EMANE network emulation services with the 5G SideLink tactical data link model created by our team. We also examine the tension between the classical MANET approach based on Layer 3 routing and forwarding and an emerging Layer 2 approach based on Wi-Fi mesh. We conclude our study with a recommendation for affordable tactical data link best practices for common FANET mission profiles.

Uncrewed aerial vehicle (UAV) swarms operating in GPS-degraded, infrastructure-limited, or contested environments require communication and navigation capabilities that remain available despite intermittent wireless links, node failures, and heterogeneous onboard resources. Conventional architectures that depend on a fixed ground station or centralized gateway introduce single points of failure and can limit the resilience of collaborative airborne missions. This paper presents SkyMesh, a decentralized airborne networking architecture that tightly integrates multi-hop communication, onboard sensing, dynamic gateway recovery, and cooperative localization to support resilient multi-UAV operations without reliance on a fixed communication coordinator. Each SkyMesh UAV operates simultaneously as a sensing platform, network relay, and cooperative localization node. The prototype integrates a Pixhawk flight controller for vehicle telemetry, Raspberry Pi edge computing for onboard visual sensing and object detection, and lightweight wireless mesh nodes for peer-to-peer information exchange. Detection events, telemetry, network state, and localization information are disseminated through multi-hop links using a compact packet structure designed for resource-constrained airborne platforms. To eliminate dependence on a single ground-connected node, SkyMesh supports dynamic gateway selection and alternate relay paths, allowing swarm information flow to continue when a gateway or communication link becomes unavailable. Shared visual observations, UAV pose estimates, and inter-UAV geometric constraints are additionally fused in a cooperative localization framework to reduce positioning errors associated with degraded or stale GPS information. We evaluate SkyMesh using a three-node UAV mesh testbed under line-of-sight and stressed two-hop relay configurations. Experimental measurements characterize packet delivery, throughput, outage duration, inter-arrival variability, and post-disruption recovery. The gateway node maintains greater than 99% packet delivery across the evaluated topologies, while relay experiments demonstrate recovery of connectivity for degraded edge nodes and distinguish communication impairments caused by link attenuation from limitations caused by onboard computation. Following communication disruptions, queued information is disseminated through controlled post-outage transmission, enabling the distributed system to re-establish synchronized operation without persistent degradation. To examine the relationship between airborne networking performance and cooperative navigation, measured mesh delay and outage statistics are incorporated into model-based localization studies. Across 600 Monte Carlo trials, cooperative localization reduces per-UAV position error relative to GPS-only estimation as swarm size increases. At three UAVs, full cooperation reduces modeled localization error from 0.90 m to 0.50 m; at ten UAVs, the corresponding error decreases from 0.42 m to 0.26 m and approaches the Cramér–Rao lower-bound trend. Hover-drift modeling further shows end-of-window error-growth rates of 0.09 m/s, 0.05 m/s, and 0.03 m/s for GPS-only, bearings-only, and fully cooperative estimation, respectively. These results demonstrate how communication resilience and cooperative navigation can be jointly designed within a lightweight airborne network, providing a practical foundation for distributed sensing and localization in UAV swarms operating under unreliable connectivity and degraded positioning conditions.

Air-ground trajectory information is increasingly important to trajectory-based operations and airborne-network services. The Automatic Dependent Surveillance-Contract (ADS-C) Extended Projected Profile (EPP) provides a predicted four-dimensional trajectory derived from the aircraft flight management system and can contain up to 128 waypoints. Successive EPP reports usually contain substantial unchanged information, making repeated full report transmission inefficient, when only a small part of the trajectory changes. While bandwidth-limited datalinks such as SATCOM and VDL Mode 2 can sometimes constrain individual message sizes to a few hundred bytes, full EPP transmissions routinely exceed these limits, forcing fragmentation, raising latency, and creating aircraft-ground synchronization risk. This paper presents an encoded-data differential transmission mechanism that preserves the existing EPP semantic model and Packed Encoding Rules (PER) encoding process. The aircraft generates the normal encoded report, compares it with the previous encoded reference, and transmits only changed contiguous regions using offset, length, and updated data fields, named as ADSDeltaChange. The ground system applies the delta to its synchronized reference, reconstructs the complete encoded report, and then performs conventional UPER decoding. A full-report path remains available for initial synchronization and recovery. Prototype measurements show substantial payload reduction. For a 128-waypoint report, the measured full encoded payload was 5,250 bytes and a one-waypoint update produced a 27-byte differential payload, giving 99.5% reduction and an approximately 194:1 size ratio. For a 32-waypoint report, the full encoded payload was 1,338 bytes; a one-waypoint update produced a 27-byte delta payload, giving 97.9% reduction, while a five-waypoint update produced a 132-byte delta payload gave 90.1% reduction. The results demonstrate that encoded-difference locality can substantially reduce repeated trajectory payloads. The work also discusses reference synchronization, integrity, failure recovery, adaptive full-versus-delta selection, and significant payload reduction. The proposed technique is a communication-efficiency layer for existing ADS-C EPP transmission over airborne network applications, not a replacement for the ADS-C EPP data model, UPER, or any conventional entropy compression.

With global air traffic becoming increasingly congested, packet collisions and co-channel interference frequently degrade the reliability of ADS-B messages. Current interference mitigation techniques rely on expensive hardware modifications, idealized simulations, or computationally expensive algorithms. This paper proposes a Successive Interference Cancellation (SIC) framework implemented on commercial off-the-shelf (COTS) hardware and validated using real-world, over-the-air (OTA) measurements. By leveraging an Ordinary Least-Squares (OLS) channel estimation, the proposed method successfully reconstructs and subtracts primary signals to recover overlapping secondary signals. Experimental results using an ADALM-PLUTO Software Defined Radio (SDR) demonstrate that the implemented SIC algorithm provides up to a 10-fold reduction in bit error rate (BER) compared to traditional decoding methods. These findings suggest that robust interference mitigation is achievable with little to no modifications to existing software-defined architectures, offering a scalable path for improving the reliability of wireless identification within crowded airspaces.

The implementation of redundant safety layers for the coordination of Unmanned Aerial Vehicles (UAVs) is crucial to prevent collisions within the highly crowded urban airspaces of the future. Using an additional communication system called DroneCAST to broadcast Detect-and-Avoid (DaA) beacons, a concept well established for instance as squitters in ADS-B, can serve as an essential building block for UAV collision avoidance. Because the communication system acts as a last layer of defense, spoofing resistance of the information within the transmitted DaA beacons is crucial to prevent attackers from misusing the system to cause false collision alarms, suppress true alarms or urge any UAV to change their flight path. Blindly accepting any received DaA would expose the system to severe adversarial attacks. Although authenticating the DaA beacons is an important element to distinguish verified sources from unverified sources, it is insufficient as the only measure of protection. Blindly ignoring (not yet) fully authenticated DaA beacons, however, could be fatal as well. Consequently, to increase the robustness of the proposed DaA approach against spoofing attacks, verifying information contained in the received DaA beacons to the largest possible extend is crucial before deriving actions upon the data. This paper makes two main contributions. First, we further expanded the current specification of DroneCAST not only to incorporate the TESLA broadcast authentication protocol for authenticating DaA beacons, but designed an out-of-band process to retrieve and verify the signed anchor of the used hash chain. Second, we utilized Received Signal Strength (RSS) to verify the credibility of kinematic information within these beacons, effectively detecting spoofing attacks where the movement of an UAV according to the data in the DaA beacons does not match the physics of the communication channel. To ensure robust and accurate detection, we modeled signal degradation across rural, suburban, and urban environments, accounting for variations with UAV flight altitude. Incorporating a proper model is essential to distinguish between non-line-of-sight propagation in urban canyons and line-of-sight conditions, e.g. above rooftops. Because isolated RSS measurements are vulnerable to distance spoofing, where adversaries manipulate signal strength, we designed a cooperative mechanism capable of detecting such manipulation through collaboration. To address the inherent vulnerabilities of cooperative networks, our design explicitly accounts for and mitigates insider attacks. A defining characteristic of the mechanism is its reliance on continuous tracking of signal strength and kinematic data over time, rather than assessing isolated data points. That way, the system can verify the reported information based on the consistency of the overall trajectory. Finally, we evaluate our approach against defined threat models across diverse environments, demonstrating robust performance under varying signal degradation conditions.

The emergence of cryptographically relevant quantum computing (CRQC) threatens the public-key cryptography that currently protects terrestrial and satellite-based communication systems. In particular, “harvest now, decrypt later” (HNDL) attacks create an immediate need to transition long-lived and mission-critical communications to quantum-resistant security before large-scale quantum computers become available. This challenge is especially important for the evolution of 5G toward 6G, where terrestrial networks (TN) are increasingly integrated with non-terrestrial networks (NTN) to provide resilient, ubiquitous connectivity for aerospace, defense, emergency response, and autonomous cyber-physical systems. While the U.S. National Institute of Standards and Technology (NIST) has standardized post-quantum cryptographic (PQC) algorithms, their substantially larger public keys, ciphertexts, and digital signatures introduce communication overheads whose impact on wireless network performance has not been fully characterized. This paper investigates the cross-layer performance implications of PQC in current 5G and emerging 6G cellular systems, with emphasis on latency-sensitive and mission-critical applications. Unlike previous studies that primarily evaluate computational overhead or end-to-end protocol performance, this work develops a cross-layer analytical model that traces PQC message expansion through transport-layer packetization, Packet Data Convergence Protocol (PDCP) processing, Radio Link Control (RLC) segmentation, Medium Access Control (MAC) transport block allocation, and radio scheduling. The analysis quantifies how larger cryptographic objects increase the number of transport packets, RLC segments, MAC transport blocks, and scheduled radio slots required to complete security exchanges over the air interface. The study further examines how these effects are amplified in NTN deployments, where satellite propagation delay, limited radio resources, link-budget constraints, and constrained scheduling opportunities place a premium on efficient use of the radio interface. In this setting, additional transport blocks not only increase serialization and scheduling delay; but also occupy time-frequency resources that could otherwise support user data or competing control traffic. The resulting latency, reliability, and radio-resource impacts are evaluated for representative NIST PQC security profiles and representative 5G/6G numerologies and transport block sizes. Potential mitigation strategies, including hybrid cryptographic deployment and adaptive security profile selection, are also discussed. The results demonstrate that successful deployment of quantum-safe security in future cellular systems requires joint consideration of cryptographic algorithms, transport protocols, and radio resource management. By establishing a quantitative relationship between PQC message expansion, wireless access latency, and air-interface resource occupancy, this work provides system designers with a framework for evaluating the performance tradeoffs associated with quantum-safe security in integrated terrestrial and non-terrestrial 5G/6G networks supporting aerospace and other mission-critical applications.

Satellites and space infrastructure are increasingly targeted by sophisticated cyber threats, including firmware tampering, command injection, signal replay, and supply-chain compromise. Yet the security architectures of most operational satellites — particularly small satellites and CubeSats — rely on perimeter-based controls that provide little assurance once physical or link-layer access is obtained. This paper argues that the Trusted Computing Group (TCG) specification suite — specifically TPM~2.0 (ISO/IEC~11889) and the DICE architecture — is.today the most mature, internationally standardised framework available for simultaneously addressing trust establishment and cryptographic secret protection in embedded systems. Crucially, TPM~2.0 is not a single chip form factor but a Functional Interface Specification: its capabilities can be realised as a discrete hardware component, a firmware implementation in a trusted execution environment, or a soft-core on a radiation-tolerant FPGA — the choice of physical implementation is precisely the mechanism by which the TPM adapts to the unique environmental demands of space. We present a structured STRIDE-based threat model for LEO satellites spanning three trust zones, map TCG mechanisms (measured boot, remote attestation, sealed storage, and DICE-based subsystem identity) to the identified threats, and propose a concrete 3U CubeSat reference architecture covering the boot chain, attestation workflow, key management, and ground-segment trust backend. We further examine the implementation consequences for space deployment, including radiation tolerance, thermal resilience, long operational lifetimes, key lifecycle management, and mass and power budgets. Regulatory alignment with the EU Cyber Resilience Act, NIS~2 Directive, and the proposed EU Space Act is analysed. Our analysis demonstrates that TCG technology provides a mature, standards-based path to verifiable, hardware-anchored security for the new space era, and we outline a companion implementation and experimental evaluation as the next stage of this research.

Cloud-enabled spacecraft ground systems rely on workload identities and automation agents whose authority can outlive the activities that justified it. Short credential lifetimes do not remove excessive delegation paths or define bounded authority during identity-service loss. Mission-phase identity containment (MPIC) joins phase-constrained authority graphs, attenuating capability leases, finite offline spending envelopes, and a recovery barrier. It distinguishes privilege paths from recovery dependencies and bounds each admission by scope, phase, policy context, workload evidence, time, and path length. A disconnected verifier cannot detect an unseen revocation immediately; signed deadlines bound that residual exposure. Evaluation combines a local cryptographic reference monitor with fully specified graph and event-order studies. In a 23-node, 22-edge example, aggregate exposure decreases from 21,600 operation-seconds with renewable phase-agnostic authority to 7,320 with phase restrictions and 6,360 with a four-edge bound. All 28 declared operational obligations remain graph-feasible. The monitor matches its specified guard decisions in 2,048 cases; an independent path enumerator agrees in 192 comparisons. Enumeration of 720 recovery orders exposes premature authorization without the recovery barrier. Continuity and depth sweeps identify the corresponding availability costs. These results characterize the declared model and local implementation. NASA's Operational Simulator for Space Systems supplies an integration reference; a full simulator deployment and observed command-delivery evaluation remain outstanding. MPIC contributes an inspectable operational authorization contract supported by bounded analytical and executable evidence.

The EU Cyber Resilience Act (CRA), binding from December 2027, requires manufacturers of products with digital elements to provide a Software Bill of Materials (SBOM) documenting top-level dependencies. Maintenance, Repair, and Overhaul (MRO) organizations in civil aviation increasingly depend on complex, third-party- and open-source-dependent ground software — asset-planning systems, airworthiness-tracking platforms, digital documentation tools — yet the CRA, EASA Part-IS, and the NIS2 Directive together address the manufacturer side almost exclusively. No regulation currently obliges an MRO operator, as the receiving party, to ingest, validate, or operationally act on a manufacturer-supplied SBOM. This asymmetry leaves open whether SBOMs deliver any practical security benefit once handed to an operator that lacks a defined receiving process. This paper addresses that operator-perspective gap, which existing SBOM literature — overwhelmingly focused on manufacturer-side generation, format, and quality — does not cover. We first conduct a structured gap analysis contrasting the regulatory target state (CRA, EASA Part-IS, NIS2) with current MRO operational practice, identifying six structural gaps: SBOM receipt and processing, continuous vulnerability monitoring, software lifecycle and patch governance, VEX processing capability, procurement policy, and cybersecurity competency. We validate these findings through a proof-of-concept: a prototypical MRO training-platform application (React/Node/Python, containerized) subjected to SBOM generation at three levels — application manifest, resolved runtime environment, and container image — using npm sbom, cyclonedx-py, and Syft, cross-referenced against vulnerability data via npm audit, pip-audit, and Grype. Results show a substantial, quantifiable discrepancy between direct and transitive dependencies: two direct backend dependencies expand to 70 components at the manifest level and 3,628 at the container-image level, demonstrating that an SBOM's informativeness is meaningless without specifying its generation layer. We further show that SBOM data alone yields no risk assessment; only combined with vulnerability tooling does it produce actionable findings, and naive tool output produces both false alarms (unreachable dev-only dependencies) and false negatives (unpatched production-path findings) without explicit build/runtime separation. A worked VEX/CSAF case study demonstrates that a component-level "known_affected" finding can be correctly reclassified as product-level "not_affected" using three independent, schema-validated arguments, confirming VEX's role in suppressing the false positives that would otherwise overwhelm operator-side monitoring. We conclude that SBOMs hold substantial but conditional value for securing critical MRO software: they establish a necessary, but not sufficient, foundation for vulnerability management, whose security benefit remains theoretical without accompanying operator-side processes, tooling, and competency. We argue this exposes a structural regulatory gap — the CRA mandates SBOM creation but not operator-side use — and propose both an extension of the regulatory framework and industry-level (IATA/ATA) standardization of MRO-specific SBOM requirements as remedies.

Heterogeneous unmanned aerial and ground swarms that route traffic over batman-adv expose their originator addresses – and, by extension, their tactical composition – to any passive observer with line of sight to the RF link. We propose a three-layer security architecture that separately addresses admission-time trust (TPM 2.0-backed enrollment and attestation, giving each joining node a hardware-rooted guarantee that its platform and software state are genuine and unmodified before it is granted network access), runtime health of already-admitted nodes (Collective Remote Attestation, which continually re-verifies platform integrity so that a node compromised after admission is detected and can be revoked rather than trusted indefinitely), and long-term identity exposure (TPM 2.0 ECDAA-based pseudonymous, rotating originator addresses, which let a node prove on every rotation that it remains a legitimate, unrevoked swarm member while denying a passive observer any way to link its old and new addresses – defeating direction-finding and traffic-analysis attacks on the swarm’s composition without requiring a trusted third party to issue new credentials). We describe a reference implementation on batman-adv topologies and report baseline routing-convergence measurements as a control dataset for the security-layer evaluation to follow.

Machine learning-based remaining useful life (RUL) prediction models can be used to support aircraft engine predictive maintenance, an integral process for aircraft maintenance professionals to ensure turbofans are operating safely to support maintenance planning and assess potential turbofan degradation. However, machine-learning-based RUL models may be vulnerable to False-Data Injection Attacks (FDIAs) from potential adversaries that manipulate sensor data used for RUL predictions, which is a threat that aircraft security professionals and maintenance teams shouldn't ignore. This study examines the sensitivity of individual sensor inputs to simulated FDIAs on an LSTM predictive maintenance model to help maintenance professionals get a better understanding of turbofan vulnerability and to create a stronger posture for turbofan telemetry security. Five LSTM models were trained on the NASA C-MAPSS FD001 dataset using different random seeds and were evaluated using root mean squared error (RMSE), mean absolute error (MAE), NASA score, and Pearson correlation. SHapley Additive exPlanations (SHAP) were then aggregated across all five models to identify which sensors contributed most strongly to the models’ predictions. During experimental testing, positive and negative perturbations were independently applied to selected sensor inputs. Perturbation magnitudes were assigned based on each sensor's training data standard deviation to account for differences in sensor variability. Attack effects were evaluated using RUL prediction displacement across sensor identity, perturbation direction, perturbation magnitude, and true-RUL state. RUL prediction displacement is the total difference between original predicted RUL and the RUL predicted from manipulated sensor input. Prediction diagnostics were then conducted on selected sensors at ±2 standard deviations to understand prediction behavior that may be hidden. Selected sensors were determined based on a variety of factors, including SHAP importance and sensor function. This diagnostic showed that increased perturbation magnitude generally increased RUL prediction displacement, and different behaviors were observed across individual sensor diagnostics. Perturbation direction analysis measured whether RUL prediction displacement was shifted upward or downward for each selected sensor and responses were compared across low-, medium- and high-RUL groups to see whether attack direction or true-RUL state had an impact on prediction displacement. Among the selected sensors, Sensor 9 produced the largest overall prediction displacement and showed distinct regional and directional response patterns. These findings indicate that while certain sensors showed distinct vulnerability over others, other state-specific elements should be considered. Perturbation direction and magnitude, true RUL state, sensor identity and variability are all important factors that affect overall sensor vulnerability. This study evaluates sensor manipulation at the model input level and does not claim that a physical engine's RUL was altered or an actual telemetry pipeline was manipulated to conduct these experiments. The results from this experiment can provide insight into which specific sensors and true-RUL states could be most vulnerable and influential in a model's predictive maintenance within a cyberattack to protect machine-learning-reliant predictive maintenance systems of tomorrow.

Space systems increasingly operate as interconnected cyber-physical infrastructures spanning spacecraft, ground stations, communication links, user services, and external supply-chain dependencies. This distributed architecture creates attack surfaces whose compromise may propagate from cyber effects to mission-level consequences. Although cybersecurity research in the space domain has expanded rapidly, the relationships among reported threats, adversarial tactics, techniques and procedures (TTPs), cyber threat intelligence (CTI), defensive mechanisms, cyber resilience, and attribution remain fragmented across the literature. This study presents a systematic evidence synthesis aimed at mapping these relationships across the principal segments of space systems. A systematic literature review was conducted using Scopus, Web of Science Core Collection, and IEEE Xplore, covering publications from 2000 through September 2026, supplemented by citation-based searching. Following deduplication, screening, eligibility assessment, structured extraction, and quality evaluation, 54 studies constituted the scientific corpus. Institutional frameworks and technical guidance were analyzed separately to contextualize operational practices without merging them with the scientific evidence base. The analysis maps adversarial activity through a common analytical chain linking threat actors and TTPs to attack surfaces, affected space-system segments, cyber effects, mission effects, detection and CTI mechanisms, attribution evidence, mitigation, and resilience. Spoofing and falsification, denial-of-service and distributed denial-of-service, jamming and radio-frequency interference, and command-and-control manipulation emerged among the most recurrent threat classes. The SPARTA framework was the predominant space-specific representation of adversarial behavior, while concepts derived from ATT&CK, structured threat representations, and knowledge-based approaches appeared with varying levels of operational integration. CTI-related concepts were present in nearly the entire corpus, but their implementation maturity varied considerably. The results reveal an important asymmetry between threat characterization and operational cybersecurity capabilities. Defensive and resilience mechanisms are comparatively well represented, whereas evidence-based cyber attribution remains substantially less developed. Persistent gaps include limited interoperability among threat representations, insufficient operationalization of CTI, scarcity of realistic datasets and forensic evidence, spacecraft resource constraints, and limited validation under representative mission conditions. The study contributes a segment-oriented threat-to-defense mapping intended to support threat-informed engineering of space systems. By explicitly connecting adversarial behavior, observables, defensive capabilities, and mission consequences, the resulting synthesis provides a structured basis for cybersecurity analysis, resilience engineering, CTI integration, and future validation of cyber defense mechanisms for space missions.

Global Navigation Satellite System (GNSS) spoofing can corrupt the navigation estimates that guide an autonomous unmanned aerial vehicle (UAV). Weak, persistent manipulation is particularly challenging when an alarm depends on a single current score. This paper investigates kinematic memory: combining a GNSS–inertial measurement unit (IMU) consistency feature with a leaky cumulative-sum (CUSUM) detector. We examine a PX4-Gazebo software-in-the-loop corpus containing 40 flights, equally divided among nominal, step, ramp, and low-acceleration stealth conditions. The implemented feature is a smoothed scalar difference between horizontal GNSS-derived acceleration magnitude and body-axis IMU acceleration magnitude. It is a physics-motivated proxy, not a fully attitude- and gravity-compensated vector residual. The comparison uses this same feature for Isolation Forest, a onesided standardized threshold labeled Mahalanobis in the code, and leaky CUSUM. We specify the implemented preprocessing, parameters, alarm latching, and evaluation conventions, and derive the accumulation rule’s finite sensitivity limit. Under the archived protocol, CUSUM obtains mean F1 scores of 0.823, 0.739, and 0.538 for step, ramp, and stealth, respectively, compared with 0.676, 0.574, and 0.000 for the standardized threshold. Neither detector produces a recorded false-alert flight. Isolation Forest achieves higher F1 scores, but alarms in all ten nominal flights and before the evaluation label boundary in 24 of 30 attacked flights. Its reported mean delays are consequently based on only one, two, and three eligible flights. Attacks begin 25 seconds after mission takeoff, as confirmed by the authors. Archived metrics have not been recomputed against this onset and do not establish takeoff-referenced latencies. The results support a limited benefit of leaky accumulation over the configured standardized threshold, not universal superiority over machine learning. The contribution is an implementation-grounded assessment of sensor consistency and temporal evidence, with explicit limits on physical interpretation, timing, and generalization.

Technical staff at the Los Alamos National Laboratory have developed a distributed space system architecture to enable rapid technology development. This system is adaptable to any host vehicle, and we present two flight missions that will implement this architecture using two ends of host vehicle class; (1) a 12U cubesat and (2) the International Space Station (ISS). The Experiment for Space Radiation Analysis (ESRA) is the latest of a series of Demonstration and Validation (DemVal) missions built by the Los Alamos National Laboratory, with the focus on testing a new generation of plasma and energetic particle sensors along with critical subsystems. The primary motivation for the ESRA payloads is to minimize size, weight, power, and cost while still providing necessary mission data. These new instruments will be demonstrated by ESRA through ground-based testing and on-orbit operations to increase their technology readiness level such that they can support the evolution of technology and mission objectives. This project will leverage a commercial off-the-shelf CubeSat avionics bus and commercial satellite ground networks to reduce the cost and timeline associated with traditional DemVal missions. The system will launch as a ride share with the DoD Space Test Program to be inserted in Geosynchronous Transfer Orbit (GTO) and allow observations of the Earth’s radiation belts. The ESRA CubeSat consists of two science payloads and several subsystems: the Wide field-of-view Plasma Spectrometer, the Energetic Charged Particle telescope, high voltage power supply, payload processor, flight software architecture, and distributed processor module. The ESRA CubeSat will provide measurements of the plasma and energetic charged particle populations in the GTO environment for ions ranging from ~100 eV to ~1000 MeV and electrons with energy ranging from 100 keV to 20 MeV. ESRA will utilize a commercial 12U bus and demonstrate a low-cost, rapidly deployable spaceflight platform with sufficient SWAP to enable efficient measurements of the charged particle populations in the dynamic radiation belts. The ESRA mission CDR was held on 15-Aug-23 and the team is currently engaged in payload-to-space vehicle integration and environmental testing in early Fall 2026. We will present calibration testing results for the WPS sensor conducted at the LANL Intelligence and Space Research Ion Beam Laboratory, and electron beam testing (10-18 MeV) of the ECP payload at Brookhaven National Laboratory. Final results of the fully integrated ESRA space vehicle will be shown as the mission hardware is stored until launch in August 2027.

Prompt optical observations of gamma-ray bursts (GRBs) and other rapidly evolving transients are valuable for time-domain and multi-messenger astronomy, but hard to capture. Events are rare, unpredictable, and fade quickly, and ground-based follow-up is constrained by weather, daylight, and geographic coverage. A low-cost autonomous spacecraft could act as an optical first responder, but a mission built around this goal faces a fundamental verification challenge: a natural transient may not occur within the spacecraft's field of regard during its operational lifetime, and a resulting non-detection cannot distinguish a true astrophysical absence from a failure of the spacecraft's pointing, imaging, or processing chain. This paper presents VECTOR (Versatile Educational Controls Testbed for Optical Response), a CubeSat-scale mission concept that resolves this challenge through catalog-omission verification. Known stars are intentionally omitted from onboard reference maps before launch, so that when the spacecraft images the corresponding region of the sky in orbit, the omitted star is detected as a new point source relative to its own prior knowledge, while ground controllers retain full knowledge of the star's true position and brightness. Since the source is real, this approach validates the complete flight pipeline, including pointing, imaging, onboard processing, and image comparison, rather than a single software component in isolation. We describe VECTOR's mission objectives and success criteria, a staged minimum- and full-success experiment plan, a preliminary CubeSat-scale system architecture built on a commercial lens optical payload, and first-order mass, power, data storage, and communications budgets that close with positive margin. VECTOR is intended to reduce the risk of future low-cost autonomous optical first-response spacecraft. CubeSats present an opportunity to act as agile, low-cost “first responders” in multi-messenger astronomy. They can rapidly search localization regions, detect and track early afterglows, and cue higher-resolution telescopes. To play this role, they must maintain accurate attitude control to autonomously identify faint transient point sources within a small field of view. We developed VECTOR (Versatile Educational Controls Testbed for Optical Response) as a 6U CubeSat designed to demonstrate an optical telescope with onboard processing to perform search algorithms. VECTOR provides a platform for testing pointing performance, image sensitivity, and detection algorithms under realistic conditions. The mission concept was developed by a student team in Summer of 2025, funded by the Air Force Research Laboratory University Nanosatellite Program Mission Concept course. This work outlines the VECTOR mission design, its role within rapid transient follow-up architectures, and its potential to expand global GRB response capacity through distributed CubeSat networks.

This paper addresses the spacecraft-side problem of maintaining centimeter-accurate inter-satellite spacing for very high-element-count picosatellite "swarm antennas" under fault-resilient operation. Broadband non-terrestrial network (NTN) and direct-to-device (D2D) services motivate ultra-high-gain apertures beyond what practical deployable reflectors can achieve. We study a swarm antenna in which 1,000 to 100,000 picosatellites, with analysis focused on the 10,000-plus regime, form a dense, quasi-planar lattice and coherently synthesize a large multi-beam aperture for handset-class users in the 0.7-2.0 GHz service band. The representative design point is near 1.8 GHz, where half-wavelength spacing is 8.33 cm for roughly 4.5 cm-class satellites. Centimeter-level relative positioning and attitude alignment are therefore required to limit phase error, control grating-lobe levels, and preserve link-budget margins. To our knowledge, no prior on-orbit system has maintained such dense spacing at this element count under realistic power and fault-response constraints. We integrate orbit design, electromagnetic actuation, scalable distributed control, and power-aware fault handling into one formation-keeping approach. Bounded relative motion is designed using a generalized circular orbit (GCO) that maintains a quasi-planar disk geometry with minimal propellant, while propellant-free electromagnetic formation flying (EMFF) provides residual-drift rejection and fine spacing adjustment. We size an air-core electromagnet compatible with a 50 mm-class picosatellite and show that the required nano-newton-scale interaction is achievable at practical coil currents, with sub-milliwatt to milliwatt electrical power per satellite at representative spacing. To make swarms of this scale tractable, a locality-aware, layered control and scheduling architecture (L0-L4) exploits the inverse-fourth-power distance decay of magnetic interaction and the slow time scale of orbital disturbances: each satellite interacts primarily with nearest neighbors, control is formulated over local groups, and time-sliced actuation energizes only non-interfering neighbor pairs. Fault-resilient operation is also examined for EMFF coil or controller anomalies. Stuck-off satellites drift away safely, whereas a stuck-on failure is treated as the critical case: abnormal motion is detected, neighboring satellites magnetically eject the faulty unit, and the local lattice is rearranged to close the gap. The resulting tradeoff among satellite mass, EMFF interaction range, brief communication suspension near the failed unit, and power margin indicates that low-mass, approximately 0.5U-class satellites can preserve useful fault-response margin while supporting several tens of simultaneous D2D beams. We validate the lower control layers on a planar water-surface testbed of ten floating mock picosatellites with overhead vision-based state estimation and wireless broadcast of coil-current commands. Starting from a deliberately perturbed configuration, the formation converged to the target honeycomb lattice and maintained it under disturbances, reaching approximately one-centimeter steady-state spacing accuracy. Although the current testbed is two-dimensional and non-orbital, it exercises key lower-layer functions of a distributed small-spacecraft control system; ongoing work extends to three-dimensional testbeds and end-to-end orbit/RF co-simulation with flight-like relative sensing. The contribution is an integration-ready path from orbit strategy and fault-resilient power budgeting through scalable distributed control to EMFF actuation for centimeter-accurate PicoSat swarm antennas.

This research aims to build a fast and reliable method for estimating electromagnetic force and torque between two small satellites flying at an ultra-close distance. Electromagnetic actuation is attractive because it can give force and torque without physical contact, propellant, or mechanical wear. It can also change the direction and size of the action by changing the electric current. These properties make it useful for satellite formation flying, remote control of small robots, and other systems where safe non-contact control is needed. The main problem is that electromagnetic interaction becomes very difficult to predict when two coils are placed only a few centimeters apart. In simple far-field models, a magnetic dipole approximation is often used. However, the force changes very rapidly as the distance becomes small, and the real shape, size, and attitude of the coils strongly affect the result. If the force and torque are estimated wrongly, the relative position or attitude of the satellites may drift. In a close-range situation, this can lead to collision or failure to follow the target motion. Therefore, the project focuses on near-field electromagnetic force and torque estimation that is both accurate and fast enough for control. The target system is a pair of small satellites, each equipped with cylindrical solenoid coils. The coil design, such as radius, length, number of turns, wire section, and winding layout, is fixed in advance. The inputs to the model are the relative position, relative attitude, and current conditions of the coils. The outputs are three-axis force and three-axis torque. The goal is to estimate these six values quickly under many possible close-range conditions. To prepare training and validation data, the project will combine commercial electromagnetic analysis software, such as Ansys Maxwell, with simulation software being developed by the research group. The commercial software will provide reference solutions for representative cases. The in-house software will include methods such as the Biot-Savart law, FEM, BEM, and FMM, and will be run on HPC systems to generate many cases systematically. The dataset will cover changes in position, attitude, and current. More data will be generated in difficult areas, especially where force and torque change sharply or where attitude has a large effect. The model is developed as Physics-Informed Neural Networks, or PINNs, not as a simple black-box regression model. Under a magneto-quasi-static approximation, residuals related to Maxwell’s equations will be included in the loss function. This design is intended to make the learned model consistent with both simulation data and physical laws. It should reduce physically unreasonable predictions, especially for untrained conditions and strongly nonlinear near-field regions. Implementing this model is more complex than ordinary supervised learning, because automatic differentiation, physics-based losses, boundary conditions, and training loops must be carefully designed.

Low Earth Orbit (LEO) Earth Observation (EO) satellite constellations are generating data at an unprecedented scale, creating downlink deficits where only a small fraction of collected data may be transmitted to the ground. Federated Learning (FL) promises to address this through collaborative on-board training across the constellation, reducing reliance on raw data downlink. However, existing FL-in-space methods remain dependent on ground stations and other dedicated external infrastructure for model aggregation, limiting scalability and autonomy. We remove this dependency by introducing AutoFLSat, the first hierarchical FL-in-space algorithm to achieve fully autonomous operation by moving parameter aggregation onto the constellation rather than on external infrastructure. AutoFLSat exploits the naturally two-tiered communication structure of satellite constellations, performing intra-plane aggregation via ring all-reduce over persistent intra-satellite links, followed by inter-plane aggregation via a bidirectional waterfall all-reduce across neighboring planes. This reduces inter-plane communication complexity from O(P^2) to O(P). Evaluated using FLySTacK, our open, hardware-aware in-orbit FL simulation platform, AutoFLSat achieves over 80% accuracy on all tasks including satellite imagery datasets like EuroSAT for Walker-Star constellations of up to 40 satellites. AutoFLSat also reduces wall-clock convergence time by up to 78% compared to ground-station-dependent aggregation and 41.7--77.8% compared to leading FL-in-space alternatives.

Distributed Space Systems (DSS) are expected to overcome the limitations of conventional monolithic space systems. Potential applications of DSS include distributed aperture and RF-distributed phased array antennas. Satellites in formation are required to maintain a separation of less than 1.0 meter, especially for an RF distributed phased array antenna. In such ultra-close separation, the Guidance Navigation and Control (GNC) subsystem is the key technology for the safe operation. Electro-Magnetic Formation Flight (EMFF) has been studied for this purpose. This EMFF utilizes electro-magnetic force working between coils on satellites in formation to control the relative position and attitude. While the formation control methods have been extensively studied, relative position estimation for multiple satellites remains an open issue. Although the optical sensor-based approaches are effective for relative position estimation, the target satellite must remain within the sensor’s field of view. Moreover, multiple optical sensors are required for estimating the relative positions of multiple satellites in formation. Therefore, the optical sensor-based approaches impose constraints on the relative attitude between the satellites. In addition, Carrier Differential GPS (CDGPS) enables the determination of the relative position without such constraints on the relative attitude. However, the relative positioning accuracy of CDGPS degrades if the GPS receivers on the satellites in formation cannot track a sufficient number of the same GNSS satellites. Consequently, this situation leads to degraded formation control performance. As an alternative approach, a relative position estimation using AC magnetic fields is proposed to complement the CDGPS. This proposed method comprises primarily of a Band-Pass Filter (BPF), an Initial Position Estimator (IPE), and an unscented kalman filter (UKF). A coil on satellites in formation generates an AC magnetic field at a specific frequency, and satellites flying nearby measure the magnetic field. The magnetic field generated by a specific satellite can be filtered out from the measured data by the BPF to estimate the relative position vector. UKF estimates the relative position vector sequentially using initial estimates provided by IPE and filtered magnetic field data for the observation update. This proposed method enables the estimation of the relative position vector of a specific satellite among multiple satellites, provided that the BPF successfully outputs the target AC magnetic field. Experimental results are presented to evaluate the validity of the proposed method. In this experiment, a satellite identifies a target among multiple coils under a cold start scenario with no prior knowledge of the relative position. Subsequently, the proposed method executes the UKF to estimate the relative position vector. The limitations and estimation accuracy of the proposed method are discussed based on the experimental results.

Accurate thermospheric density observations are critical for predicting atmospheric drag on low Earth orbit (LEO) satellites, directly affecting orbit prediction, collision avoidance, reentry forecasting, and space weather operations. Existing thermospheric density retrieval methods based on satellite drag primarily rely on reduced-dynamic precise orbit determination (POD) using Global Navigation Satellite System (GNSS) observations, either by estimating stochastic non-gravitational accelerations during orbit determination or by deriving drag accelerations through numerical differentiation of reduced-dynamic orbit products. Although these approaches have demonstrated promising performance, they generally retrieve density along individual satellite trajectories and provide limited direct information on the spatial structure of the thermosphere. Furthermore, the dynamic constraints, empirical acceleration parameterization, and smoothing commonly used in reduced-dynamic POD can attenuate short-timescale drag signatures, thereby limiting the temporal resolution of retrieved density variations. This paper assesses whether relative GNSS constraints can improve the observability of drag-induced orbital perturbations for thermospheric density retrieval. Rather than treating each spacecraft independently, relative carrier-phase observations are incorporated into a network-constrained kinematic orbit determination framework in which the motions of multiple GNSS-equipped LEO satellites are estimated jointly. By exploiting common-view GNSS measurements and relative carrier-phase ambiguity resolution, the network adjustment is expected to improve relative orbit precision and enhance the recovery of drag-induced accelerations through numerical differentiation. The assessment examines whether these improvements translate into measurable benefits for thermospheric density retrieval compared with conventional independent kinematic orbit determination. A validated GNSS-based thermospheric density retrieval framework has been established by reproducing published results using conventional orbit determination, providing the reference for the present assessment. GRACE Follow-On is employed as a proof-of-concept mission representing the simplest two-satellite network. Relative GNSS constraints are being incorporated into the network-constrained orbit estimation framework, and the resulting orbit solutions are being evaluated against independent kinematic solutions and calibrated onboard accelerometer observations. Complementary simulations are being conducted to investigate the effects of satellite separation, observation arc length, measurement noise, ambiguity resolution performance, and spacecraft drag modeling on the recoverability of drag-induced accelerations and thermospheric density variations. The study will provide a quantitative assessment of the utility of relative GNSS constraints for thermospheric density retrieval and identify the orbital configurations and measurement conditions under which they offer meaningful advantages over conventional independent orbit determination. The results are expected to provide guidance for the design and implementation of future network-constrained thermospheric density sensing using GNSS-equipped LEO satellite constellations.

Satellite constellations in Low-Earth Orbit (LEO) have emerged as an exciting new application area for Federated Learning (FL). While research in this domain has primarily focused on algorithmic efficiency and simulation, the design considerations for deploying FL on resource-constrained small satellite hardware remain underexplored. We present FLyCube (Federated Learning Ready CubeSat), a low-cost prototype platform to bridge this gap. This system integrates open-source and commercial off-the-shelf components, with custom flight software to facilitate decentralized on-board learning. In a validation with five FLyCube prototypes in a single orbital plane setting using inter-satellite links, we demonstrate that decentralized learning tasks can be successfully executed under the power and bandwidth constraints of 1U CubeSats. Operating at <3.5 W peak power consumption and 12.9 kbps data rate, FLyCube achieves global model convergence in several computer vision tasks including land cover classification within 3% of typical mission lifetime (1-2 years). Quantized parameter transmission accelerates this convergence to within 1% of mission lifetime while closely matching full precision performance. FLyCube demonstrates that FL in space is practically feasible on constrained low-cost hardware, thereby challenging the narrative that onboard GPUs are a necessity for FL-in-space operations. Instead, FLyCube shows that for low-cost hardware which can handle the computational load, bandwidth for parameter communication becomes the primary bottleneck in scaling to multi-orbital-plane constellations.

The rapid expansion of commercial space activities, lunar exploration, and cislunar infrastructure is creating demand for a resilient, high-capacity communications and computing architecture that extends well beyond conventional Earth-centric satellite networks. This paper presents the INterplanetary SPace InteRnEt (INSPIRE), a scalable optical networking architecture built upon SpheroSat, a modular Autonomous Optical Infrastructure Node (AOIN) that integrates optical communications, navigation, distributed sensing, and edge computing into a single spacecraft platform. SpheroSats employ the Inter Spacecraft Optical Communications (ISOC) system to establish high-speed optical links, enabling the autonomous formation of resilient optical mesh clusters in Low Earth Orbit (LEO). These clusters constitute the fundamental building blocks of a multi-layer optical network spanning LEO, Geostationary Earth Orbit (GEO), and the Earth-Moon L1 and L2 Lagrange points, providing the foundation for a scalable optical Space Internet. The proposed architecture extends beyond communications by integrating distributed computing resources directly into the network infrastructure. In addition to providing 100 Gbps optical relay services, SpheroSat nodes are envisioned as modular computing platforms capable of supporting edge processing, autonomous network management, artificial intelligence (AI) inference, distributed data storage, and delay-tolerant data services. As part of this effort, we are conducting an architectural study to evaluate the feasibility of SpheroSat-based distributed space data centers deployed in LEO, GEO, and the Earth-Moon L1/L2 regions. The study investigates candidate system architectures, optical interconnect strategies, distributed timing and synchronization, resource orchestration, power generation and thermal management, and the advantages of performing computation closer to space-based users, spacecraft, and scientific assets rather than relying exclusively on terrestrial cloud infrastructure. This paper presents the current results of this ongoing investigation, including the proposed INSPIRE network topology, SpheroSat clustering methodology, ISOC optical networking concepts, and preliminary trade studies for distributed orbital data centers. Particular emphasis is placed on the scalability of the architecture, autonomous formation and maintenance of optical mesh networks, and the evolution of SpheroSat from a communications relay node into an intelligent, distributed computing platform. We also discuss the role of orbital data centers in supporting future AI-enabled space operations, autonomous spacecraft swarms, scientific data processing, and commercial in-space services. The results establish a framework for a resilient, high-capacity optical networking and computing infrastructure capable of supporting commercial space activities, scientific missions, cislunar operations, and sustained human exploration throughout the Earth-Moon system, while providing a technology path toward a future Solar System-scale Internet.

Distributed lunar satellite swarms for low-frequency radio astronomy must satisfy interferometric requirements while maintaining a communication network under orbital evolution. This paper evaluates three swarm initialization strategies by combining baseline statistics, UV coverage and point spread function (PSF) analysis with time-varying network characterization. Both interferometric and communication performance are evaluated over ten independently randomized realizations to account for initialization variability. Under the evaluated source direction, UOEI produces the smallest mean beam width and sidelobe level, but approximately 20% of its baseline samples exceed the adopted 100-km science limit and it requires the largest inter-satellite communication range. Restricting its PSF calculation to baselines within 100 km broadens the beam and increases the sidelobe level, but both metrics remain lower than for the other initialization strategies. LBI remains within the science baseline limit and requires the shortest links for robust connectivity, while POEI provides the densest local network. These results show that maximum baseline extent alone is insufficient to characterize either scientifically useful sampling or the resulting network requirements, motivating joint evaluation of the useful baseline distribution and the network topology generated by candidate swarm configurations.

The Realizing Rapid, Reduced-cost high-Risk Research (R5) project has seen three unique spacecraft through design, build, test, launch, and operations in fiscal year 2026. Spacecraft 7 (R5-S7) performed suborbital proximity operations around its launch vehicle’s upper stage using an industrial LiDAR for relative navigation. Spacecraft 10 (R5-S10) will follow R5-S7’s mission by completing proximity operations in orbit around a client vehicle and using an event camera for celestial navigation. Spacecraft 9 (R5-S9) will demonstrate additional image processing and autonomy capabilities as well as a novel optical communication system. The R5 team has focused on refining a lean process to accomplish this rapid cadence, with emphasis placed on moving quickly from design and analysis into assembly, integration, and test (AI&T) of the system. This is particularly important for the avionics subsystem, where a mixture of commercial off-the-shelf (COTS) and custom components are used. The R5 avionics subsystem has developed a framework outside of the traditional CubeSat architecture that features flexible mechanical and electrical integration of these elements and their interfaces to allow for changes to the spacecraft configuration very late in the flow. Each spacecraft bus serves as the baseline for the next, and revisions are made to the custom power, command and data handling (C&DH), and harnessing elements to integrate the additional sense, compute, and payload components required for the mission. This paper describes the evolution of the avionics subsystem and modifications to accommodate each mission’s concept of operations and hosted payloads while maintaining very low cost and rapid schedule to achieve mission success. This includes discussion of the R5 process, flight results, and ensuing lessons learned throughout the lifecycle each spacecraft.

Fins in student-built sounding rockets are critical structural elements that provide aerodynamic stability for the entire vehicle during flight. In the traditional manufacturing approach, often utilised within our student research group and similar-scale projects, this assembly is produced in multiple stages. Within the framework of the Turbulence student sounding rocket project, for which all three tailcone iterations were specifically developed, five independent components were historically fabricated separately: four fin blades and a tailcone. This process requires not only forming each part individually, but also carrying out a complex assembly to integrate them into a single structure. Such a methodology generates key technological challenges, consuming significant time and material resources, which is particularly problematic when working with composite materials. Beyond the high labour demand, splitting the structure into separate components introduces severe design risks, including difficulties in ensuring adequate stiffness, strength, fin symmetry and alignment, as well as joint reliability at the interface between the fins and the tailcone. This paper presents a solution to this problem based on the analysis and comparison of previous solutions used in earlier development phases of the Turbulence student sounding rocket project. These experiences clearly demonstrated the essential limitations of former manufacturing methods, prompting the team to seek a new technology. To eliminate the identified drawbacks, a monolithic composite structure manufactured as a single integral unit was developed. In the context of composite materials, this approach is of paramount importance. Furthermore, eliminating separate bonding steps and employing simultaneous resin curing for the entire module guaranteed excellent structural homogeneity of the composite and higher mechanical integrity of the entire joint. As part of the work, a monolithic tailcone assembly with integrated fins was designed and fabricated. The unified composite structure yielded several technological advantages. Most importantly, it significantly reduced processing and preparation time for individual components. A key aspect was the ability to form the side walls of the element from a single sheet of reinforcement material, which ensured consistent control over fibre placement and effective filling of the internal junction zones. The article details the individual stages of project execution. It presents the CAD modelling of the components and specialised moulds equipped with dedicated channels and joining clamps. The practical phase of tooling preparation is also discussed, including material selection, precise mould sanding, application of chemical release agents, and proper cutting and lay-up of composite fabrics. The work concludes with implementation conclusions. Based on the manufacturing process, key areas for optimisation were identified. These observations will allow for further refinement of the newly developed monolithic fabrication method in future project iterations for the team and others.

University space programs must often become adept at doing more with less. The common challenges of space, such as size, weight, power, cost, complexity, and mission scope are made more difficult with limited budgets and student graduation. The NSF Center for Space, High-Performance and Resilient Computing (SHREC) is a lab focused on the advancement of mission-critical computing from spacecraft to supercomputers. SHREC has developed four university technology-demonstration missions in the last decade with a focus on onboard computing for spacecraft. This wide-ranging mission portfolio has resulted in a considerable number of lessons learned during the development and integration process. This paper serves to communicate many of these lessons to the wider community. Many university teams are very well versed in their specific research areas. However, dilemmas manifest in the details and complications of integration. The SHREC lab has faced past issues due to cost limitations, interface complexity, and knowledge transfer. These issues ranged from complex constraints in thermal simulation, to timelines for printed circuit board fabrication and assembly, to overlooked interface requirements. Student graduation can also bring complications, with turnover of experienced personnel a desired and expected part of the process. This paper explores the context, impact, and resolution of these issues across all four SHREC space-computing payloads. Each of these systems was integrated and deployed by the Space Test Program – Houston (STP-H) team and would not have been possible without their support. Lessons learned, issues encountered, and improvements made with each new mission are explored. The authors consider the key changes made to development and integration processes in the lab across these four payloads. In particular, the reduction of interface complexity across missions and the reduction of avionics complexity through the adoption of systems-on-module (SoM) bring significant reductions in development time. The authors also detail significant improvements realized in the most recent payload integration. These improved development and integration strategies can be employed in other labs and educational institutions to help students craft impactful research payloads while reducing overhead.

This paper will focus on the development of a low cost sun sensor for the attitude determination capabilities of the SpartanSat CubeSat project at San Jose State University. The sun sensor uses an orthogonally mounted two-slit mask configuration with a photodiode array underneath each to determine the two-axis CubeSat sun vector. The photodiode arrays detect the strip of light passed through the mask slit as a relative peak in light, and find the centroid of the illuminated photodiodes to map back to an angle. Typical industry standard sun sensors can cost thousands of dollars each, and many satellites require 2-4 sensors for sun vector determination in all orientations, which is very significant for the budget of a low cost small spacecraft system. This project aims to develop a sun sensor for less than $100 to better fit the budget of our CubeSat project while still providing specifications of angle accuracy about each axis within 3-5 degrees and a field of view of at least 60 degrees. The paper will present several prototypes made and the results of the tests performed to develop the sun sensor. Data examining the effect of the distance between the mask and sensor and the thickness of the mask will be correlated to the accuracy and field of view specifications. The hardware-in-the-loop tests are used to validate our accuracy and field of view requirements and demonstrate the viability of the low cost alternative.

Autonomous inspection of an uncooperative target spacecraft requires a servicing vehicle to maneuver in close proximity while relying on degradable relative navigation and onboard guidance subject to computation deadlines. For small servicing spacecraft, this challenge is intensified by limited propellant, restricted actuation authority, onboard computing limits, and reduced sensor redundancy. This paper addresses an off-nominal condition: a nominal planner may continue to issue geometrically admissible commands while the navigation belief supporting them becomes unreliable because measurements are lost or biased, or because the planner fails to return an updated command within its execution deadline. Existing safe rendezvous and proximity-operation methods commonly handle uncertainty within the estimator or nominal planner. Less attention has been given to planner-independent supervision that checks whether a proposed command remains justified by the current navigation belief and feasible within spacecraft resource limits. The scenario considers a 150 kg-class servicing spacecraft transferring from an 80 m hold point to a 25 m inspection region around an uncooperative target in low Earth orbit while remaining outside a 15 m spherical keep-out zone. Relative translation is represented in the target Hill frame. The simulation truth model includes higher-order orbital effects, bounded disturbances, actuator limitations, and model mismatch, while the onboard supervisor uses a reduced-order prediction model suitable for short-horizon onboard execution. Relative navigation is provided by an extended Kalman filter combining optical bearing measurements with intermittent range updates. The nominal planner, navigation filter, attitude-control loop, and actuators are treated as existing mission components, but their interfaces, rates, timing behavior, and physical limits are represented explicitly. At each guidance cycle, the supervisor receives the proposed command, relative-state mean and covariance, planner timing status, and resource constraints. It predicts the post-command navigation belief over a finite abort horizon while conservatively assuming that no additional measurements become available. At every prediction step, the position confidence region is compared with an uncertainty-adjusted keep-out boundary derived from a per-step allocation consistent with a total horizon risk budget. The command is accepted only if the predicted belief satisfies the safety condition throughout the horizon and the required maneuver remains feasible. Otherwise, or after a planner deadline miss, the supervisor evaluates a trusted set of hold, braking, and retreat actions and applies an admissible fallback while preserving a predefined retreat reserve. The method is evaluated in a Basilisk-based Monte Carlo campaign against direct planner execution, point-estimate supervision, and fixed-margin supervision. Test cases include correlated camera outages, measurement bias, covariance inconsistency, actuator saturation, reduced thrust authority, limited propellant, planner deadline faults, and mismatch between supervisor and truth dynamics. The evaluation reports keep-out-zone violations with binomial confidence intervals, lower-tail minimum separation, inspection completion, propellant expenditure, remaining abort capability, override frequency, fallback duration, and supervisor execution time. The paper contributes a compact assurance layer for resource-constrained small spacecraft that separates advanced autonomy from trusted safety logic without modifying the nominal planner. It investigates whether explicit supervision of the evolving navigation belief and maneuver resources can improve proximity-operation safety without excessive interruption or propellant cost.

The growing population of Resident Space Objects (RSOs) strains Space Situational Awareness (SSA) networks built around ground-based radar and optical sensors, which struggle to track the small debris objects that make up most of this population. Dual-purposed, space-based star trackers offer a cost-effective way to close this gap, but an integrated, autonomous multi-target tracking framework for distributed observer networks remains an open problem. A central challenge is the association ambiguity of angles-only measurements: with multiple close, comparably-confirmed targets, marginal association methods (e.g., LBP/JPDA) can violate one-measurement-one-track exclusivity, while a single observer leaves range fundamentally ambiguous. This paper addresses both problems within a single delta-GLMB filter on the GCRS state. Its three contributions are: (1) a tracklet-level association scheme that decides each new tracklet's identity once from the joint evidence of its frames, including asynchronous claims across platforms, while updating state frame by frame; (2) a particle representation of range-ambiguous orbits that stays calibrated for a single observer, combining an admissible-region birth model (PAR/PUB) with resample-move regeneration on the exact window posterior; and (3) a variable-N observer-network study quantifying catalog-grade OD yield against star-tracker constellation design. On an opportunistic fleet of six real hosts, the filter produced catalog-grade orbits for 403 of 2258 observed objects within 2000 s, with 97 % of estimates consistent with their object and honest to slightly conservative uncertainty. Simultaneous stereo views bring most observed objects to catalog grade quickly, while asynchronous observers still reach catalog grade by fusing passes that individually could not. A host's pointing law proves to matter far more to yield than any filter setting.

Compact CubeSat telescopes must accommodate optical shielding, structural support, and deployment hardware within a limited launch envelope. This paper presents an integrated optomechanical design study of a passive deployable baffle with concentric light-trapping vanes, combining efficient axial packaging with off-axis light rejection. A compression-spring drive and guided translation provide an architecture for cost-conscious implementation. The analysis couples aperture sizing and ray transport, shell finite-element analysis, spring and stop dynamics, and deployed alignment. A prescribed 2:1 accommodation case compares an 87.5 mm fixed baffle with a 175 mm deployed baffle: complete direct-ray blockage shifts from 61.7◦to 42.9◦off axis while preserving the required 6.9◦ full field. Assuming common 5% Lambertian reflectance, direct-plus-one-reflection pupil-power transfer is 99.2% lower at 50◦for the deployed geometry. At the full operating length, five annuli raise the ideal clamped-sleeve first natural frequency from 635 to 1284 Hz. The matched 87.5 mm spring-travel model gives 0.184 s to first contact and a 360 N aggregate arrest-force peak, while the optical geometry permits 0.634◦tilt at zero decenter. The results establish a coupled design framework in which passive deployment accommodates a larger optical structure, vanes support light rejection and structural reinforcement, and arrest loads and alignment govern the usable configuration. This framework connects compact launch accommodation to enhanced optical shielding for CubeSat payloads.

The growth of activity near the Moon is increasing the need for space situational awareness capabilities that can operate beyond Earth orbit. This paper presents flight results from using the Cislunar Autonomous Positioning System Technology Operations and Navigation Experiment (CAPSTONE) as an electro-optical observer of other spacecraft in the lunar and cislunar environment. During its extended mission, CAPSTONE has acquired star-tracker imagery of targets including the Artemis II spacecraft Integrity and the Korea Pathfinder Lunar Orbiter, Danuri. These observations demonstrate how an existing spacecraft sensor can be repurposed to support cislunar detection and tracking without a dedicated surveillance payload. The central contribution is an end-to-end methodology that connects uncertainty-aware tasking with multi-frame detection and tracklet validation. Observation planning uses propagated target-state samples to represent the region in which the spacecraft may appear on the focal plane at each candidate epoch. These samples are grouped with clustering methods that select a limited set of pointing centers while covering the largest possible fraction of the predicted angular search region. The resulting tasking strategy can distribute observations across both pointing direction and time, using temporal dithering to capture changes in the target’s apparent position and motion while reducing the risk that a single pointing or epoch misses the object entirely. Because geometric coverage alone does not guarantee a usable observation, each candidate pointing and epoch is also evaluated for expected brightness. A Cook-Torrance BRDF visual-magnitude model captures uncertainty in illumination geometry and target optical properties, allowing the scheduler to favor opportunities that are both likely to contain the target and likely to produce a detectable signal. This creates a practical balance between covering more of the angular uncertainty region and concentrating observations where brightness and motion conditions are most favorable. Flight imagery is then used to assess how well these combined geometric and photometric tasking metrics predict successful acquisition and to identify which planning assumptions most strongly affect detection performance. The image-processing pipeline is designed to recover weak moving targets from sequences in which no single frame may provide a decisive detection. Background subtraction and difference imaging suppress the static star field and other persistent structure, while co-adding along candidate motion paths increases the effective signal-to-noise ratio of faint sources. Statistical outlier rejection is then used to remove isolated detector artifacts and other nonpersistent features before detections are associated across time. The resulting candidate tracklets are evaluated for consistency in angular position, apparent motion, and brightness, with track-before-detect methods used when the target remains below the single-frame detection threshold. Validated detections are transformed through the astrometric plate solution into line-of-sight measurements and compared with reference ephemerides to assess astrometric accuracy and tracklet consistency. The paper evaluates the end-to-end performance of the approach using flight imagery, with emphasis on whether uncertainty-aware tasking leads to successful acquisition and whether the processing pipeline produces reliable angular tracklets. Together, the results demonstrate a practical path toward using operational spacecraft as distributed cislunar sensors for maintaining custody throughout the lunar environment.

This paper presents a formulation for the rendezvous and synchronization of an underactuated chaser spacecraft with a target on a circular orbit using minimal thrust. The chaser is equipped with full rotational actuation and a single-axis unidirectional thruster. The proposed formulation features a controllable linear time-invariant system, where linear control methods can be applied to generate the rendezvous and synchronization trajectories. The key to achieving a linear system is to assign the body-fixed y-axis along the thrust direction. Subsequently, a linear–quadratic regulator is designed to generate fuel-optimal, continuous-thrust rendezvous trajectories. Then, the design is evaluated with a representative scenario, demonstrating relative position and velocity convergence while achieving synchronization between the chaser and the target. The thrust profile indicates that an initial impulse propels the chaser away from the target, after which both thrust and the natural drift of the orbital dynamics are exploited to guide the chaser to rendezvous. The proposed approach provides a practical means of enabling both rendezvous and synchronization for underactuated spacecraft, a capability that is particularly relevant to small satellite missions where stringent mass, volume and fuel constraints can preclude fully actuated spacecraft designs.

Validating CubeSat Attitude Determination and Control Systems (ADCS) before launch requires reproducing the time-varying geomagnetic field encountered in Low Earth Orbit (LEO). Previous magnetic testbeds have reproduced orbital magnetic-field profiles at accelerated time scales or with manually tuned feedback controllers. This paper presents a real-time orbital magnetic field generation system based on a three-axis square Helmholtz cage and experimentally compares four methods for reproducing dynamic orbital magnetic fields. The Helmholtz cage is designed to generate controlled magnetic fields up to 200 µT, is designed for ±1% uniformity within a 0.35 m cube, accommodates CubeSats from 1U to 12U, and actively cancels the ambient magnetic field. Orbital magnetic field commands are computed using the International Geomagnetic Reference Field (IGRF) model. Four methods are evaluated for generating the corresponding coil current commands: a PI controller tuned using the Nelder–Mead optimization method, a PI controller tuned using a Genetic Algorithm (GA), a linear calibration method with magnetic-field feedback, and a Gradient Boosting regression model. Experimental results show that the linear calibration method and the Nelder–Mead-tuned PI controller achieve similar tracking performance, with accuracies above 99.4% on all three axes and per-axis RMSE below 0.03 µT relative to the in-loop magnetometer across all ten orbital cases. The Gradient Boosting regression model achieves tracking accuracies above 99.3%, whereas the GA-tuned PI gain set shows greater variability across orbits, with accuracies ranging from 94.2% to 99.5%. These results demonstrate that the Helmholtz cage is a repeatable platform for reproducing orbital magnetic-field profiles for pre-launch CubeSat ADCS testing under dynamic orbital magnetic-field conditions.

Detumbling is an essential part of satellite operations. If the tumbling rate of a satellite isn’t constrained, the performance of essential subsystems such as radios or solar panels’ Maximum Power Point Trackers (MPPTs) cannot be ensured. One of the control algorithms most commonly used in simple, low-cost satellites is B-dot. In B-dot, the control applied to each satellite axis is the time derivative of the measured external magnetic field in that axis. Here, we propose a low-cost testbed for the validation of the performance of satellites using a B-dot detumbling controller and magnetorquer actuators. While it is generally difficult to experimentally test the performance of Attitude Determination and Control Systems (ADCS), due to the need to emulate the frictionless environments these systems operate in space while allowing free rotation in 3 axes, the properties of B-dot control allow for the design of simplified testbeds where the satellite can be kept static. The core component of the proposed testbed is a set of Helmholtz coils, used to create a variable magnetic field simulating the rotation of the satellite in Earth’s magnetic field. External equipment can be used to measure the magnetic field generated by the magnetorquer actuation. The phase between the two can then be checked to validate the performance of the system. This approach allows for testing fully assembled flight models inside clean rooms, since the satellite is static, the equipment is small and portable, and no extra risk is involved in integrating the models in testing rigs. Two setups are proposed to measure the magnetic field generated by the magnetorquers, one based on standard laboratory equipment, i.e., a waveform generator, oscilloscope, and gaussmeter, and one based on a custom platform designed using low-cost electronic components, i.e., magnetometers, current sources, and microcontroller integrated circuits. The proposed testing method has been used to validate the B-dot detumbling performance of two CubeSat missions, ISTsat-1 and Argus 3. The experimental tests highlighted where the real systems diverged from the numerical models, exposing coding bugs that would otherwise remain concealed and that threatened the CubeSat’s ability to successfully detumble.

CubeSats have emerged as a dominant platform for low cost space missions. They enable rapid technology demonstration, scientific experimentation and increasingly complex mission concepts such as constellations and swarms. As these missions scale in number and functional complexity, traditional centralized avionics architectures, where a single onboard computer is responsible for the majority of spacecraft functions, become increasingly constrained. Centralized designs are inherently vulnerable to single-point failures and exhibit reduced scalability as software, integration effort and subsystem coupling grow with mission complexity. This work presents a distributed avionics architecture for CubeSat-class spacecraft. This architecture is developed as a foundational technology to support constellation and swarm missions. In the proposed approach, core spacecraft functions such as power supervision, mission control and subsystem interfacing are decomposed across multiple semi-autonomous avionics nodes. Each node is responsible for a bounded functional role and communicates with other nodes through a shared message-based spacecraft data bus with defined priorities and timing behavior. A hierarchical authority structure is retained to preserve coherent mission-level control while reducing dependence on a single centralized processor. The architecture is defined with strict CubeSat constraints in mind, like mass, volume, power and development effort. Functional partitioning is guided by fault containment, timing requirements and subsystem locality, rather than by board-level convenience. Standardized interfaces and node-level state machines are employed to support modular integration and controlled degradation in the presence of faults. Fault detection and recovery mechanisms such as heartbeat monitoring, supervisory reset authority and node isolation are incorporated to prevent local failures from propagating across the avionics system. A representative multi-node avionics system is implemented using microcontroller-based hardware and evaluated using a FlatSat and hardware-in-the-loop test methodology. Verification activities emphasize architectural behavior rather than flight qualification. These activities focus on communication latency, bus loading, power consumption, fault response behavior and long-duration operational stability. Fault injection scenarios, including node resets and communication disruptions, are used to examine system robustness and degradation characteristics. While the primary focus of this work is the internal avionics architecture of a single CubeSat, the system is evaluated in the context of constellation and swarm applications. Standardized health and status messaging interfaces are defined to enable uniform telemetry reporting across multiple spacecraft. The scalability of these interfaces is examined through software-based emulation of multiple spacecraft instances. This allows analysis of communication bandwidth, processing overhead and integration effort as the number of spacecraft increases. This demonstrates how distributed avionics architectures support swarm-level capabilities such as fleet health monitoring, role reassignment and graceful degradation without introducing inter-satellite dependencies for basic spacecraft safety. This work presents the proposed architecture, design methodology and verification approach. It also discusses key design trade-offs relevant to the adoption of distributed avionics as a scalable and fault-tolerant foundation for future CubeSat constellation and swarm missions.

Safe Mode is an operating mode that is essential for CubeSats to conserve energy and safeguard onboard equipment during anomalous situations, such as low battery charge, communication loss, or radiation-induced failures. In a long duration of Safe Mode, the On-Board Computer (OBC) is typically turned off or put into a "low-power" mode, which means that continuous navigation and spacecraft state awareness are not possible. This will take more time to get back into operational mode before mission operations can resume. Current solutions are based on continuous Global Navigation Satellite System (GNSS) measurements or on-board orbit propagation, which are power consuming and not suitable for long duration low-power operation. An ultra-low-power autonomous state estimation coprocessor is proposed that can keep an estimation of the current overall state of the spacecraft while the primary OBC is offline. The coprocessor does not perform full orbital propagation, but rather determines the minimum operational state needed for autonomous recovery based on measurements from low-power sensors such as Sun sensors, magnetometers, inertial measurement units (IMUs), real-time clocks (RTCs), and electrical power subsystem telemetry generated by a dedicated microcontroller. These measurements are used to validate whether the recovery conditions have been met and if the OBC should return to nominal operations. The proposed architecture features a dedicated sensor interface, a finite-state-machine (FSM)-based controller, a fixed-point sensor-fusion engine, and a wake-up decision module optimized for ultra-low-power operation. The digital hardware is developed using the SkyWater SKY130 open-source process and tested using the OpenLane application specific integrated circuit (ASIC) design flow. The performance will be evaluated based on silicon area, timing, power consumption, estimation error, and wake-up decision reliability for representative mission scenarios of CubeSats. This proposed coprocessor design will allow for low power situational awareness while allowing for mission computing to be performed independently on a spacecraft, which will minimize standby power consumption. This architecture offers a scalable digital solution for resilient CubeSat avionics and future autonomous spacecraft with strict energy constraints.

Future space-based imaging missions will require computationally intensive on-board signal processing. Commercial terrestrial processors can meet these performance demands, but they are unreliable in space, where ionizing radiation induces logical faults that cause a range of malfunctions. This vulnerability limits their use in critical imaging applications such as planetary horizon alignment. Radiation-hardening techniques can make electronics operate reliably in space, but the resulting systems are cost-prohibitive for most commercial missions. This paper presents RadPC, a computing technology that provides fault-tolerant operation in space using lower-cost commercial off-the-shelf (COTS) components. RadPC is a quad-core, majority-voted RISC-V computer implemented on a field-programmable gate array (FPGA). It combines quad-modular redundancy with memory scrubbing and partial reconfiguration to repair faulted cores and return them to service. A distinguishing feature of RadPC is its ability to combine otherwise unused FPGA resources into memory-mapped hardware accelerators, or coprocessors, that integrate directly with the computer's architecture. This approach places high-performance hardware-based signal processing in the same FPGA fabric as a fault-tolerant controller that keeps the overall system operational in the presence of radiation. In this paper, we describe the RadPC hardware accelerator architecture and demonstrate its use for horizon detection and image filtering on images captured during low Earth orbit (LEO). We present image and error-mitigation data from on-orbit operation showing that the system detects and recovers from radiation-induced faults while continuing to process imagery. Together, these results show that RadPC can deliver both the fault tolerance and the processing performance required for space-based imaging, offering a low-cost path to advanced on-board processing for small satellites and other commercial missions.

The Space Test Program (STP) at the Department of War (DoW) has long supported the development, evaluation, and advancement of new technologies crucial for the future of spaceflight. STP-Houston provides opportunities for DoW and civilian space agencies to perform on-orbit research and technology demonstrations from a variety of space-faring platforms including the International Space Station. The STP-H12 payload is scheduled for launch in December 2027 via the Japan Aerospace Exploration Agency. It will feature new technologies, including an RGB-neuromorphic camera fusion experiment developed by the National Science Foundation Center for Space, High-Performance and Resilient Computing, working closely with several government and industry partners, known as the Visual And Neuromorphic Tracking and Geosensing Experiment (VANTAGE). This paper discusses the design methodology for the SHREC Space GPU (SGPU) carrier card, one of the modules on the upcoming VANTAGE mission. The GPU-SoC used for the SGPU, the NVIDIA Jetson Orin NX, is available only as a System on Module (SoM), thereby posing additional reliability hurdles. Mitigation techniques, such as watchdog timers and current monitoring for single event latchup, were implemented within the SoM carrier board to address known radiation vulnerabilities in the GPU and requisite auxiliary electronics. Support circuitry on the carrier comprises a mix of commercial and rad-hard components, the selection of which was informed by the mission’s science objectives, duration, environment, and budget. The SoM-carrier design approach will hopefully prove to be a reliable option for enabling the usage of advanced computational devices in space.

CLARREO Pathfinder launched to the ISS on May 15, 2026 in the trunk of the SpaceX Dragon as part of Crew Resupply-34. On May 22, the payload was then installed robotically to the external site 3 of ELC-1 and powered on, which commenced the on-orbit commissioning phase of the mission. The HySICS Pointing System (HPS) then underwent many weeks of both functional and performance checkouts to verify its ability to provide the required pointing capability for the HySICS instrument to conduct science observations. During commissioning and early operations, performance measurements both directly and indirectly showed that the HPS was meeting its requirements, and example results are included in the paper, including but not limited to: post-launch gimbal pointing performance, solar pointing jitter, measurements of ISS base motion jitter (inc. comparisons to other payload measurements such as 8 years of similar TSIS measurements), measurements of ISS structural attitude accuracy and stability, Earth pointing accuracy (inc. confirmation of ground alignment information), and geolocation accuracy. As previously described in The HySICS Pointing System: Precision Pointing of CLARREO Pathfinder from the ISS [1], the ISS is a challenging operating environment, and several of the experiences and lesson learned are described, including specific experiences and observations of the highly-constrained viewing environment and challenges with ISS-provided position, attitude, and timing information. CLARREO Pathfinder is now well underway toward achieving its ambitious objectives to provide unprecedented highly radiometrically accurate measurements of Earth-reflected sunlight and using that accuracy to calibrate other orbiting sensors. Author's note: we are currently in the midst of the ambitious commissioning campaign (hence the delay in submission of this abstract), so new, exciting results are incoming on a daily basis that will be incorporated into the paper as available. Sufficient results will be available for the October 2 paper submission deadline. [1] Brown, P., Holden, T., “The HySICS Pointing System: Precision Pointing of CLARREO Pathfinder from the ISS”, IEEE Aerospace Conference, 2024. https://doi.org/10.1109/AERO58975.2024.10521121

Searching the solar system for signs of past and present habitable environments and of life itself is a major goal directing research for planetary exploration in the coming decades. The OrganiCam instrument aims to assist with this search in the form of a lightweight and compact instrument that can perform both time-resolved luminescence and Raman spectroscopy. Both methods are sensitive to organic compounds and can be used in combination to search for organics (luminescence) and identify them further after initial detection (Raman). The scientific goals and traceability are provided in more detail in Gasda et al. (2021), while this work focuses on the instrumentation and testing of optical properties. The OrganiCam instrument builds upon the successful design and methods of the Mars 2020 SuperCam and SHERLOC instruments, including the direct incorporation of SuperCam’s intensifier and relay optics systems. The “fore-optics” (optical systems before the intensifier) differ from those of SuperCam and consist of two optical subsystems, the objective assembly and the spectrometer assembly, for a total of four optical subsystems within the instrument. OrganiCam’s optical performance metrics are based on the science traceability metrics described in Gasda et al. (2021). These metrics are quantified as Modulation Transfer Function (MTF) requirements in the form of the system’s contrast at a specified spatial frequency. To measure the overall system’s performance, each of the four optical subsystems is analyzed individually, then the results are multiplied together to calculate a total system MTF. The MTF requirements vary between operational modes, with image collection requiring higher contrast at higher spatial frequencies than spectra collection. The intensifier and relay optics assemblies have known optical characterizations from the manufacturer and SuperCam development, respectively. Here, we focus only on the performance of the objective and spectrometer assemblies. In addition to nominal performance metrics, to ensure robustness of the OrganiCam optical system, we are modelling manufacturing and assembly tolerances and conducting thermal analyses to show that the instrument meets performance requirements in a variety of situations. For tolerance analysis, the Monte Carlo method is applied to create thousands of system variations and calculate the MTF of each system individually. For the thermal analyses, only two scenarios, “hot” (+55°C) and “cold” (-55°C), are modeled, representing the temperature extremes expected for operation during cruise and a nominal mission to the equatorial region of the surface of Mars. These same temperatures were used to qualify the SuperCam instrument onboard the Mars 2020 rover. In both the tolerance and thermal analyses, some compensation occurs with the instrument’s focus stage. Initial tolerance analyses at room temperature have proven the instrument to be robust to both manufacturing and assembly errors. However, initial thermal analyses resulted in several test cases that showed inadequate performance, suggesting a need for system modification under these extreme temperature scenarios.

The upcoming Landsat 10 mission is expected to generate substantially larger volumes of multispectral image data than previous Landsat satellites due to its increased spatial resolution and expanded spectral dimensionality. This can create significant challenges for data storage and transmission. To address these limitations, the CCSDS 123.0-B-2 standard, developed by the Consultative Committee for Space Data Systems, provides a framework for multispectral and hyperspectral image compression. This standard has been selected for Landsat 10 to support near-lossless compression of the mission science data. This work presents an FPGA-based VHDL test bench for investigating the real-time hardware acceleration of the CCSDS 123.0-B-2 image compression algorithm. The implementation supports Band Interleaved by Pixel (BIP) order, adaptive prediction, and the CCSDS hybrid entropy encoder for efficient compression. The development of this work is performed using AMD Vivado and Vitis. A MicroBlaze soft processor provides runtime configurability of the compression parameters, while Vitis High-Level Synthesis (HLS) is used to accelerate computationally intensive components. Compared with software implementations in C, FPGA-based compression offers the potential for higher throughput through parallelism and pipelining. This enables real-time processing for bandwidth-constrained embedded systems. This work provides a scalable hardware test bench for evaluating the CCSDS 123.0-B-2 standard as a pathway toward real-time, onboard near-lossless compression of Landsat 10 imagery.

This study presents the design, characterization, and implementation of an optical Hardware-in-the-Loop (HIL) testbed for the quantitative evaluation of Image Quality Control (IQC) models in Earth observation nanosatellites. Onboard classifiers are usually validated in software with stored images, which neglects how the camera actually forms the image: optics, sensor, and tone response. The objective is to provide a laboratory platform that approximates far-field imaging, so that digitally defined variations in blur and illumination are physically reacquired by the processing hardware rather than processed as ideal data. The testbed integrates, along a single axis, a high-resolution display, a collimator lens, and an OpenMV H7 Plus, whose camera recaptures the projected Earth-observation scenes and runs the classifier on board. The design covers the geometric dimensioning of the bench, the definition of the active display area from the sensor field of view, and the mapping of each scene to the classifier input. The assembly operates in an enclosure isolated from ambient light; it was mechanically aligned, its tone response was characterized, and its stability was monitored throughout the campaign (NSAD between 0.34% and 0.39% over 104 min). The testbed was applied to the MobileNetV2 classifier of Waskiri-Sat, the first nanosatellite developed in Bolivia, which labels scenes as CLOUD (>70% coverage) or CLEAR (<30% coverage) for autonomous image quality control. One hundred test tiles were evaluated under ten blur and illumination levels, both read from storage (software-in-the-loop, SIL) and recaptured through the optical path (HIL), on the same hardware. In the nominal case, the HIL path reduced accuracy from 0.90 to 0.75 and cloud recall from 0.94 to 0.52, a statistically significant change, while class separability remained high (AUC 0.944). The losses concentrated on dark overcast scenes, which the display-camera chain darkens disproportionately, and digital under-exposure proved far more harmful in HIL than in SIL. Such effects remain hidden in a purely digital test, showing that successful software validation alone does not guarantee that a classifier's operating threshold transfers to physically captured imagery. Including optical reacquisition in the validation of onboard classifiers therefore allows downlink decisions to rely on thresholds verified with the actual imaging chain, making more efficient use of the limited bandwidth and of the power and processing resources of the On-Board Computer (OBC).

Gamma-ray bursts (GRBs) are the most energetic electromagnetic events in the universe, and although high-energy gamma-ray observations have provided significant insights; the mechanisms of GRB origins remain unclear. More study of their afterglows, the broad-spectrum radiation following a GRB, is required for further understanding. The Antarctic Demonstrator for the Advanced Particle-astrophysics Telescope (ADAPT), a NASA mission led by James Buckley of the Washington University in St. Louis Physics Department was developed to address this question. Using a high-altitude weather balloon over Antarctica, ADAPT is designed to demonstrate a Compton scattering detector for a future space-based gamma-ray and cosmic-ray observer. Planned for a December 2026 flight, ADAPT will provide unprecedented sensitivity and a wide field of view, producing localizations and polarization constraints on a few GRBs in seconds. However, ADAPT alone cannot obtain precise GRB localization or capture optical afterglows, limiting its ability to fully capture these bursts. AIRIS (ADAPT Incidence Resolution & Imaging Subsystem) is an optical telescope mounted on the same platform as ADAPT with the ability to quickly reposition itself after a detection event. When ADAPT detects a gamma-ray burst (GRB), AIRIS quickly repositions itself, slewing at up to 30 degrees per second to refine the GRB's location and capture optical images of its afterglow. Due to the combination of ADAPT’s rapid computing pipeline, the location of AIRIS on the same platform as ADAPT, and AIRIS’ rapidly slewing mount, AIRIS is quickly able to observe the optical afterglows of GRBs. Furthermore, AIRIS’ near-space vantage point, low-noise SONY CMOS sensor that produces clear, high-quality images, and large aperture 200mm f/1.8 lens, enables the observation of very faint signals. Through burst localization and time evolution tracking, AIRIS enables observations essential for studying GRBs’ development. The data AIRIS gathers will feed into NASA's General Coordinates Network, guiding telescopes worldwide to observe GRB afterglows and contributing to multi-messenger astronomy.

The EAGLE-VSWIR (Visible to Short Wavelength Infrared) free-flyer spacecraft is host to a solar-reflectance imaging spectrometer instrument. The instrument measures global land and water reflectance and will provide data for investigations ranging from terrestrial and aquatic systems to cryosphere and geologic composition on Earth. Launch is scheduled for 2028 with a 16-day revisit time sun synchronous orbit at an altitude of 620 km to 660 km. VSWIR measures from 380 nm to 2500 nm wavelengths with a 10 nm spectral sampling resolution. It expands on the heritage design of the Earth surface Mineral dust source InvesTigation (EMIT) instrument that utilized a telescope operating near room temperature, a Dyson spectrometer operating at 240 K and an actively cooled 155 K focal plane array. In the case of VSWIR, there are dual telescopes, spectrometers and focal plane arrays each achieving a large swath of 92.5 km. The VSWIR thermal control architecture consists of active and passive elements and implements a single cryocooler similar to EMIT. An overview of the overall thermal control design approach is presented along with discussion of various trade studies that have been performed.

NASA’s Dawn mission studied the largest main belt asteroids Ceres and Vesta, which inform us of early solar system formation and evolution. Dawn’s Visual and Infrared Spectrometer (VIR) was the primary instrument investigating the mineralogy of these bodies. The spectra taken by VIR have revealed rich deposits of hydrated minerals highly variable at regional scales. However, the VIR spectra contain multiple types of noise which must be corrected prior to analysis. These include odd-even channel noise, spectral spikes, vertical stripes, and artifacts. Odd-even channel noise is caused by the incoming signal being read by IR sensitive cells on two multiplexers, with one reading the evens and another reading the odds. Non-identical material properties of the multiplexers manifest as different gains, with the combined signal appearing to have a saw-tooth pattern. Spectral spikes are due to random events, vertical stripes are due to differences in the relative response of the sample detectors, and artifacts are due to the non-homogeneous instrumental response of the detector. We are carrying out an improved artifact correction procedure to enable machine learning analyses and wider use of the VIR spectra by the planetary science community. We will demonstrate efficacy of applying an artifact correction matrix derived from a statistical sample of observations, specifically in correcting spectral spikes, vertical stripes, and artifacts. We characterize here the performance of this existing method on addressing the odd-even channel noise, and discuss improvements to address this source of noise. Dual-multiplexer spectrometers are commonly flown on spacecraft, and improvements in corrections to the often associated odd-even noise has implications for interpretability of the infrared spectra taken by other spacecraft

Estimating Leaf Area Index (LAI), a key representative for vegetation biomass prediction, is costly to measure by hand at the scale precision agriculture requires. Uncrewed aerial systems (UAS) capture plant-level detail over small areas, while satellites cover large areas at coarse resolution, motivating models that can bridge UAS and satellite scales. This work trains 16 multi-channel CNNs to predict LAI in California vineyards from six co-registered UAS channels (RGB, near-infrared, Digital Surface Model (DSM), and Canopy Height Model (CHM)), comparing Early Fusion (one shared convolutional stack) against Late Fusion (per-channel-group stacks) across every channel combination. On held-out test flights, Early Fusion using RGB and canopy height achieved the best fit (R2 = 0.91), and Early Fusion generalized better than Late Fusion overall despite Late Fusion having up to 4 times as many parameters. DSM alone consistently hurt predictions by conflating terrain elevation with CHM, while CHM data was the most valuable auxiliary channel. Aggregating field-scale predictions to the Landsat 8 30 m grid showed good agreement with NDVI and surface reflectance, though the relationship with EVI was unexpectedly negative. These results identify which channel combinations and fusion strategies best support scaling UAS-based biomass estimates toward satellite resolution.

Near-infrared imaging spectrometers are critical instruments for planetary and Earth-observing missions, generating high-dimensional datasets characterized by complex, nonlinear spectral structures. Extracting physical meaning, like mineralogy, from these datasets remains an operational challenge for aerospace data systems. In scans of Ceres by the Visible and InfraRed mapping spectrometer (VIR) on NASA’s Dawn mission, each pixel represents a spectrum of 432 channels that covers the carbonate, phyllosilicate, and ammonium absorptions. Unsupervised clustering is a promising approach to determining which pixels belong to the same surface unit, but existing methods fall short by treating pixels independently or exhibiting high sensitivity to tuning parameters. We present Graph K-Subspaces (GKSS), a method that clusters spectra by fitting a union of subspaces (UoS) while incorporating a spatial graph regularization term that encourages co-located pixels to share cluster assignments. Because each cluster is a subspace rather than a single mean spectrum, it can absorb within-unit changes in brightness and grain size, and we choose the graph weight without labels by trading subspace fit against spatial cohesion. On benchmark datasets, GKSS obtains competitive performance with state-of-the-art methods while remaining robust to the choice of its graph parameters. To measure accuracy under mission conditions, we inject coverage gaps, brightness seams, and noise measured from VIR scans into labeled benchmark scenes, so that every degraded result can still be scored. Under these perturbations, GKSS loses at most 5.2 points of balanced accuracy, while the best competing method loses up to 13.8 points. On VIR data from Ceres, GKSS gives stable clusters that do not follow observation seams and, when tuned on the mapped units of Occator crater, gives Cerealia Facula a cluster of its own, separate from the crater floor.

Recent advancements in processing and sensing hardware combined with improved affordability in the satellite launch process have led to the increasing adoption of CubeSats for modern maritime domain awareness (MDA) capabilities. Likewise, line scanner hyperspectral image (HSI) sensors provide unique capabilities on these payloads, offering high spatial resolution at low Earth orbit (LEO) with high spectral resolution. This allows for a diverse set of environmental missions including illegal fishing tracking, water quality monitoring, and chemical/oil spill detection. With high spectral resolution comes large data volumes, which must be processed onboard the satellite to provide near-real time information that is actionable on the ground. To process and filter this data, artificial intelligence and machine learning (AI/ML) models are deployed onboard the spacecraft, turning the full dimensional image data into concise information suitable for low bandwidth downlink. During image capture, unexpected motion by the spacecraft leads to distorted imagery that subsequently gets processed by onboard AI/ML models. This paper examines the performance impact of miscalibrated HSI data on neural networks tasked with a ship detection objective. Across a range of common model architectures, we examine performance impact due to spectral misalignment along with geometric distortions from unaccounted for spacecraft rotation. Additionally, we investigate the degree to which different levels of model quantization might compound or mitigate performance degradation under these imaging conditions, as CubeSats have tightly constrained resources and model quantization is common to reduce size and power usage on payloads. Notably, we found that smaller model architectures tended to be more robust to artificial spectral misalignment artifacts. We also found that FP16 quantization effectively halves the memory footprint with negligible degradation in baseline accuracy. These results are informative for model selection on a HSI CubeSat mission and should be considered when balancing SWaP constraints with the extent to which imagery is preprocessed before inference. Keywords: Hyperspectral imagery (HSI); pushbroom sensor; onboard processing; machine learning; convolutional neural network (CNN); vision transformer (ViT); quantization; maritime domain awareness (MDA)

Atmospheric turbulence degrades ground-based satellite imagery through blur that varies across frames. Multi-frame blind deconvolution (MFBD) estimates images and optical states but incurs substantial computational cost. We introduce \textbf{LUCID-A}, a physics-informed MFBD method that learns atmospheric-state initialization without per-frame optical supervision. A burst encoder predicts 32-dimensional atmospheric states and frame weights, enabling differentiable Fourier optics, analytic image reconstruction, and one safeguarded refinement. On simulated SILO satellite-imaging bursts spanning 11 turbulence conditions, replacing eight low-sharpness frames using fine-tuned Real-ESRGAN super-resolution (SR) increases LUCID-A's structural similarity index (SSIM) from 0.433 to 0.544. With SR replacement, measured restoration-stage latency is 0.20 s for LUCID-A versus 3.01 s for MFBD, excluding SR preprocessing, with lower peak signal-to-noise ratio (PSNR): 16.68 versus 18.38 dB. Downstream classification experiments show that these restoration gains do not uniformly translate to recognition gains, highlighting the need to evaluate both image fidelity and task utility. These results establish a measured tradeoff between reconstruction fidelity and computational cost for satellite imaging. More broadly, learned initialization offers a route to accelerating inverse imaging while retaining explicit optical modeling.

We consider the problem of autonomously tracking surface vessels using a proliferated LEO satellite constellation. Each satellite has a software application stack that includes sensor signal processing, local fusion and tracking, global fusion and tracking, sensor tasking, mission planning, and communications management. To support decentralized sensor tasking and track picture merging at each asset, we developed message-sharing and fusion algorithms. The global track picture is truly distributed across the constellation with no single asset holding the complete picture. The track picture over any area-of-interest can be obtained by combining the track pictures from assets that have responsibility over regions that compose the area-of-interest. The sensor tasking is decentralized with each asset making sensor tasking decisions based on its track picture, which covers its own sensor’s field-of-regard and those of its primary neighbors. Through this approach, each asset can anticipate the actions of its neighbors when generating its own sensor tasks, leading to collaborative decision-making without a centralized resource manager. The developed algorithms enable monitoring of commercial shipping traffic using, for example, optical sensors on pLEO constellations. Results are given from a simulated scenario with a Walker satellite constellation, a set of surface vessels of interest, and surface background traffic.

The proliferation of low-altitude, non-cooperative small Uncrewed Aerial Vehicles (sUAVs) introduces significant security vulnerabilities that legacy airspace monitoring systems cannot resolve due to terrain masking and severe electromagnetic blind spots. Establishing a continuous tracking perimeter against targets with minimal Radar Cross-Sections (RCS) requires deploying heterogeneous sensor networks that layer disparate modalities, such as radar and electro-optical sensors. While operationally necessary, optimizing the joint spatial deployment of these networks is traditionally an NP-hard Maximal Covering Location Problem (MCLP). Earlier works have proposed Synergistic Heterogeneous Active Placement Engine (SHAPE) methodology which successfully resolves this combinatorial explosion through multi-stage spatial decomposition and an active decision-level OR-fusion solver, reducing high-fidelity deployment execution times to a matter of minutes. To push this operational agility even further into near-real time execution, this paper extends the SHAPE architecture by introducing Deep Neural Network (DNN) based surrogate models. This approach replaces the initial deterministic physical wave propagation calculations, specifically Longley-Rice tropospheric attenuation evaluated over 3D Digital Elevation Models (DEMs), with rapid predictive models. While the underlying SHAPE architecture is designed to accommodate modular plugins for any sensor modality, the DNN surrogates developed in this work are specifically tailored for radar and electro-optical networks. The proposed methodology leverages these trained models to instantly approximate radar signal attenuation, multi-path fading, and optical geometric limits without the computational overhead of raw raytracing. During spatial evaluation, topographic truncation and multidimensional K-Means++ clustering extract structurally prominent candidate coordinates, which the DNN surrogate evaluates to generate marginal probability matrices. These matrices feed directly into the dynamic hybrid optimization engine, which calculates the synergistic fusion gain between the long-range radar macro-layer and the gap-targeted optical micro-layer. Simulation results demonstrate that integrating deep learning surrogate models into the K-Means++ spatial partitioning pipeline elevates the already highly scalable SHAPE framework to near-real time execution. This architecture empowers network planners with instantaneous spatial agility, enabling the dynamic and rapid deployment of resilient, multi-modal surveillance perimeters capable of continuously tracking Group 1 stealth threats (0.1 m² RCS at 100 ft altitudes) across complex urban canyons and high-relief mountainous topography.

Understanding the changing surface topography and vegetation structure of Earth is a major priority for NASA's Earth Science Division. However, critical algorithmic gaps exist in traditional surface mapping methods that hinder their use for observing dynamic landscapes. If acquisition parameters vary widely (e.g., illumination, view angles, seasonal shifts), or if the surface target has complicated physical characteristics (e.g., harsh shading, diffusive reflectance, homogeneous surface texture), traditional methods often struggle to create accurate 3D terrain models. To overcome these limitations, we present a novel continuous scene representation method using heterogeneous Earth-observation datasets. Our method, which we call SPECTRA-NeRF (Shadow-aware, Physics-informed Estimation of Color and Topography for Reconstruction and Appearance NeRF), utilizes Neural Radiance Fields (NeRFs) to encode diverse physical and biophysical information. SPECTRA-NeRF successfully integrates and fuses multi-modal data from various sensors, including high-resolution multispectral 8-band WorldView-3 imagery, orthorectified surface bidirectional reflectance products from the National Ecological Observatory Network (NEON) Imaging Spectrometer (NIS), and Digital Surface Models (DSMs) from the NEON Airborne Observation Platform (AOP) LiDAR instrument. In this paper, we discuss the technical details that enable SPECTRA-NeRF to accurately reconstruct and relight dynamic Earth observation targets and compare results with traditional multi-view stereo (MVS) photogrammetry methods. Initial findings demonstrate that unifying these heterogeneous datasets within a physics-based NeRF architecture significantly enhances 3D terrain and vegetation mapping. Our approach offers a pathway to inform science traceability matrices for future missions concepts, such as the Surface Topology and Vegetation (STV) incubation study, by leveraging the plethora of Earth-observation data and encoding their information in a unified 3D scene representation.

In recent years, space situational awareness (SSA), which involves the detection and continuous monitoring of resident space objects (RSOs), has attracted increasing attention for ensuring safe and sustainable space activities. In particular, there is growing interest in autonomous and high-frequency optical observations using spaceborne sensors to complement ground-based observations, especially for the detection and continuous monitoring of small and dim RSOs. Star trackers (STTs), which are widely used for satellite attitude determination, are expected to serve as cost-effective optical observation sensors because they have the potential to detect not only stars but also RSOs. However, when multiple RSOs are simultaneously observed, associating each light source with the corresponding track and maintaining track identities become challenging. This study proposes a multi-object tracking method based on the Simple Online and Realtime Tracking (SORT) algorithm for multiple RSOs observed in STT images. The authors previously proposed a SORT-based method for RSOs observed as point sources, in which tracks and observed RSO candidates are associated based on Euclidean distance. However, data association based on a fixed distance threshold does not consider the prediction uncertainty of each track. Therefore, it may cause incorrect associations when multiple light sources are located near a predicted position. To address this issue, this study introduces a Mahalanobis distance-based data association method into the SORT-based tracking framework using the predicted positions and prediction error covariances obtained by a Kalman filter. In addition, a brightness gate based on the difference between predicted and observed brightness is introduced to address cases in which data association is difficult using position information alone. Position and brightness gates are applied to each combination of a track and an RSO candidate, and assignment is performed using the Hungarian algorithm, with the Mahalanobis distances of the combinations that satisfy both gates used as costs. Moreover, the initial states and error covariances of new tracks are estimated by applying a least-squares method to observations over multiple consecutive frames. The effectiveness of the proposed method is evaluated using simulated STT images containing multiple RSOs and other light sources such as stars. The evaluation examines the basic multi-RSO tracking capability of the proposed method and compares its data association performance with Euclidean distance-based association, as well as with and without the brightness gate. Preliminary simulation results show that the proposed method can continuously track multiple RSOs simultaneously present in the STT field of view. Under the preliminary evaluation conditions, the track identities of the RSOs were maintained, and no ID switches were observed. The proposed method is expected to be applicable to stable multi-RSO tracking in onboard optical observations using STTs.

This paper presents a vision-based perception pipeline and a Ground Obstacle Emergency Braking System for uncrewed aircraft, designed to work within all phases of aircraft operation. Taxiing, takeoff roll, and landing rollout are routine yet demanding phases that contribute significantly to mission success, yet they have received far less attention than the surrounding flight phases. Autonomy on the ground rests heavily on perception: the aircraft must detect, classify, and track other aircraft, ground vehicles, and personnel who share the apron, taxiways, and runways. In the event of a potential collision, a brake command must be applied. Three main challenges shape the task and motivate the approach: general-purpose and automotive obstacle detection systems do not transfer to the airport surface, labeled airport imagery is scarce, and a mismanaged braking response would disrupt the ground-operation sequence. The work presents a solution to the problem by proposing and integrating two intelligent systems. On the one hand, a vision-based obstacle perception module has been developed that detects, classifies, and tracks objects common to airport environments. A curated corpus of airport imagery from several heterogeneous and unevenly labeled public sources has been assembled, unifying them into a single format and labeling them automatically with the state-of-the-art foundation segmentation model SAM3. A YOLO detector has been fine-tuned and specialized to three airport-relevant classes, i.e., aircraft, person, and ground vehicles, with the source mixture studied systematically to identify the combination that generalizes best across domains, reaching an mAP50 score of 0.537 on the "Computer Vision Dataset for Aircraft Taxi Operations" produced by NASA. To enable real-time obstacle tracking, each frame's raw detections pass through lightweight filters that discard physically implausible detections, with temporal continuity enforced through motion- and appearance-information-based trackers. The obstacle perception module, trained on a data source mixture generalized across unseen domains the best, where single-source models did not, and was carried into deployment. It recovers the range and bearing of each obstacle relative to the aircraft using a depth estimation module. On the other hand, an emergency braking system has been developed that automatically judges if the obstacles visible in the scene require a braking command. The deployed system anticipates impacts through a constant-bearing-decreasing-range criterion, and dynamically interacts with a speed controller. Landing and rejected takeoff regimes are handled conservatively, since a missed obstacle can be catastrophic. Low-speed taxiing is also managed, aiming to avoid needless stops that would disrupt ground operations. The performance of the system has been demonstrated end-to-end in a MATLAB/Simulink highly accurate multibody simulation environment, using X-Plane as a graphical engine. The system reliably brings the aircraft to a stop, with obstacles detected up to 553 meters, while leaving the planned taxi undisturbed in cluttered scenes and under degraded visibility. The work thus brings machine-learning perception to the ground phases of flight, tackling the scarcity of labeled airport imagery, fine-tuning detectors specifically for the airport surface, and validating the system in a closed loop with the aircraft's longitudinal control for safe, camera-only ground operation.

Satellites are an appealing sensing modality to track ships after they stop transmitting Automatic Identification System (AIS) reports because of their global coverage. However, they require scheduling over uncertain, changing locations with constraints on access, illumination, and delivery times. In addressing these challenges, we make four contributions. First, we define a forecast model that combines route hypotheses with persistent off-route motion constrained to water. Route motion includes stops and correlated lateral uncertainty, while off-route motion includes random heading changes. Detections localize the ship, while informative misses reduce probability where detection was possible. Second, we formulate scheduling as minimization of expected position entropy integrated over a planning horizon. Entropy is measured in nats, and its time integral in nat-hours, so that an image delivered earlier reduces uncertainty over a longer part of the horizon. The formulation accounts for tasking lead, delivery delay, orbital access, daylight, and cloud obstruction. Each decision uses only delivered outcomes. Third, we solve the selection problem through sequential linearization with mixed-integer linear programming subproblems and hourly replanning. Fourth, we evaluate twelve SkySat optical and nine Capella synthetic aperture radar (SAR) satellites on a synthetic, 36-hour AIS-silence scenario in the Black Sea and the Turkish Straits. We compare six-hour scheduling, a myopic policy, and prediction without imaging under three sensor and weather settings. In the cloudy case, six-hour scheduling reduces integrated posterior entropy from 154.89 to 39.82 nat h and mean position error from 31.64 to 0.99 km, against 43.41 nat h and 3.53 km for the myopic policy. In clear sky with lower SAR detection probability and equal delivery delays, both policies reduce integrated entropy by about 68%.

Closely spaced object tracking requires a tracker to main- tain continuity through missed detections, ambiguous associations, false detections, and close or crossing trajectories. This work eval- uates a reinforcement learning (RL) multi-agent track-management framework against two Track-Oriented Multiple Hypothesis Track- ing (TOMHT) references: a project-specific Python implementation and MathWorks trackerTOMHT used as an external reference point. The RL tracker couples a Kalman-filter state estimate with a Prox- imal Policy Optimization (PPO)-trained policy that selects track- management actions including update, coast, reinitialize, and termi- nate. Monte Carlo testing was performed across a synthetic scenario bank using common detector realizations and metrics including Multiple Object Tracking Accuracy (MOTA), fragmentation, purity, probability of track, probability of false track, timeliness, and RMS localization error. Reward-policy studies show that moderate gap- tolerant coasting can substantially reduce RL fragmentation without requiring aggressive reinitialization penalties. The two TOMHT references exhibit different strengths: the in-house tracker maintains very low fragmentation but produces many false tracks, while the MathWorks reference provides much stronger MOTA and false- track suppression but higher fragmentation on the present test bank. Current RL policies occupy a competitive region of this trade space. Ongoing candidate-list development further indicates that remaining RL failures are increasingly associated with association ownership rather than track-management reward selection alone.

Satellite tracking systems commonly employ the monopulse technique due to its high angular accuracy and relatively simple implementation for estimating target position from received RF signals. However, RF tracking performance may degrade under adverse conditions such as multipath propagation, atmospheric disturbances, and rapid target dynamics. This can lead to pointing errors and reduced communication quality. Additionally, ground stations operate in a wide range of environments, including land and maritime applications, and use antennas of different sizes, making robust and autonomous tracking strategies essential. This paper presents a hierarchical estimation and data fusion framework to improve the performance and autonomy of satellite tracking systems. First, a state estimation technique is employed to enhance the RF tracking measurements by reducing the impact of measurement noise and propagation-induced disturbances, providing a more reliable estimate of the target trajectory. Subsequently, information from the ephemeris and RF tracking modes is combined through a multiple-model data fusion architecture to exploit the complementary characteristics of both tracking modes. While the ephemeris mode provides smooth trajectory predictions, it may suffer from orbital prediction errors, bias, and time delays. Conversely, the RF tracking mode provides real-time corrections but is more sensitive to measurement noise and propagation effects. By adaptively combining both information sources according to the operating conditions, the proposed framework compensates for their individual limitations and provides a more robust and accurate target position estimate. The proposed approach enables the development of an autonomous tracking mode capable of maintaining accurate antenna pointing and maximizing received signal quality under varying operational conditions. The proposed framework is validated using experimental data acquired during real satellite tracking tests, demonstrating improvements in trajectory estimation accuracy, robustness against measurement disturbances, and overall tracking performance compared to conventional single-mode tracking approaches.

Maintaining continuous custody of high-dynamics targets is a persistent challenge in aerospace surveillance, spanning aviation conflict monitoring and ballistic trajectory tracking. Building on the earlier GoDSAT reinforcement-learning framework for satellite sensor tasking, this paper introduces a hybrid agentic architecture that layers decentralized per-satellite policy reasoning over deterministic safety-constrained tracking control. Agents operate through a bounded observe-reason-validate-act loop, inferring urgency and ownership context from shared state, proposing local actions, and executing only those actions that satisfy line-of-sight, field-of-view, and slew constraints. Coordination emerges through a shared ownership registry and urgency map rather than direct peer-to-peer messaging, preserving auditability for safety-critical missions. Across 100 independent seeded experiments the hybrid framework produces statistically significant gains of +6.57 percentage points in tracking coverage (95% CI [5.80, 7.33]) and +4.33 percentage points in strict custody rate (95% CI [3.85, 4.81]). An ablation study isolates the contribution of ownership stickiness, urgency-aware prioritization, and bounded slew continuation. The results indicate that hybrid agentic coordination can improve constellation-level continuity while retaining interpretability and deterministic safety compatibility.

Automated maneuver detection helps prevent orbital collisions and maintain accurate tracking catalogs as the number of objects in low Earth orbit rises. However, geomagnetic storms can expand the thermosphere and increase atmospheric drag, causing orbital decay that resembles the signature of an intentional propulsion maneuver. The purpose of this study was to determine whether geomagnetic activity increases the false-alarm rate of a standard two-line element (TLE) based maneuver detector, and whether a storm-adjusted detection threshold could restore quiet-time performance. A negative control population of 396 non-maneuverable objects, restricted to debris and rocket bodies between 300 and 600 km altitude, was assembled from the public Space-Track catalog. Because these objects have no propulsion, every flagged maneuver indicated a false positive. The mean motion was converted to a semi-major axis, and a first-degree polynomial was fitted to each ten-element window to predict the expected value at each epoch. Residuals were normalized through a median absolute deviation scale estimate, and epochs above three scale units were flagged. All 281,277 epochs from the 2024 orbital histories were matched to the concurrent planetary K-index and categorized by activity level. The 2024 study period includes the May 2024 Gannon Storm, the most severe geomagnetic storm in twenty years. The false-alarm rate increased from 6.1% under quiet conditions to 55.5% under severe storm conditions, with standard errors below 0.009 in every group. Intermediate groups fell between these values, at 13.3% under active conditions and 24.3% under storm conditions. Holding the false-alarm rate at its quiet-time value required raising the detection threshold from 3.0 to 15.3 scale units. The recovered thresholds increase monotonically with activity level, from 4.66 under active conditions to 7.20 under storm conditions, yielding an empirical threshold schedule k(Kp). A per-object case study during the Gannon Storm confirms this mechanism. Storm-influenced decay in the semi-major axis produces residuals that exceed the quiet-time threshold, generating flags that are unambiguously false positives. These findings suggest that scaling detection thresholds to geomagnetic activity may improve the reliability of automated conjunction screening during periods of severe space weather. Future research will extend the current framework to maneuvering satellites with published maneuver logs, enabling a receiver operating characteristic analysis of the sensitivity cost of storm-adjusted thresholds.

Reliable nonlinear state estimation becomes challenging when deployment-time disturbances differ from the white-Gaussian noise assumptions adopted in the nominal system model for state estimation. This paper presents the Kalman N-Filter, a physics-guided, self-supervised variational estimator trained under a nominal white-Gaussian system model and designed to remain effective under deployment-time noise mismatch. Unlike many learning-based estimators, the proposed network requires no ground-truth state trajectories for training; instead, measurement consistency, dynamics consistency, and posterior regularization are constructed from noisy measurements and nominal physical models. For deployment on the actual nonlinear system, we establish a conditional mean-square error bound for the posterior-mean estimate under explicit regularity, bounded-disturbance, residual-boundedness, and input-to-state stability (ISS)-type conditions. In simulated nonlinear constant-turn-rate-and-velocity (CTRV) radar-tracking experiments, when the network is trained using the nominal model and measurement data corrupted by white Gaussian noise, the Kalman N-Filter achieves a root-mean-square error (RMSE) of 0.582 m under first-order autoregressive (AR(1)) process noise, corresponding to a 52.5% reduction relative to the unscented Kalman filter (UKF), and remains comparatively stable as temporal correlation increases. On the real-world North Campus Long-Term (NCLT) GPS/IMU dataset, it achieves a mean RMSE of 4.55 m across 20 trajectory segments, compared with 4.84 m for the UKF, without retraining on the real test data.

Modern spacecraft flight software relies on stateless real-time operating systems that cold-boot after any power interruption, discarding all runtime context and learned behavioral parameters. We present ÆON, a persistent cognitive operating substrate that replaces the traditional RTOS with a three-layer architecture: a non-volatile-memory-resident microkernel (the Pulse) that resumes execution instruction-accurately after power loss; an onboard cognitive engine (the Mind) that performs continuous on-device learning; and an autobiographical memory graph (the Self) that accumulates mission-long experience. A capability-based Mesh secures all inter-component communication with cryptographic tokens, eliminating ambient authority. A prototype was implemented on an ATmega328PB microcontroller (Arduino Uno R3, 16 MHz, 2 KB SRAM, 32 KB Flash) with an external AT24C256 I²C EEPROM serving as non-volatile main memory. Measurements were collected using an external ESP8266 NodeMCU configured as a microsecond-accurate hardware interrupt timer. Power interruptions were induced via physical disconnection of the 5 V supply—a hard fault model representative of unexpected power loss on spacecraft. Four experimental conditions were characterized, each with 20 valid trials. With BOOTRST disabled, ÆON achieved a mean recovery latency of 166.652 ms (σ = 26.60 µs), and with BOOTRST enabled, 166.867 ms (σ = 129.95 µs). The corresponding FreeRTOS v10.4.3 baseline measured 161.568 ms (σ = 132.20 µs) with BOOTRST disabled and 161.811 ms (σ = 155.36 µs) with BOOTRST enabled. The BOOTRST effect was approximately +0.2 ms in both systems, while the ÆON–FreeRTOS difference remained approximately +5.1 ms under both configurations, indicating that the additional recovery cost arises primarily from ÆON's persistent-state recovery and application initialization rather than from the bootloader configuration. Memory footprint: the ÆON persistent kernel requires 10,364 bytes of Flash and 513 bytes of SRAM, compared to 8,070 bytes of Flash and 353 bytes of SRAM for the FreeRTOS baseline performing the same sensor task without persistence—an overhead of +2,294 bytes of Flash (+28.43%) and +160 bytes of SRAM (+45.33%) . Both values remain within the available ATmega memory, with Flash utilization at 32.13% and SRAM utilization at 25.05%. Model fidelity retention: the learned anomaly threshold (θ) was initialized to 8.0000, and the device was power-cycled 10 times. Across all trials, θ exhibited zero numerical drift, with each cycle returning the exact reference value after persistent-state restoration from the external EEPROM. These results demonstrate that persistent execution, continuous on-device learning, and capability-based security can be integrated into a single operating substrate on severely resource-constrained flight hardware. ÆON therefore provides a foundation for fault-tolerant, self-optimizing avionics in next-generation small satellites and deep-space missions.

Modern synthetic aperture radar (SAR) payloads generate large volumes of raw echo data, making timely image product delivery increasingly challenging under finite on-board storage, intermittent ground-station access, and limited downlink bandwidth. Frequency-domain SAR image formation relies on repeated fast Fourier transform (FFT) and inverse FFT operations across the range and azimuth dimensions, separated by corner-turn data reordering. Consequently, on-board SAR processors require compact, low-power, and high-throughput FFT accelerators that satisfy stringent size, weight, and power (SWaP) constraints. Moreover, SAR missions often trade image resolution, scene coverage, and revisit rate, allowing shorter range or azimuth processing windows that reduce FFT size and expose additional opportunities for parallel processing. This work presents a SAR-oriented stream-scaling FFT core that exploits these reduced-size workloads to improve throughput while maintaining a compact hardware footprint. The proposed architecture is based on a fixed 32-lane, 32×32 Cooley–Tukey decomposition comprising a fully pipelined 32-point FFT, a banked streaming transpose stage, a compensated twiddle-rotation stage, and a second pipelined 32-point FFT. The 32-point sub-FFTs employ multiplierless fixed-twiddle rotators, while the transpose stage uses 32 local memory banks to reorder data efficiently between the two FFT dimensions. Unlike conventional multi-mode FFT accelerators that simply execute shorter transform schedules for reduced FFT sizes, the proposed architecture dynamically repartitions the same physical 32-bank, 32-lane fabric into two independent 16-lane FFT streams for 256-point operation or four independent 8-lane FFT streams for 64-point operation. This stream-scaling capability enables concurrent processing of multiple range or azimuth lines, significantly improving aggregate throughput instead of leaving hardware resources underutilized. A key challenge in supporting multiple FFT sizes is accommodating different twiddle-factor sets without increasing coefficient storage or introducing long-distance coefficient distribution. Rather than storing separate twiddle tables for each transform size, each local bank stores only the twiddle coefficients required for 1024-point operation. For 256- and 64-point modes, the controller selects reusable entries from the existing local table and applies a lightweight local phase correction whenever the required reduced-size twiddle is not directly available. This storage-efficient twiddle reuse scheme preserves mathematically exact FFT outputs while eliminating coefficient-memory growth and minimizing coefficient movement across the chip. The proposed FFT core is verified on an FPGA prototype against a Python reference for 1024-, 256-, and 64-point operation and is implemented in a 28-nm CMOS process. It operates at 500 MHz, sustains 32 complex samples per cycle, and achieves an aggregate throughput of 16 GS/s while consuming 65 mW under random input activity, corresponding to an energy efficiency of 4.06 pJ/sample. As a first silicon step toward longer SAR FFT workloads, the proposed 1024-point architecture serves as a reusable Cooley–Tukey leaf tile that can be hierarchically extended toward million-point FFTs through additional inter-tile transpose and twiddle stages, providing a scalable foundation for future high-resolution on-board SAR processors.

The ever-growing demand for efficient AI and Big Data processing has fueled a rapid development of new hardware architectures specifically designed for compute-intensive tasks. Alongside these advancements, software solutions are emerging to leverage this specialized hardware by offloading tasks. However, proprietary software often presents a significant learning curve for users, hindering adoption and flexibility. This paper proposes OFFLOAD, an open source hardware-agnostic software hardware framework. OFFLOAD distributes tasks to various hardware units, including both novel accelerators and existing system-on-chip (SoC) architectures. Our framework seamlessly interfaces with popular databases and application development tools. Through multi-level abstractions at the compiler, operating system, and driver levels, OFFLOAD translates high-level code and data into hardware-optimized binary instructions for open architectures. To our knowledge, OFFLOAD represents a novel approach in this domain. We demonstrate the feasibility of OFFLOAD by integrating it with popular tools like MySQL, Apache Spark, and Apache Arrow in Python at the user level. Tasks are then offloaded for execution on hardware using memory-mapped I/O. Raspberry Pis serve as examples in this demonstration, showcasing the entire workflow from software based data query to hardware execution. To underscore the framework’s efficiency, we performed a predictive analysis with our Raspberry Pi based prototype, revealing that it can handle 10,000 queries in just 39.73 milliseconds using 256 worker nodes.

The performance of computing systems deployed in space does not solely depend on hardware, but also on the chosen software application. Ideally, space systems opt for a hardware-software co-design approach and target any issues that arise. Orbital Edge Computing (OEC) seeks to provide missions with high-performance computing capabilities that can be deployed in space. This technology has the ability to run Artificial Intelligence (AI) software as a stand-alone system without the need for terrestrial communications or cloud computing support. In addition, OEC intends to optimize the use of resources by focusing on low size, weight and power (SWaP) characteristics. Emerging hardware architectures focusing on optimal AI computation promise advantageous SWaP but also bring their own list of challenges to consider. This work describes lessons learned at the Space Architecture, Characterization, and Evaluation Research (SPACER) laboratory when assessing the possible use of hardware-software AI systems for space deployments. We compile and discuss experimental results from graphic-processing unit (GPU), neuromorphic, and analog hardware while executing relevant AI algorithms for satellite imagery and space domain awareness applications. Suboptimal behaviors observed in these systems might have been triggered by radiation while others are more systemic and include time varying throughput, temporal changes in performance, forgetting weights, temporal drifts, sudden catastrophic performance degradation, and incomplete recovery. Some GPU-based OEC results have demonstrated orbital viability with increasing number of flight heritage examples appearing in the news, but they come at the expense of power consumption. Neuromorphic hardware reduces energy needed per AI inference, but recent radiation testing experiments have revealed specific challenges these systems need to overcome. Analog computing also consumes low energy on each inference, but problems with temporal drift and long reprogramming times limit system availability and must be addressed before deploying in space. This work provides an overview of technical strengths and vulnerabilities of hardware-software AI systems for space applications. The purpose of this paper is to inform the aerospace sector of these important issues to consider in this era of AI productivity, commercialization, and deep-space exploration. We thank the AFRL Information and Spectrum Warfare Directorate for providing guidance support to make this possible. We also thank the AFRL Space Warfare Directorate's SPACER laboratory for providing infrastructure to host the equipment necessary for data analysis. The views expressed are those of the author and do not necessarily reflect the official policy or position of the Department of the Air Force, the Department of Defense, or the U.S. government. Distribution Statement A: Approved for Public Release. Distribution is Unlimited. Public Affairs Release Approval # AFRL-2026-2835.

The Lunar Environment Monitoring Station (LEMS) is a compact, autonomous seismometer suite designed for deployment near the lunar South Pole to provide continuous, long-duration monitoring of the Moon’s seismic environment. By recording ground motion from moonquakes and meteorite impacts, LEMS will characterize the regional crust and mantle, improving models of lunar formation and evolution while informing the geophysical environment for future lunar surface operations. Following successful environmental qualification and mission testing, this paper presents application-driven lessons learned from implementing onboard seismic science processing on a CPU- and I/O-constrained flight computer. LEMS is built around a single-soft-core LEON3 processor operating at 40 MHz with 16 MB of RAM and 16 GB of flash storage. Seismic data are sampled at 50 Hz on a daughter card and transferred to the processor card over an SPI bus. Consequently, system performance is constrained not only by processor throughput and memory capacity, but also by I/O bandwidth, bus timing, storage management, and changing resource priorities across autonomous mission modes. The principal lesson is that onboard science processing cannot be optimized in isolation; it must be designed and verified as part of an end-to-end, resource-constrained system. In real-time embedded systems, processor scheduling, data movement, storage, communications, and mission autonomy must be integrated and evaluated as a complete architecture. To reduce processor load, a finite impulse response (FIR) filter and sample-conditioning functions were implemented in FPGA, offloading repetitive signal-processing operations from the LEON3. The processor was reserved for supervisory control, flight software execution, telemetry and command handling, communications, data packaging, compression, storage coordination, health monitoring, and fault management. Autonomous mission modes were designed to execute without ground intervention. Seismic waveform data are compressed using the STEIM2 algorithm, an established community standard that significantly reduces flash storage and downlink bandwidth while preserving compatibility with existing seismic analysis workflows. The paper also describes implementation strategies within NASA’s core Flight System (cFS) running on an open-source real-time operating system. A child-task I/O architecture isolated FPGA interactions, device transfers, and other potentially blocking operations from the primary scheduler, enabling deterministic execution on a single processor. Static buffer ownership, bounded queues, and minimized memory copies were essential to sustaining continuous 50 Hz acquisition while managing large data volumes. Command security, implemented using CryptoLib and CCSDS Space Data Link Security concepts, was treated as an additional system constraint because authentication and encryption introduce processing overhead, communication latency, and operational complexity. System verification emphasized sustained data acquisition, processor utilization, SPI and I2C bus performance, buffer management, compression efficiency, command latency, and integration observability. The resulting lessons demonstrate that substantial onboard science capability can be achieved on CPU- and I/O-limited flight computers through careful co-design of FPGA acceleration, efficient compression, disciplined cFS tasking, and comprehensive consideration of security and resource constraints across the entire system.

EagleCam 2 builds upon the original EagleCam mission's pioneering objective of capturing third-person views of an extraterrestrial lander during its descent and surface operations. Post-mission analyses of previous iterations identified critical constraints in telemetry capacity, software reliability, and real-time image processing, which collectively restrict the retrieval of high-fidelity visual data over limited communication links. To directly address these bottlenecks, EagleCam 2 implements a robust telemetry prioritization strategy centered entirely on intelligent onboard data processing. The core of this architecture is a software pipeline that autonomously compresses large, uncompressed RAW images and encoded video arrays into low-resolution, tiled thumbnails prior to initial transmission. This data reduction strategy largely decreases the initial bandwidth footprint, enabling a highly efficient human-in-the-loop downlink process. During communications, ground operators quickly review this thumbnail telemetry to evaluate spatial context, illumination, and overall image quality. Based on this assessment, operators selectively command the targeted downlink of specific, high-priority full-resolution files. By dedicating strictly limited bandwidth exclusively to operator-selected data, this approach prevents the transmission of redundant or compromised imagery, thereby optimizing downlink capacity and significantly improving the retrieval speed of high-quality mission data. To maintain high throughput, the payload architecture utilizes GPU-accelerated video encoding and native onboard JPEG compression. This hardware utilization ensures that large data arrays are processed rapidly and allows the system to dynamically re-encode video onboard to accommodate unexpected, degraded downlink conditions. Concurrently, command uplink interfaces allow operators to adjust camera capture configurations in-flight based on their initial evaluation of the thumbnail views. Finally, this flight architecture is comprehensively supported by custom ground-support tools developed to validate software resilience to injected faults and to provide rigorous operational training for the human-in-the-loop protocol, ensuring reliable data management during constrained communication windows.

An orbital data centre is a data-management problem before it is a compute problem. Earth-observation constellations already generate more data than they can return to the ground, the links that carry it are intermittent by construction, and onboard storage cannot be repaired. This paper specifies the data platform for a constellation that performs time-critical analysis in orbit, using autonomous wildfire detection as the driving mission. The platform organises all data movement into six explicit zones — an input/output zone providing partitioned, atleast- once event streaming over a delay-tolerant transport; a computer-vision zone; a machine-learning zone; and storage, caching and lifecycle zones — under one structural rule: zones never call one another directly, so every cross-zone interaction is a durable, replayable, attributable event. Within that substrate we develop the parts that an intermittently connected system depends on most. A three-tier cache hierarchy spanning the sensing satellite, the orbital aggregation node and the sovereign ground core is specified by what each tier holds, how it is filled by prefetch from the contact plan, and what happens on a miss; invalidation is by locally evaluated staleness budget rather than by message, consistency under at-least-once delivery rests on content-hash idempotency, and eviction follows data class and mission value. A feature layer with a single definition for training and serving, point-in-time joins and declared degradation makes a disconnected cache honest rather than merely available. Delay-tolerant transport with contact-graph routing, an optical inter-satellite mesh with a predictively armed radio-frequency fallback, corridor-cropped and delta-encoded uplink under congestion, and standards-based external integration complete the design. A five-level data classification binds residency, key custody and derivation rights to every object and to every cached copy of it. We state the metrics and baselines under which the platform should be evaluated, and the conditions under which its central claims would fail.

The proliferation of small satellites and mega-constellations has intensified demand for autonomous operations that reduce reliance on ground control. In response, AI at the edge is being integrated into imaging satellites for real-time onboard Earth observation tasks, including surveillance, environmental monitoring, and disaster response. As these edge-computing applications evolve, computational demands grow correspondingly to support mission adaptability. Among available edge-computing platforms, Field-Programmable Gate Arrays (FPGAs) are particularly well suited for edge acceleration as they offer a favourable balance between reconfigurability, low latency, and performance-per-watt — unlike ASICs, which achieve higher power efficiency but lack reconfigurability, and GPUs, which offer flexibility but at lower performance-per-watt. Widely adopted for images, signal and sensor data processing, these attributes make FPGAs especially critical for small satellites operating under stringent size, weight, power, and cost (SWaP-C) constraints. In a conventional workflow, a static hardware design is loaded onto the FPGA's Programmable Logic (PL). FPGAs with limited PL resources present a bottleneck, making it difficult to accommodate large designs and thereby constraining deployment at the edge. Furthermore, to support mission adaptability, AI/ML workloads at the edge require periodic updating. Under a static design framework, any such change necessitates loading a complete bitstream, which interrupts FPGA operation. Dynamic Partial Reconfiguration (DPR) overcomes this limitation by allowing selected FPGA regions to be updated while the rest of the system remains operational. Although its potential has been explored in domains ranging from data centers to edge devices, prior studies have primarily demonstrated feasibility, with limited focus on optimization techniques to improve energy efficiency, throughput, and PL area utilization. In this paper, we present an energy-efficient reconfigurable edge compute engine leveraging two key techniques: 1) limiting the active region on the PL fabric, and 2) pipelining tasks at the block level. In the first technique, heterogeneous reconfigurable slots are defined on the PL, and complex computational workloads such as Convolutional Neural Networks (CNNs) are partitioned into discrete subtasks assigned to these slots. In the second technique, an Integer Linear Programming (ILP) solver maps subtasks to available slots, while block-level pipelining of consecutive tasks further improves throughput. After each pipeline stage completes, intermediate results are stored in off-chip memory, the corresponding FPGA slots are reconfigured via ARM-controlled Processor Configuration Access Port (PCAP), and the process repeats until all subtasks are executed. To validate the proposed techniques, a MobileNetV1Lite network for onboard marine vessel detection is implemented and trained on the MASATI-v2 dataset. CNN kernels are developed using high-level synthesis (HLS) tools. Implementation results on the Kria KV260 platform demonstrate that the proposed approach achieves approximately 2.75× lower power consumption with only an 8% increase in inference time compared to a static implementation, while maintaining an F1-score of 0.867. These results highlight a practical path toward scalable, power-efficient, and adaptive edge AI accelerators for next-generation autonomous satellite missions. The full paper presents the formulation, methodology, system design, hardware implementation, and comprehensive power and performance benchmarking, demonstrating that energy-efficient reconfigurable edge engines are a promising solution for resource-constrained small satellite applications.

High-speed onboard data-handling systems must route bursty payload, telemetry, and processor traffic through FPGA switching and buffering structures while operating within constrained hardware resources. This paper examines the arbitration stage of an onboard switching-cache design whose data path is organized around an 8-by-8 crossbar. Each output port uses an 8-way round-robin arbiter to resolve input contention. Although the baseline arbiter is simple and provides long-term service fairness, its decisions primarily depend on input requests and data-valid signals. It does not explicitly account for downstream output readiness, backpressure, output-buffer high-water conditions, or input waiting time. As a result, it can issue selections that cannot produce effective transfers while an output is unavailable. This paper presents a flow-control-aware round-robin arbitration method for the crossbar of an onboard FPGA switching cache. The method retains the baseline round-robin structure and adds output-availability gating, high-water request compensation, and long-wait request compensation. Output availability is determined from ready, backpressure, and output-buffer high-water signals. When an output is unavailable, the arbiter returns an invalid-selection code. When it is available, requests associated with high buffer occupancy or a wait counter above a configurable threshold are selected from a boosted request set using round-robin arbitration; otherwise, selection proceeds from the normal request set. The method also includes protection against a selected request being withdrawn before data becomes valid. The method is evaluated through lightweight arbitration simulation, local VHDL validation in Vivado Simulator, staged crossbar integration checks, and module-level synthesis estimates. In the modeled backpressure and burst-recovery scenarios, invalid grants decrease from 219 and 250, respectively, with the baseline arbiter to zero with the proposed method. Maximum waiting time also decreases in selected contention scenarios, while throughput is unchanged in the reported simulations. The candidate arbiter adds an estimated 95 Slice LUTs and 47 Slice Registers, with no additional BRAM or DSP usage. These results indicate that flow-control information can improve arbitration behavior with a modest module-level resource cost. Full replacement of all eight output arbiters and full-design timing closure remain future work.

Deep neural networks (DNNs) are continuously improving in performance but at the same time increasing in size and computational complexity. This makes them more challenging to deploy on size-, weight-, and power-constrained platforms, such as spacecraft. Typical inference for DNNs relies on GPUs that use dense computation, leading to high energy consumption. While exploiting sparsity is a promising method for reducing energy and latency, most prior work has focused on weight sparsity methods like pruning. This research instead focuses on reducing the number of non-zero activations in DNN models to achieve activation sparsity, serving as the foundation for a custom, sparse FPGA inference engine that skips multiply-accumulate (MAC) computations on zero-value activations. Such an implementation could provide a power-efficient method of performing neural network inference on space systems through existing radiation hardened/tolerant FPGAs and system-on-a-Chip devices such as the Microchip RTG4, Microchip PolarFire, and AMD Versal. We investigate whether activation sparsity can be induced in Convolutional Neural Networks while minimizing accuracy loss. Unlike previous works that target classification networks for sparse activation, this research applies activation sparsification to both image classification and object detection. We evaluate the ResNet-50 and YOLOv8 architectures for a maritime awareness application: ship detection. Furthermore, we expand thresholding to the SiLU activation function in YOLOv8, alongside ReLU in ResNet-50, by introducing a per-layer threshold value that forces certain activations to zero. This research evaluates two methods of selecting these thresholds. The first is a per-layer method that independently selects the largest possible threshold for each layer that keeps the accuracy loss within an acceptable margin. The second method, a novel approach, accounts for cumulative accuracy degradation by sequentially determining the thresholds. For both methods, models are retrained with the thresholded activations to recover accuracy. The baseline ResNet-50 model, without any sparsity methods applied, inherently exhibits an activation sparsity of 50.19%, leading to a ~60% theoretical MAC savings. The baseline YOLOv8 model, however, has essentially no initial sparsity, offering significant room for optimization. Both sparsification methods with retraining yielded substantial activation sparsity and theoretical MAC reductions. For ResNet-50, the per-layer method increased activation sparsity to 79.61% and MAC savings to 87.93% with a 2.19% relative accuracy drop. The cumulative method achieved 77.39% sparsity and 84.69% MAC savings with a minimal relative accuracy drop of 1.78%. In contrast, YOLOv8 was more sensitive to thresholding with the SiLU function. The per-layer method achieved 44.41% sparsity and 37.80% MAC savings with a relative accuracy drop of 8.34%. The cumulative method achieved 33.70% sparsity and 27.28\% MAC savings with a small relative accuracy drop of 3.81%. These two threshold-based techniques successfully induce sparsity in ResNet-50 and YOLOv8 models, opening the door for reduced power and latency during DNN inference on edge hardware. The methods strike a strong balance between increased sparsity and preserved accuracy. Neither method is superior; rather, each serves a specific optimization objective. Per-layer thresholding maximizes compute savings at the cost of higher accuracy loss, whereas the cumulative method minimizes accuracy loss while still providing sparsity gains.

Future long-duration and deep-space missions will increasingly require spacecraft to maintain mission capability despite radiation-induced faults, progressive hardware degradation, changing power availability, thermal constraints, communication delays, and evolving computational demand. Current fault-tolerant avionics primarily preserve predefined hardware and software configurations through redundancy, fault detection, isolation, recovery, and limited task reallocation. Emerging onboard computing platforms provide substantially greater processing performance and reconfigurability, but the spacecraft computing architecture itself generally remains a configuration established before launch. This paper proposes a Health-Aware Adaptive Avionics Architecture (HAAA) in which the instantaneous health of spacecraft computing resources becomes an explicit input to onboard architectural reconfiguration. The principal contribution is a fault-governed framework that combines a protected deterministic safety core with a heterogeneous pool of processors, reconfigurable logic, intelligent accelerators, and distributed subsystem computing resources. A continuously updated computational-health model estimates the availability, degradation state, fault history, timing capability, power cost, thermal condition, and environmental exposure of these resources. Based on mission criticality and current vehicle state, an independent reconfiguration manager selects among pre-verified computing topologies and redistributes eligible functions while preserving deterministic command authority and fault-containment boundaries. Unlike conventional FDIR approaches that primarily recover failed components or restart predefined functions, the proposed architecture treats computing topology as a controlled spacecraft resource. A mission function may therefore migrate between heterogeneous processing domains, execute with reduced redundancy, invoke an alternative hardware accelerator, or transition to a degraded implementation when its original computational path becomes unsuitable. Reconfiguration is restricted by a formally defined safe adaptation envelope, preventing autonomous optimization from modifying safety-critical control functions or selecting configurations outside verified timing, power, reliability, and interface constraints. The architecture introduces three coupled mechanisms: a spacecraft computational-health state model, a mission-aware resource allocation function, and a runtime assurance layer capable of approving, rejecting, or reversing architectural transitions. A practical validation methodology is proposed using heterogeneous processor/FPGA hardware, representative spacecraft workloads, injected processor and memory faults, simulated radiation events, thermal and power constraints, and controlled degradation scenarios. Performance is evaluated against static and conventional redundancy-based architectures using mission availability, function recovery time, computational latency, power consumption, fault propagation, and retained mission capability. The proposed approach shifts spacecraft fault management from preserving individual computing elements toward preserving mission-level computational capability. If experimentally validated, the architecture could provide a practical foundation for resilient autonomous avionics capable of operating for extended periods with reduced dependence on ground intervention, particularly for deep-space exploration, distributed spacecraft, orbital infrastructure, and other missions in which repair, replacement, or continuous ground supervision is unavailable.

Earth-observation networks are reported at benchmark accuracy and parameter count; neither predicts whether a model will run on the accelerator a CubeSat carries—or whether several will run together. This paper presents a reference architecture for the onboard EO processing chain on one flight-representative module—the processing unit of a 1U remote-sensing instrument, a Zynq UltraScale+ XCZU4EV system-on-module with two physically isolated DPUCZDX8G INT8 cores and an Arm application processor, vendor-tested for vibration, thermal cycling, and radiation. First characterized for a two-model downlink gate, the module is extended to five applications spanning the chain—scene screening, target detection and multi-object tracking, learned and reference-guided compression, rule-based downlink gating, and SAR-guided restoration of cloud-contaminated optical imagery. Four already execute on the module; the fifth is taken through the same flow. We claim integration and measurement, not board design. The architecture has three layers: a ground flow trains in FP32, quantizes post-training in Vitis AI, compiles to a fingerprint-matched xmodel, and admits a model only if it compiles to a single DPU subgraph—the vendor inspector's report is an insufficient gate. An onboard runtime feeds native-resolution windows to the isolated cores, fuses outputs on the application processor, and retasks the payload by swapping compiled models in a quarter-second; hardware isolation lets a mission model and a data-reduction model run concurrently with bounded interference—detection on one core beside compression on the other. A deterministic integer entropy path with fixed cumulative-distribution tables closes the chain, byte-identical on re-encode on the EM board. A four-step lightweighting checklist for fixed-operator accelerators—judge deployability by compiled subgraph count; prefer exact reformulation; accept an approximating substitute only where none exists, with retraining budgeted and hardware constants recorded; route dynamic-shape operators to the host—derived from the detector, is tested against four networks with disjoint operator families: hyperprior codecs with GDN and sub-pixel layers, embedding heads, and two-stream fusion with skip gating. The result is an operator compatibility matrix where one substitution reverses direction when the target is the Arm cores, not the DPU, grounded in a compile study of EO building blocks, attention and mixture-of-experts included. Latency and energy for all five are consolidated across the flight module, a Kria KV260 (same DPU IP, different host chip), a Versal AI Edge VEK280, and an Orin NX; a deployment-maturity table states per stage what runs on the DPU, on the Arm cores, or only as derived composition. Cross-stage lessons: application-processor pre- and post-processing co-bounds end-to-end time and can flip the optimal DPU configuration; post-training quantization matched FP32 for detection and quantization-aware training added nothing; per-tensor INT8 imposes a rate–distortion wall on learned codecs that widening does not remove; and reference-guided selective coding pays only under a measurable sparse-change condition. The chain targets a 6U mission with an 8.7-minute contact window and 37.7 MB native frames, launching late 2027. We state what remains uncharacterized before flight—codec-specific power, thermal envelope, radiation-induced configuration upsets, a full end-to-end run—so the envelope reads as engineering-model evidence, not flight heritage.

Microwave Kinetic Inductance Detectors (MKIDs) have emerged as leading superconducting detectors for submillimeter and millimeter-wave astronomy because they enable large-format detector arrays with intrinsic frequency-domain multiplexing. However, existing MKID readout architectures become increasingly resource-intensive as detector counts grow, limiting their scalability and suitability for future space-based and large-aperture astronomical instruments. This work presents a scalable, reconfigurable, and ASIC-compatible MKID readout architecture that integrates frequency-comb generation, real-time tone calibration, and spectrometer-based signal processing within a unified framework. The architecture is organized using modular, resource-efficient building blocks suitable for future ASIC implementation while preserving FPGA-like flexibility through hardware reconfigurability to accommodate diverse detector-array sizes and mission requirements. A numerically controlled oscillator (NCO) and a polyphase filter bank (PFB) efficiently generate and process excitation tones while enabling continuous monitoring of detector responses. In addition, an on-chip calibration module dynamically tracks detector resonance shifts and adaptively retunes excitation frequencies during observations, eliminating interruptions and maintaining stable readout performance. A reconfigurable frequency-comb generator enables the same hardware architecture to operate as either a single-channel or multi-channel readout system, maximizing hardware reuse across different instrument configurations. Furthermore, a custom low-complexity fast Fourier transform (FFT) engine significantly reduces computational complexity, hardware resource utilization, and power consumption compared with conventional implementations while maintaining high-throughput operation. Experimental results demonstrate support for up to eight readout channels, each capable of reading 1,000 detectors, allowing a single implementation to interface with up to 8,000 MKIDs. The modular architecture also supports synchronization of multiple readout chips, enabling seamless expansion to detector arrays substantially larger than 8,000 pixels without architectural modifications. The proposed architecture has been implemented on a Xilinx RFSoC ZCU111 platform to validate its functionality and scalability, demonstrating its practicality for high-performance MKID readout systems for submillimeter imaging. By combining ASIC-oriented efficiency with FPGA-like reconfigurability and multi-chip scalability, the proposed architecture provides a practical and power-efficient solution for next-generation large-format MKID arrays and establishes a clear pathway toward low-area, low-power ASIC implementations for future spaceborne astronomical instruments.

As further space exploration and science activity increase, so does the need for power generation and distribution. NASA's Moon to Mars Objectives highlight the need for sustainable power infrastructure for continuous support of robotic and human operations. The work in this paper describes a scalable and adaptable power management and distribution (PMAD) network that can be installed and rearranged for temporary and growing power needs. Demonstrated using NASA’s Automated Reconfigurable Mission Adaptive Digital Assembly Systems (ARMADAS) project, the PMAD network exhibits the ability to provide scalable power distribution capability while also being reconfigurable to fit the changing needs of mission operations. ARMADAS employs an autonomous multi-robot system in conjunction with building-block-based construction, capable of supporting applications such as power and data routing, habitation, and material transport. The system features cooperative autonomous robots that can assemble a modular lattice structure for the lunar surface. This paper presents the design and demonstration of a PMAD system that is integrated into the modular lattice structure and adapts to the growing structural environment. The PMAD system requirements and mission overview are also described. Modified for the ARMADAS lattice structure, an interface integrates into the modular building block architecture for embedded power distribution functionality. This modified building block creates a power distribution voxel (PDV) that can act as an electric conductor. Additionally, the PDV retains its previous attributes of structural integrity, modular assembly, integrated fastening, and robotic collaboration. This expands the capabilities of the present robotic construction agents and ensures smooth integration with the robotic assembly workflow. A PDV segment is created using two PDVs in tandem containing a passive-release and self-retracting cable spool between them, allowing the segment to accommodate variable distances and support reconfiguration or relocation across the structure. The PMAD network consists of segments that span the global lattice structure and connect to local subsystems throughout the lattice. Passive alignment features on the PDV ensure reliable mechanical coupling between PDVs and their interfaces, with embedded electrical connectors, during network assembly. The interface connection facilitates the stable power transfer between PDVs, suited for permanent or temporary connection. The adaptability of the PDVs enables dynamic modifications to the PMAD network as global and local electrical requirements change with new mission objectives. As the PMAD network scales in its complexity and growing power needs, PDVs can be adapted to suit changing electrical loads. With the reconfigurable component of this network, lower load-bearing PDVs can be substituted for higher load-bearing PDVs to ensure suitability for voltage and current requirements of added subsystems across the infrastructure. To validate this network, a proof-of-concept PDV prototype was fabricated and evaluated, and a system feasibility assessment was conducted. Its limitations, scaling considerations, and its application for off-Earth environments were analyzed. This technology demonstration has the potential to standardize PMAD networking systems, utilizing modular components to fully integrate and facilitate systematic electrical connections throughout space structures and beyond.

As small satellite missions grow in complexity and ambition, the need for rigorous early-stage system-level verification has never been greater. Conducting verification activities prior to spacecraft assembly is critical to reducing integration risk, advancing flight software maturity, and ensuring overall mission readiness. The Flat Satellite (FlatSat) has emerged as a widely adopted verification asset, offering a representative spacecraft environment in which hardware and software can be thoroughly validated before flight units become available. Yet the utility of a FlatSat extends well beyond the engineering hardware it hosts—its effectiveness is equally shaped by the design of the verification platform itself and the degree to which that platform can support integration, testing, and mission operations activities throughout the project lifecycle. This paper presents the development and application of an integrated FlatSat verification platform, namely a FlatSat Bench, designed and deployed during the development of a CubeSat mission. The platform was designed as a modular bench environment that accommodates engineering model avionics, subsystem interfaces, power distribution networks, communication links, test equipment, and verification resources within a single, configurable architecture. To reduce harness complexity and improve long-term maintainability, the bench incorporated an internal interconnection architecture based on the spacecraft’s standardized 1U form-factor connectors, allowing avionics modules to interface internally without the need for extensive external cabling. Configurable connection paths further enabled individual electrical interfaces to be selectively engaged or isolated as needed, supporting subsystem-level testing, fault isolation, interface verification, and controlled evaluation of distinct operational configurations. In this way, the bench facilitated concurrent engineering activities across multiple disciplines while providing controlled access to spacecraft hardware for system verification, flight software development, interface validation, and anomaly investigation. The FlatSat Bench supported the complete progression from Engineering Model (EM) verification to Flight Model (FM) Assembly, Integration and Testing (AIT). Over the course of the project, it served as the primary environment for validating subsystem interfaces, verifying command and telemetry functionality, conducting end-to-end functional testing, evaluating flight software updates, and troubleshooting integration anomalies prior to implementation on flight hardware. This approach reduced integration risk, improved test efficiency, and minimized disruption during the critical spacecraft integration phase. The platform also contributed substantially to mission operations readiness. Operational procedures, command sequences, and contingency scenarios were validated on the FlatSat ahead of the Launch and Early Orbit Phase (LEOP), enabling operator training and building confidence in mission execution. The sustained use of a common verification platform across development, integration, and operational readiness phases established traceability between verification activities and reduced the likelihood of late-stage system anomalies emerging during flight. This paper details the design philosophy underlying the FlatSat Bench, its integration into the spacecraft development process, and the lessons learned from its application across multiple project phases. The methodology presented here demonstrated how a well-designed, integrated verification platform can enhance mission readiness—fostering closer collaboration among development, AIT, and operations teams while reducing technical risk at each stage of the spacecraft lifecycle.

This paper presents the design, realization, qualification and performance evaluation of a Discharge Power Supply (DPS) developed for ISRO's first 300mN Stationary Plasma Thruster (SPT)-based Electric Propulsion System (EPS) used in a Technology Demonstration Satellite. The DPS converts a 70 V spacecraft bus to a regulated 350 V discharge supply and delivers up to 7A discharge current. To meet the high-power requirements of the propulsion system while maintaining redundancy and scalability, the DPS employs a dual-module architecture operating in current-sharing mode, enabling a maximum output power capability of 5kW. The paper discusses the converter topology selection, control architecture, compensator design, and current-sharing implementation adopted to achieve stable operation under the highly dynamic load characteristics of the SPT. Particular emphasis is placed on the interaction between the power supply and the plasma discharge process, including the impact of low-frequency breathing mode oscillations on output regulation and loop stability. A compensator tuning methodology is presented to ensure robust performance and stable operation across the complete thruster operating envelope. Significant challenges were encountered during hardware realization and environmental qualification, primarily due to the high-voltage nature of the discharge stage. Multiple thermal-vacuum qualification anomalies necessitated a comprehensive review and refinement of high-voltage design and manufacturing practices. The paper details the lessons learned and the process improvements implemented to mitigate arcing risks, material selection and high-voltage zone management. These measures resulted in reliable operation under thermal-vacuum conditions. Thermal management strategies adopted to maintain acceptable component temperatures while maximizing efficiency are also presented. Experimental results from standalone and integrated thruster testing demonstrate stable discharge regulation, effective handling of plasma-induced oscillations, successful load sharing between the power supply modules, and a peak practical conversion efficiency of 91%. The realized DPS provides a flight-qualified, high-efficiency solution for high-power Hall thruster applications and represents a significant milestone in the indigenous development of electric propulsion power processing technologies for future spacecraft missions

Small spacecraft electrical power systems commonly employ solar charge controller architectures in which maximum power point tracking (MPPT) and battery charging functions are closely coupled to the battery state. Under battery undervoltage, disconnection, or fault conditions, this coupling can disable power conversion and prevent available solar-array power from supporting the spacecraft bus. This paper presents a reconfigurable solar power conditioning unit (PCU) architecture that decouples solar-array power regulation from battery availability and enables a battery-fault-tolerant operating mode for small spacecraft electrical power systems. The proposed PCU is based on an LT8705A 4-switch buck-boost DC/DC converter with digitally programmable feedback control. An STM32L496ZG microcontroller performs supervisory control, ADC-based monitoring of solar-array and battery voltage/current, and incremental conductance MPPT. Two DAC outputs are used to program the converter input and output feedback networks, enabling source-side MPPT regulation and output-side constant-voltage regulation on the same power stage. Together, these modes implement an MPPT-to-CV charging sequence analogous to conventional CC-CV battery charging, while allowing the first charging phase to be governed by solar-array maximum power tracking rather than fixed current regulation. The controller operates in two primary modes: an MPPT mode, in which the solar-array operating point is adjusted to track the maximum power point and deliver charging power, and a CV/EOC mode, in which the output bus is regulated at a predefined end-of-charge-equivalent voltage. The CV/EOC mode allows the bus to remain powered when the battery is fully charged, disconnected, unavailable, or intentionally bypassed. To validate the proposed concept, experimental evaluation was conducted using the Analog Devices DC2285A LT8705A-based buck-boost demonstration platform, an 8-cell lithium-ion NMC battery pack, and an ITECH IT-M3623 programmable power supply operated with SAS1000 solar array simulation software. The simulator was configured to emulate solar-array I–V characteristics and a 90-minute low-Earth-orbit irradiance profile, allowing the measured PCU operating point to be compared with the simulated maximum power point. Experimental validation was performed under constant-power load conditions, including a 100 W load case during MPPT-mode battery charging and a 50 W load case demonstrating mode transition between MPPT, CV/EOC, and MPPT operation. The results show that the DAC-programmed feedback network can command both input-side MPPT regulation and output-side CV regulation. During MPPT operation, the PCU tracks the solar-array maximum power point over varying voltage, current, power, and irradiance conditions. During CV/EOC operation, the controller intentionally departs from the maximum power point to regulate the output bus voltage, demonstrating controlled mode reconfiguration rather than loss of tracking. These results indicate that a reconfigurable MPPT/CV solar PCU can provide a practical fault-tolerant power path for small spacecraft operation, including commissioning, daylight operation, degraded modes, and emergency recovery scenarios. The proposed architecture does not replace energy storage for missions requiring continuous eclipse operation, but it enables battery-absent bus regulation under defined solar-input and load conditions.

The Dragonfly mission's thermal management system relies on two distinct heat transfer mechanisms across mission phases: during cruise, a pumped fluid loop provides conductive/radiative cooling of the Multi-Mission Radioisotope Thermoelectric Generator (MMRTG), on the surface of Titan, fan-driven convective airflow takes over this cooling function. To support both phases, a common motor controller has been developed capable of driving the pumps and fans as required across the mission timeline. The controller design prioritized power efficiency, electromagnetic interference/compatibility (EMI/EMC) performance, and control flexibility. These priorities led to several key architectural decisions: motors are driven directly off the power bus without an isolated converter stage, a very low corner frequency input filter was implement, and a microcontroller-based (rather than FPGA-based) architecture was selected to maximize control flexibility. This design has achieved a standby power of less than 1.5 W and an operating efficiency greater than 90%. With less than 2 years from project start to flight hardware delivery, the team prioritized rapid hardware fabrication to enable early, iterative testing. This approach surfaced key issues early, including a fan motor compatibility problem that drove a flight fan design update, layout-driven self-compatibility noise issues that were resolved in the engineering model (EM), and EMC characterization enabled by a custom conducted emissions test cart used to iterate on the input filter design. The controller's microcontroller-based architecture also makes it readily adaptable to different motor current/speed ranges and tuning, positioning it well for reuse on future missions. This paper presents the controller's design, key trade-offs, and hardware test results.

Dragonfly is a rotorcraft lander designed to fly in the cold, dense atmosphere of Saturn’s moon, Titan, and is the latest mission in NASA’s New Frontiers program. The rotors which will provide the thrust for lander flight are driven by Brushless DC (BLDC) motors, which are commonly used in rotor drives for terrestrial drone applications, as well as for various actuators on the Mars 2020 mission. While BLDC motors require a more complex controller compared to a traditional brushed DC motor, the lack of mechanical commutation via mechanical brushes is an advantage for reliability and environments where cleanliness and arch safety are a concern. Typically, in BLDC drives, rotor position feedback is provided by Hall effect sensors which determine the commutation state of the motor drive. However, Hall effect sensors perform poorly at the cryogenic temperatures to which the motors will be exposed to on Titan. To overcome this challenge, a sensorless control methodology is selected which determines commutation intervals based on direct measurements of the back-EMF voltage on open/undriven phases. The implementation and robustness analysis of this sensorless control methodology will be discussed, along with numerical verification.

In amateur rocketry, sustainability, cost-effectiveness, and autonomy are becoming as important as performance. Telemetry boards, flight computers, and GPS transmitters critical for search and recovery are typically powered by single-use or frequently recharged heavy battery packs, resulting in ongoing operational costs and the risk of chemical battery waste if a rocket is lost in the field. This work presents a six-step design methodology for an autonomous power system featuring a smart charging architecture fed by small photovoltaic panels integrated onto fin surfaces. The methodology comprises: (1) defining the avionics power profile and aerodynamic constraints; (2) arranging panels in a series-parallel electrical configuration to achieve sufficient voltage and current capacity; (3) mitigating reverse-current losses caused by in-flight shading through blocking diodes and distributing series groups across opposing fin surfaces; (4) managing energy storage via a BMS-equipped Li-ion battery pack protected against overcharge, over-discharge, and short circuit; (5) implementing mechanical power switching to prevent unintended operation during pad wait and transport; and (6) employing a vibration-resistant, polarized modular connector architecture designed to withstand high-G launch loads while preventing reverse-polarity connections. The proposed system enables the rocket to continuously charge its own battery whenever sunlight is available—both while waiting on the pad and during post-landing search operations—thereby eliminating dependence on single-use batteries and rendering the avionics autonomous against loss or power-depletion risk.

Components and subsystems that make up equipment for human exploration on the lunar surface require continued performance and operation. Lunar dust, or regolith, is extremely fine, abrasive, and clingy due to its electrostatic properties. System failure cannot be tolerated for it is extremely difficult to repair and expensive to replace. Therefore, navigation lighting for lunar applications must function when deployed and continue to function throughout the respective product life-cycle (days, month, or years depending on mission). This analysis evaluates GaN FET (gallium nitride field-effect transistor) based LED driver technology using a uniquely designed Indium Tin Oxide (ITO) conductive coated lens. In this test case, lunar rover headlights must be designed to resist dust accumulation and remain functional in the presence of this challenging material. Since the Moon has long shadows and uneven terrain, the headlights are designed to provide focused, directional illumination to help astronauts navigate craters, rocks, and other obstacles safely. This evaluation includes deployment of a newly designed Earth based lunar dust chamber for dust mitigation and resistance. NASA approved GreenSpar 250 regolith simulant was used to simulate missions intended for the lunar highlands region.

Helium is routinely utilized in spaceflight for flight preparation, launch operations, and cold-gas thrusters for reaction control. Helium is typically utilized due to its inert properties and small atomic size, allowing for high pressures and low mass. Although the element is nonreactive, it can infiltrate certain electronic components and temporarily alter the performance of sensitive devices such as MEMs oscillators, sensors, and laser systems. Understanding the behavior of a system following helium exposure is essential for ensuring mission success. Despite the wide use of helium during space operations, relatively limited documentation has been publicly released that focuses specifically on the effects of helium exposure to COTS instrumentation operating under a flight-like environment. In this paper, comprehensive helium exposure tests are conducted to evaluate the performance and reliability of a COTS camera intended for space applications under deep vacuum and transient helium exposure. Lab controlled helium exposure was utilized to assess potential impacts on system functionality and camera operation. Preliminary testing conducted on several MEMs and crystal oscillators indicated a systematic failure of MEMs oscillators within the controlled helium environment.

Microelectronic systems encounter distinct challenges compared to their terrestrial counterparts. These systems must account for constraints related to size, weight, and power, as well as environmental factors specific to extraterrestrial environments. Radiation is a critical environmental hazard that can cause detrimental effects. The radiation environment experienced by a system is determined by solar weather cycles, the system location, and the flight vehicle in which it is housed. Ionizing radiation exposure can induce errors in microelectronic devices through single-event effects (SEEs) and total ionizing dose (TID). SEEs occur when charge from ionizing radiation is deposited into sensitive volumes of the device, whereas TID results from the accumulation of trapped charge at the interface between the semiconductor and insulating material. These phenomena can lead to software execution errors, device performance degradation, or catastrophic device failure. Before deploying a system in space, mission designers must verify that radiation exposure will not substantially disrupt operations. This verification is accomplished by testing the components using terrestrial radiation sources that replicate the ion types and energies present in space. This procedure is commonly employed for microelectrodes, resulting in an expected error count that will be observed during testing of space deployment, ensuring that errors induced by radiation effects remain within acceptable limits. Once qualified, devices are integrated into systems and deployed for space missions. However, post-deployment evaluations of devices or systems to verify the precision of the error predictions are rarely conducted. The Indiana University Center for Reliable and Trusted Electronics (IU CREATE) and the Radiation Team have developed a payload to address this knowledge gap in the post-deployment evaluation of devices and systems to validate the error predictions. The modular payload will be deployed on a satellite to collect data on radiation-induced errors using commercial and custom components. A key component of the system is the IU Radiation Effects Integrated Circuit (IU RadFX IC), a custom device engineered to provide comprehensive data on radiation effects. Additionally, Commercial Off-The-Shelf (COTS) components were selected due to their extensive testing at terrestrial radiation sources. By analyzing the data from these devices, the mission seeks to enhance model accuracy by identifying discrepancies between the predicted and actual error counts. In addition to improving device-level error predictions, system-level modeling that temporally predicts error rates can be validated using in situ data. This validation will enable mission designers to have greater confidence in the reliability of their systems throughout the mission lifetime and facilitate the more rapid adoption of new state-of-the-art technologies.

Entering year 3, Sandia National Laboratories’ Radiation‑Assured Design and Testing for Electronics and Computational Hardening (RAD‑TECH) research campaign is a seven‑year portfolio that enables state‑of‑the‑art (SotA) microelectronics use in extreme radiation environments, including space. The effort is organized around three thrusts: discovery of novel radiation effects, maturation of assessment methodology, and hardening of electronics from the transistor level to system‑level architectures. Continuing FY26 projects align with these thrusts. The SoC‑level hardening effort explores built‑in hardening blocks and architectural techniques to improve resilience of complex system‑on‑chip designs, with iterative testing in mixed‑radiation environments and development of predictive models. The rapid annealing work investigates ultra‑fast thermal cycles and localized heating to recover radiation‑induced degradation in advanced nodes, targeting sub‑microsecond anneal times and minimal performance loss. The PLL hardening program focuses on stabilizing embedded timing circuits against dose‑induced drift, employing radiation‑tolerant bias networks and on‑chip monitoring. The watchdog‑demonstration project integrates a high‑performance processing core with a radiation‑hardened supervisory circuit that can detect faults in real time, validating end‑to‑end system reliability. Four new investigations broaden the campaign’s scope. One study evaluates charge‑accumulation hardening for commercial 3‑D NAND non‑volatile memory, aiming to enable terabyte‑scale capacities with dramatically reduced single‑event upset susceptibility and enhanced total‑dose tolerance. Another utilizes an ion‑beam source to produce tunable neutron spectra, providing high‑flux, low‑activation beams. The final effort designs a radiation‑hard power‑converter that meets the low‑voltage, high‑current demands of advanced computing modules while integrating radiation‑aware protection and sensing circuitry. Together, these investigations refine radiation‑assessment tools, produce validated hardening techniques, and will enable high‑performance computing systems to operate reliably in the most hostile radiation environments.

Planetary missions increasingly depend on onboard neural network inference for perception and autonomy, running on SoC-FPGA platforms whose radiation-tolerant variants are now flight-qualified or flying. Radiation-induced upsets are a defining constraint of this environment, and a growing body of work characterizes how neural networks respond to bit-level faults. Nearly all of it, however, stops at static metrics: inject faults, report the drop in classification accuracy. Missions do not experience accuracy numbers. They experience behavior. A fault that silently corrupts a terrain classifier matters only insofar as it changes what the rover does next, and the mapping from static error to behavioral consequence is neither obvious nor, for planetary robotics, systematically studied. This paper studies that mapping directly. We deploy two representative onboard workloads on a current commercial SoC-FPGA development platform from a device family with space-grade variants: an INT8-quantized convolutional terrain classifier trained on a public NASA Mars terrain dataset, and a quantized reinforcement-learning navigation policy trained in a physics-based rover simulation. We then run a two-stage fault-injection campaign. The first stage injects emulated single- and multi-bit upsets broadly across weights, activations, bit positions, and layers, characterizing static sensitivity and how INT8 quantization reshapes fault response relative to floating point. The second stage takes the sensitive fault sites identified in stage one into closed-loop hardware-in-the-loop experiments, with the simulator in the loop and all inference running on the target device, and measures behavioral outcomes: trajectory deviation, hazard encounters, task completion, and recovery. This two-stage design makes closed-loop evaluation tractable while grounding it in a broad static characterization. Against these results we evaluate two low-cost mitigations suited to power-constrained onboard use, activation range clamping and selective redundancy for the layers the campaign identifies as critical, and quantify their behavioral protection per unit of added latency, power, and resources. Faults are emulated at the application level; configuration-memory injection and beam testing are discussed as validation paths but not claimed. We expect three contributions. First, a behavioral-level characterization of fault resilience for quantized onboard inference, including where static accuracy metrics mispredict mission risk in both directions. Second, a sensitivity map across layers, bit positions, and network types that practitioners can use to target hardening effort. Third, practical, measured guidance on cheap mitigations, together with an open-source fault-injection and hardware-in-the-loop evaluation harness for the community. The work extends a line of inquiry we began with an FPGA architecture for deep learning in planetary robotics at this conference in 2017, and addresses what we now believe is the binding question for onboard autonomy: not whether these networks fit on flight-class hardware, but whether they can be trusted there.

Developing fault management software for embodied systems, such as lunar rovers, poses a number of challenges, with a significant one being a lack of available data. Obtaining sufficient representative data requires time-consuming test campaigns on expensive hardware, which typically is in high demand and requires human teleoperation. For lunar rovers, it may be impossible to gather enough flight-like data to train a machine learning (ML) system. Further, using such hardware also often means it is impossible to gather data from all relevant failure modes without damaging the physical test artifact. To address these constraints, our solution is to develop physics-based models of the lunar rover and its environment, generate training data using those models, and then evaluate the system against a set of faults on the actual rover test hardware using sim-to-real ML techniques. A core challenge in simulation-based training is developing models that allow algorithms to bridge the sim-to-real gap. For Voyager Technologies’ Vertical Solar Array Technology Optimized for Lunar Traverse (VOLT) lunar rover, this includes accurately simulating complex interactions between flexible wheels and lunar soil, as well as the structural flex throughout the rover base. These characteristics also change as the rover undergoes long-term operational wear, meaning that any model calibrated for a new rover could become progressively invalid over the course of the mission. To help our system bridge some of the challenges of the sim-to-real gap, specifically the challenge of complex interactions between the rover and the environment, we use domain randomization. Domain randomization involves building simulations with variable parameters and generates randomized scenarios by sampling parameters from predefined distributions. The simulation parameter distributions are constrained by physics-based bounds, and the resulting models are validated against data collected from the real VOLT test hardware. Additionally, bridging the sim-to-real gap requires a generalizable algorithm. MAIFLOWER provides generalizability by combining model-based reasoning (MBR) with a novel thermodynamics-inspired ML algorithm. We demonstrated that MAIFLOWER can be trained on simulated data and subsequently identify faults in data collected through a test campaign with prototype VOLT hardware. This campaign consisted of 36 individual tests across 5 different cases, where we used a gantry crane, pulley, test fixtures, and a sandpit to evaluate system performance under a range of anomaly conditions. The VOLT platform was operated with closed-loop, optimization-based control of the three-arm gimbal to demonstrate fault detection efficacy with real-time sensing and actuation. The hardware-in-the-loop setup enabled verification of ground-truth conditions for both the introduction and removal of fault conditions. MAIFLOWER accurately detected all anomalous soil conditions across all 36 tests.

The rapid expansion of mega-constellations in Low Earth Orbit (LEO) has increased the population of tracked resident space objects (RSOs), making collision avoidance a critical challenge for operators. Conjunction events requiring assessment and maneuver planning have become more frequent, necessitating scalable autonomous decision-support capabilities. Conventional workflows rely on ground based analysts to assess conjunction data, design avoidance maneuvers, and verify safety through secondary collision screening. As conjunction volumes continue to rise, these processes become increasingly resource intensive and difficult to scale. Machine learning based approaches offer a promising solution by automating safe maneuver recommendation and reducing the per event analysis burden. While machine learning and reinforcement learning techniques have been applied to satellite collision avoidance, two key limitations remain. First, existing approaches do not evaluate post-maneuver trajectory safety with respect to secondary conjunction risks from neighboring resident space objects, potentially introducing new collision threats after the primary avoidance maneuver. Second, most reported methods rely on multi-step sequential decision making frameworks that incur planning latency and execution complexity, while single-step maneuver generation remains unexplored. To address these limitations, this paper presents a novel application of machine learning for autonomous satellite collision avoidance maneuver generation through two approaches. Both methods explicitly account for post-maneuver secondary conjunction safety while incorporating fuel consumption and orbital deviation constraints. The first approach employs a neural network safety classifier that evaluates a predefined library of candidate maneuvers in a single forward pass, classifying each maneuver as safe or unsafe and producing a filtered set of avoidance actions, thereby addressing both identified limitations. The second approach employs a Proximal Policy Optimization (PPO) agent operating over a continuous four-dimensional action space to plan sequential avoidance burns across multiple orbital periods with a maximum of 20 decision steps. Post-maneuver secondary safety is addressed through neighbor proximity penalties within the reward function. For both approaches, fuel efficiency and orbital stability are promoted through penalties on propellant consumption and semi-major axis deviation from the initial orbit. The methods are trained and evaluated using 5,000 realistic Low Earth Orbit conjunction scenarios derived from space surveillance catalog data. Both approaches are developed and evaluated in a custom Python-based orbital mechanics simulation environment. The supervised classifier achieves training and validation accuracies of 92.32% and 90.07%, with a false-safe rate of 1.89%, enabling autonomous, risk mitigated decision making in 9 out of 10 scenarios and requiring only minimal human intervention. The PPO agent's safety rate improves from 3% to 84% over training, achieving a mean miss distance of 20.96 km using an average of 4.0 decision steps. Unlike the classifier, which is constrained to a discrete candidate library, the PPO agent optimizes over a continuous action space, enabling more efficient fuel consumption and reduced orbital deviation. The results demonstrate the effectiveness of both approaches for autonomous collision avoidance. The full paper will present the formulation, implementation details, performance evaluation, and trade-off analysis of both approaches, highlighting the complementary benefits of single-step supervised maneuver selection and multi-step reinforcement learning based maneuver planning for autonomous satellite collision avoidance.

This study presents a hierarchical multi-objective optimization framework for solar array steering guidance of a non-sun-synchronous low-Earth-orbit (LEO) satellite equipped with asymmetric solar array wings. The steering angles affect not only solar incidence and power generation but also the projected area and center of pressure relative to atmospheric flow, thereby influencing aerodynamic-drag-induced delta-V and reaction-wheel angular momentum accumulation. Consequently, conventional Sun-tracking strategies that maximize power generation are not necessarily optimal for orbit maintenance and attitude control. To support practical satellite guidance and control, several rule-based steering patterns were parameterized using a limited number of design variables. The resulting steering laws were optimized to satisfy power requirements while reducing orbit-maintenance and reaction-wheel angular momentum accumulation. Each candidate steering law is evaluated using an integrated numerical simulation model including orbital motion, eclipse determination, solar array power generation, and environmental disturbances. A mission-representative numerical case was constructed based on the orbital conditions and asymmetric solar array configuration of the Precipitation Measuring Mission (PMM) satellite under development by Japan Aerospace Exploration Agency (JAXA). Generated power is calculated from the relationship between the Sun vector and each array surface normal. Aerodynamic drag is evaluated using the satellite geometry and the array-angle-dependent projected area normal to atmospheric flow. Reaction-wheel angular momentum was obtained by integrating the aerodynamic, solar-radiation-pressure, gravity-gradient, and residual-magnetic torques. The problem is formulated as a two-objective optimization minimizing drag-induced orbit-maintenance delta-V and maximum accumulated angular momentum, subject to array angle and rate limits, required power, and reaction-wheel momentum capacity. In the first stage, two basic steering patterns are optimized independently to obtain their Pareto-optimal sets. In the second stage, two composite patterns are constructed by assigning a different basic pattern to each wing and interchanging the left-right assignments. The composite patterns are reoptimized using combinations of first-stage Pareto solutions as the initial population. For each pattern, a representative solution is selected by minimizing its distance from the ideal point in the normalized objective space. The simulations revealed clear trade-offs among generated power, drag-induced delta-V, and maximum accumulated angular momentum. The first-stage results identified the operating conditions under which each basic pattern was advantageous for reducing either delta-V or momentum accumulation. The composite patterns yield additional non-dominated solutions beyond those obtained from the basic patterns, demonstrating that coordinated use of the asymmetric wings can improve the trade-off between the objectives. The final Pareto set included solutions that satisfied the required power while reducing delta-V by approximately 25% and maximum accumulated angular momentum by approximately 7% relative to the baseline continuous Sun-tracking strategy. These results demonstrate that solar array wings can serve not only as power-generation devices but also as controllable elements for managing aerodynamic characteristics and reducing attitude-control demand. The proposed hierarchical framework enables steering laws with different levels of complexity and performance to be quantitatively compared and selected according to mission-specific power, propellant, and momentum-management requirements. It provides an integrated guidance-design framework for asymmetric LEO satellites and can contribute to reduced propellant consumption and reaction-wheel momentum-management burden.

Formation flying (FF), in which multiple spacecraft cooperatively synthesize a large effective baseline, is an enabling technology for next-generation interferometry, coronagraphy, synthetic-aperture imaging, and gravitational-wave detection. Recent on-orbit demonstrations have achieved millimeter-level relative control over hundred-meter baselines, and next-generation concepts, such as infrared stellar interferometers and space-based gravitational-wave antennas, push relative-position requirements into the sub-millimeter to sub-micrometer range. At these levels, post-launch sensor-to-payload misalignment and slowly varying thermo-mechanical deformation can no longer be ignored: they introduce persistent biases that fixed-gain, ground-calibrated models cannot represent. A 10 μrad pointing drift maps to hundreds of micrometers of lateral motion at a 40 m inter-satellite distance, indistinguishable from true relative displacement, corrupting the navigation estimate and hence closed-loop pointing and pose (position and attitude) control. The quantity to be navigated is the relative pose of the mission payload frames, not those of the sensor-bearing bodies. Because each sensor is misaligned with the payload, the sensor-frame pose differs from the payload-frame pose that the mission actually requires. The navigation sensors retain a wide field of view and stay usable throughout operation, but they drift. Star-tracker differencing reaches arcsecond-level relative attitude. The laser ranger achieves several tens of micrometers along the line of sight. Quadrant-photodiode sensing achieves tens of micrometers laterally. The mission payload attains markedly higher precision, but only over a narrow capture range and at intermittent epochs. The problem therefore becomes one of reference-triggered bias estimation. Each payload reference re-identifies the sensors' biases, which are carried forward to compensate them until the next reference. Temperature and orbital telemetry serve as priors for the slowly varying bias. We propose a Learning-Augmented Extended Kalman Filter (LA-EKF) to target the payload-frame relative position and attitude directly. A physically consistent EKF backbone carries the well-modeled dynamics, while a learned residual model, calibrated by the intermittent references, compensates the effective, temperature-dependent error between the sensor-derived and true payload-frame pose. In the offline setting, the corrector is instantiated as ridge-regression and neural-network mappings from thermal features to a residual offset. In the online setting, it is re-calibrated on orbit at each reference event via recursive least squares. The proposed framework is evaluated on a three-spacecraft infrared interferometer, whose optical path difference (OPD) between the payload arms serves as the intermittent reference. The evaluation employs realistic multi-node thermal histories with a sun-angle train/test split that holds out unseen conditions. The numerical simulation shows that a plain EKF cannot represent the thermo-mechanical payload-frame OPD bias, and a LA-EKF substantially reduces it, markedly improving payload-frame accuracy over the EKF baseline. The decisive case is a post-training pose shift, in which the mounting geometry at inference differs from that seen in training: offline ridge and neural-network correctors, fixed at upload, retain the resulting payload-frame bias, whereas the online recursive correctors re-identify it on orbit once a reference becomes available, with the residual gap limited only by reference availability. The study thus isolates when on-orbit reference-triggered adaptation, not offline calibration alone, sustains high-precision navigation of the payload frames across intermittent-reference gaps.

This paper presents novel algorithms for reducing fuel usage in spacecraft control scenarios using reinforcement learning (RL) techniques. While most of the existing literature aims to reduce fuel usage through negative fuel rewards, we instead construct algorithms that intermittently incorporate drifting actions where the spacecraft does not use any fuel. The first algorithm, Drifter-Refine, reduces fuel usage through refining a previously trained RL policy. At each time step, Drifter-Refine uses model-based lookahead and is forced to drift if total fuel usage is expected to decrease. Experimental comparisons are performed in the popular Aerospace SafeRL simulated docking environment from the 2022 IEEE Aerospace Conference. Drifter-Refine reduces fuel usage by 39.2% when compared to a base model trained with Proximal Policy Optimization, showing that training with a negative fuel reward does not necessarily produce an optimal policy in practice. Motivated by these results, we construct a second algorithm, Drifter-Learn, that is forced to incorporate drifting actions during the initial training process. The key algorithmic idea is to allow a variable-length final time step in each episode, allowing the spacecraft to drift for a relatively long amount of time while only performing the Bellman backup for a single time step. Multiple such models are trained on different sectors of the state space and stitched together to form Drifter-Learn. Experiments show a decrease in fuel usage of 87.7% when compared to the base model. An ablation study was performed to show that the Drifter algorithms can effectively operate model-free and handle common safety constraints while maintaining performance in noisy and sparse reward conditions. Moreover, Drifter policies considerably reduce the number of time steps with thruster engagement, which aids in the transfer of the policies to physical systems. The totality of these results highlights a new approach for designing and implementing RL algorithms for spacecraft control that achieve significantly higher fuel efficiency.

Future space missions involving proximity operations for debris removal or servicing will require controllers that can handle the binary on/off constraints posed by the spacecraft thrusters. While previous works have proposed Mixed-Integer Model Predictive Controllers (MIMPC) to tackle this problem, these are computationally expensive due to the mixed-integer optimization problems being NP-hard. Reinforcement learning (RL)-based controllers have also been proposed and tested as they offer a potential tradeoff by moving the computing complexity to training time and offer a low-compute inference possibility on the satellite. However, these controllers are generally sample inefficient, hard to formally verify, and lack adaptability to new mission scenarios. This work proposes an RL-Augmented Model Predictive Controller (MPC) where the RL policy only decides whether to activate a thruster and the MPC optimizes the firing duration and the remaining continuous inputs. This controller is compared with a baseline MIMPC and an end-to-end RL policy for the REACSA setup from ESA in simulation regarding their performance in approaching a goal and station keeping. The controllers are also subjected to various robustness tests across sensor noise, actuation errors, model errors, and actuation delays. RL+MPC proves to be a promising strategy, with better sample efficiency and robustness than end-to-end RL and, at the same time, a faster control rate along with robustness to various actuation, model, and sensor errors while only sacrificing a little optimality compared to MIMPC. The results suggest RL+MPC as a practical and realistic intermediate step between MIMPC and RL.

Magnetorquer-based attitude control is attractive for small satellites due to its simplicity, reliability, low mass, and power efficiency, but is instantaneously underactuated. Consequently, magnetorquers have traditionally been used primarily for detumbling, whereas full attitude pointing with magnetorquers has typically required either slow projection-based feedback laws or computationally expensive optimization-based planners. This work investigates whether a neural-network policy can provide accurate magnetorquer-only attitude pointing while retaining the implementation simplicity of a traditional feedback controller. During controller development, we find that memoryless policies fail to generalize reliably across arbitrary pointing objectives because the instantaneous observation does not contain sufficient information to resolve the underlying dynamics. We relate this limitation to the finite-horizon nature of magnetic attitude control: just as variation of the geomagnetic field accumulates control authority over time, temporal measurements can accumulate information about otherwise hidden control-relevant dynamics. This motivates a stacked observation history that substantially improves training stability and pointing performance. The resulting controller performs simultaneous detumbling and arbitrary attitude pointing from initial angular velocities up to 0.2 rad/s. Across 100 matched Monte Carlo simulations with realistic measurement noise and environmental disturbances, the history-augmented policy achieves a final mean pointing error of 16.2°, compared with 64.4° for the memoryless policy. Comparisons against representative magnetic feedback controllers demonstrate substantially faster transient convergence, while robustness tests show stable pointing under spacecraft, actuator, and orbital variations outside the training distribution. Developed in the context of Carnegie Mellon University’s ARGUS CubeSat mission, these results indicate that learned feedback policies offer a computationally simple approach to rapid magnetorquer-only detumbling and pointing.

We describe and analyze a distributed compute architecture for SSO computational satellites that can potentially provide >100 kW compute power per launched metric ton (including deployment and station keeping mass). The architecture co-locates and integrates the solar cells, radiator, and compute functions into multiple small panels arranged in a large array. The resultant large vapor chamber radiator area per panel should permit ICs to operate at junction temperatures near 40*C with benefits in compute efficiency and reliability. Using the structure of the radiator to support the solar cells may also yield a specific power of about 500 W/kg compared to less than 100 for existing conventional implementations. Assuming development of custom solutions for all components, a 16 MW computation, 150 ton satellite comprising a 20 m x 2200 m grid of 16,000 panels can fit in a single Starship hold. The concept is scalable to much larger satellites with higher mass payloads or using on-orbit assembly. We consider panel sizes from 1 to 4 m2 to allow trading vapor chamber heat transport with compute efficiency and inter-panel communication. Assuming a 1 kW/panel design, 512-panel subarrays of the satellite can run a representative inference-only LLM with 500,000 token context window and 128 attention blocks, at a rate of 553 tokens/sec/session, across 256 simultaneous in-flight sessions. A full satellite could support 31 such subarrays, for >7900 inferences at a time.

Spacecraft cryptographic migration must preserve command authority, real-time execution, and recovery while retiring an obsolete policy. Algorithm inventories alone do not establish that the implementation assessed on the ground is the implementation selected after an interrupted onboard update. This paper presents MOSAIC, an analytical framework connecting a Mission Cryptographic Dependency and Fault-Propagation Graph, a Mission Cryptographic Cut-set Index, an Assurance-Constrained Agility Envelope, and a Monotonic Policy-Epoch Transition Protocol. A common identity binds the cryptographic use, planning and target graphs, exact build, policy core, operational scope, and qualification assumptions. Availability loss and unauthorized effects use separate monotone predicates; admission requires simultaneous resource, evidence, safety, and recovery closure; protected state separates software-bank recovery from restoration of retired policy. The evaluation uses fully specified synthetic inputs and standardized object lengths, without operational mission records or an external dataset. Complete enumeration of a five-vertex graph produces equal-weight priorities of 0.90, 0.75, and 0.60. An exact 286-point weight grid yields 215 strict preferences for the shared dependency, 65 for the longer-exposure branch, and six ties. A second Boolean case distinguishes availability from integrity cuts. A declared 160-byte body with fixed-width P-384 and ML-DSA-65 signature values requires at least 3,565 bytes, or 17, 8, and 5 protected fields at usable capacities of 220, 480, and 880 bytes. Constructed contract checks examine record substitution and shared-budget admission; bounded journal and restart checks expose omitted freshness, live-state, and recovery guards. These results establish reproducible decision semantics and conditional identity preservation, not measured flight performance or full protocol verification. The framework makes withheld admission an explicit outcome and preserves current-policy recovery under its protected-service assumptions. Flight use still requires complete mission evidence, exact-target resource bounds, independent qualification, and a demonstrated implementation of the protected operations.

The transient and cumulative radiation responses of gate-all-around (GAA) devices with various backside power distribution (BPD) configurations are investigated with technology computer aided design (TCAD) software. A suite of architecture parameterizations are simulated to observe how single event effect (SEE) and total ionizing dose (TID) responses are affected. TCAD models are built according to published specifications for nanosheet FETs and BPD devices. BPD architectures using through-silicon vias are the focus of this work, and both bulk and buried dielectric isolation (BDI) GAAFETs are investigated. Backside metal isolation layer thicknesses, TSV position, and NS widths are varied. Changes to the drift/diffusion dynamics of ion strikes in the example BPD models are analyzed. In our bulk models, the substrate thickness, backside metal isolation layer thickness, and nanosheet width have the greatest effect on the heavy ion charge across the parameterizations studied. At low LETs, our bulk models accumulate up to 35% more charge than “traditional” frontside power models due to pinning of the electrostatic potential in the substrate by the backside metal layer. At high LETs, diffusion current becomes a larger component of the transient, and charge collection is more dependent on substrate thickness regardless of power delivery. Across all bulk model parameterizations, it is found that heavy ion collected charge differs no more than 50% between frontside and our backside power models. As a result, our SRAM models made with both frontside and backside power architectures exhibit similar single event upset (SEU) performance. For devices with BDI, the heavy ion transient is truncated substantially, and charge collection is reduced over an order of magnitude. For both bulk and BDI GAAFET models, TID degradation is found to be minimal up to high charge concentrations. BPD elements do not meaningfully affect the TID response in our models. Across models and conditions studied, BPD elements did not significantly change radiation response relative to frontside power, despite the addition of dielectric regions (TID susceptibility) and substrate potential gradients (SEE susceptibility). Additionally, differences in backside power parameters in our models did not drastically affect response. This absence of increased degradation suggests promising responses for similar devices if used in space. As the semiconductor industry continues to modify and optimize the BPD architecture for performance and density, changes in the radiation response of these devices need to be continually monitored as the architectures could vary significantly. SEE testing at low LETs will continue to be important for qualifying BPD parts for space applications. Regardless of power delivery method, eliminating the drain-substrate junction will likely exhibit the best radiation response, both from an SEE and TID perspective. Based on these results, BPD GAA devices appear to be promising for radiation environment applications.

Event-based sensors are rising in popularity for vision tasks due to their high temporal resolution, low power consumption, and high dynamic range. These properties enable vision in environments where resources are limited, which makes frame-based vision applications much harder to carry out. Environments such as space and the lunar surface are two prime examples. Performing vision tasks in such environments requires as little resource consumption as possible, while maintaining high-quality data acquisition for various vision and science efforts. The detection and tracking of lunar regolith ejected from the wheels of lunar terrain vehicles (LTVs) is important to the overall design process for such vehicles. By understanding the characteristics of wheel-loft ejecta, engineers are able to more informatively iterate and improve upon design architectures to properly mitigate the damages lofted lunar regolith causes to other LTVs and lunar habitats. Using event-based sensors to detect and track the wheel-loft ejecta aids in saving resources where they are otherwise limited. This research presents a multiple-object tracker (MOT) making use of event-based data for a preliminary experiment to detect and track wheel-loft ejecta. By combining an unsupervised clustering algorithm, DBSCAN, with the well-known Kalman filter for tracking purposes, this work assesses the viability of characterizing wheel-loft ejecta through estimating initial velocities and launch angles of detected objects. Our efforts validate the Kalman filter motion models through an analysis on synthetically generated data that behaves as ideal projectiles to assess the unlabeled event-based data. We find that our MOT is capable of detecting and tracking objects in a dense environment using a simple tracking-by-detection architecture.

Triple Modular Redundancy (TMR) is the legacy solution for SEU tolerant electronics, in which three identical copies operate in parallel and resolve a single SEU error through voters. Semiconductor ASIC renditions of a TMR structure consist of three logic tracks. Together, the three tracks contain three sets of combinational logic, two sets of 3-input voters, and three storage cells. The tracks must be separated by a minimum distance Dmin, which guarantees that a single SEU strike does not impact logic in more than one track. This paper presents three enhancements to legacy TMR that reduce the area required by semiconductor ASIC implementations. The first SEU tolerant structure focuses only on combinational logic; legacy TMR voters and storage cells are employed as is. Two enhancements are employed to reduce the chip area: (1) more efficient logic structures than classical CMOS and (2) the ability to allow PMOS and NMOS logic functions to reside adjacent to each other. Legacy TMR, using classical CMOS, must separate logic functions in ASIC tracks at least Dmin apart; Dmin is the minimum distance separating sensitive circuit nodes. Pass logic is used to implement the logic functions. In addition to requiring fewer transistors, pass logic produces hazard-free outputs. The classical CMOS track consists of PMOS and NMOS regions; at any one time, either the PMOS or the NMOS region is OFF. The pass logic structure also has PMOS and NMOS regions, but both regions are ON and each produces the same output value. Through proper coding, these regions can be adjacent and an SEU cannot affect both PMOS and NMOS outputs at the same time. The first SEU structure can reduce combinational logic area by 50%. The second SEU tolerant structure utilizes aspects of the first structure with the goal of removing one set of TMR voters. The combinational logic consists of a pair of PMOS and NMOS pass logic circuits, producing four independently generated outputs. When an SEU strikes one of the PMOS/NMOS pairs, at most one output is affected. One cannot determine which output is correct, but SEU fault detection is realized. A simple interface circuit detects the SEU error in real time and passes the correct output to legacy TMR storage cells. The combinational ASIC area is reduced by 33%, and one set of voters is eliminated. The third structure builds upon the second structure and replaces legacy TMR storage cells and voters with a single Self-Restoring Logic (SRL) storage cell. Individual SRL storage cells are used today in space applications in a classical ASIC semiconductor structure, but not in the structure presented here. This SEU tolerant structure also saves more than 33% of the combinational logic area required by legacy TMR. Since SRL is smaller than TMR with voters, additional area is saved. No voters are needed

As launch costs decline and small-satellite missions become increasingly common, spacecraft designers are adopting commercial off-the-shelf (COTS) electronics to reduce cost, improve availability, and expand design flexibility. Although numerous CubeSat and commercial space missions have demonstrated the viability of COTS electronics, their use requires application-specific radiation assessment because electrical performance under total ionizing dose (TID) exposure varies significantly across devices. The cost of radiation characterization remains a major obstacle to COTS adoption, as testing effort scales with the number of candidate parts, device samples, and accumulated dose levels. During early component selection, exhaustive qualification is often impractical; the primary objective is instead to identify promising candidates for comprehensive radiation testing. This work presents an AI-assisted upscreening framework that predicts higher-dose electrical degradation from sparse low-dose TID measurements to support early-stage radiation-aware component selection. The framework transforms measured electrical characteristics into a compact latent representation of degradation, learns its evolution with accumulated dose, and propagates prediction uncertainty into mission-specific performance metrics. Rather than applying a universal pass/fail criterion, the method estimates the probability that a device will violate designer-defined specifications, such as allowable on-current degradation, on-resistance increase, leakage-current growth, or threshold/gate-drive margin, thereby enabling screening decisions that reflect mission requirements. The proposed framework is evaluated using low-cost COTS N-channel power MOSFETs, a widely used component for spacecraft power switching and load management. Devices are irradiated through a TID campaign to 50 krad(Si), while the AI model is trained using only the early portion of each device's measured dose history. The remaining higher-dose measurements are withheld during training and used to evaluate the model's ability to predict later-stage degradation. Predicted degradation trajectories are converted into application-relevant screening metrics that quantify the probability of specification violation as a function of accumulated dose. The Bayesian framework was trained on 0-20 krad measurements across seven COTS 2N7002 samples and evaluated against withheld 20-50 krad data. The threshold-voltage trajectory was reconstructed with a test-set R² of 0.914 and a median prediction error of +32.3 mV at 50 krad, less than 1.5% of the pre-irradiation value. Extending the analysis across the full campaign reveals the framework's core utility as a dose-dependent risk profiler: at 40 krad, survival probabilities fall to 39.0% for on-current retention and 19.7% for gate-drive margin, declining further to 15.5% and 8.6% at 50 krad. Together, these results show that sparse, low-dose measurements can supply well-calibrated probabilistic spec-survival estimates across multiple device-level performance metrics, supporting partial-dose campaigns as a low-cost first screen before full-dose qualification. By enabling many candidate devices to be screened using reduced sample counts and partial-dose measurements, the proposed approach provides a low-cost, uncertainty-aware decision layer for prioritizing COTS before committing to full-dose, full-sample-count radiation qualification. The framework is intended to complement, rather than replace, conventional radiation qualification.

With the ongoing evolution of "New Space", spaceborne digital systems increasingly embody commercial off the shelf (COTS) integrated circuits (ICs) for their high functionality/sophistication and low cost. However, COTS ICs, designed with COTS (library) cells, are susceptible to radiation induced effects, including single event-transients and single event-upsets. To enhance radiation tolerance, radiation hardened by design (RHBD) cells are often employed to replace COTS cells at critical circuit nodes, i.e., cells whose faults significantly impact system reliability. For such designs incorporating RHBD cells, including RHBD-only and mixed‑COTS/RHBD designs, early‑stage simulation-based Functional‑Safety (FuSa) evaluation is crucial. This is because FuSa estimates system-level reliability based on cell-level reliability information of the cells employed, enabling early identification of critical nodes. However, prevailing FuSa tools are not directly applicable to RHBD cells, because they rely on technology‑dependent parameters pre‑defined for COTS technologies, and not for RHBD. Specifically, directly applying existing FuSa flow to designs incorporating RHBD cells implicitly and inadvertently assumes that RHBD cells feature identical radiation hardness to COTS cells. This assumption may lead to an overly-pessimistic reliability estimation for RHBD-only and mixed-COTS/RHBD designs. Put simply, most prevailing FuSa tools are unable to precisely ascertain the system-level reliability of designs incorporating RHBD cells. To address this limitation, we propose a radiation‑calibrated FuSa flow for early-stage reliability estimation of RHBD-only and mixed-COTS/RHBD designs. Our proposed flow augments the default (COTS-based) parameters within the FuSa tool with radiation‑calibrated parameters derived from RHBD-cell characterization. To the best of our knowledge, this work is the first to integrate heavy-ion-derived RHBD cell-hardness calibration into a COTS-oriented FuSa flow. The proposed FuSa flow comprises three steps. First, a hardness‑correlation factor between the RHBD and COTS cells is ascertained from heavy-ion characterization. Second, this hardness‑correlation factor re-calibrates the technology‑ dependent parameters of the FuSa tool, augmenting it with radiation‑calibrated parameters for the RHBD cell-library. Third, the re-calibrated parameters are applied to designs to ascertain their system‑level reliability to identify the critical nodes for subsequent placement of RHBD cells. To, in-part, demonstrate the proposed flow, we design a linear-feedback-shift-register (LFSR2) and an Advanced-Encryption-Standard (AES) module. As an initial feasibility study and a proof of concept, each design is implemented in COTS-only and RHBD-only versions, and evaluated by the proposed FuSa flow. Thereafter, their system-level reliability results are compared to verify whether the proposed flow correctly captures the reliability improvement introduced by RHBD cells. Using RHBD-only implementations of LFSR2 and AES, the proposed flow predicts system-level failure-rate reductions of ~3.7 and ~1.3 orders of magnitude, respectively, relative to their COTS-only implementations. These improved reliability estimates reflect the cell-level reliability from our heavy-ion characterization as expected. The results provide an initial simulation-based verification that heavy-ion-derived RHBD cell-hardness information can be incorporated into pre-silicon FuSa analysis. Consequently, our proposed FuSa flow benefits the early-stage evaluation of RHBD-only and mixed-COTS/RHBD designs. Our validations of mixed-COTS/RHBD implementations and post-silicon verification are ongoing. Upon successful completion of the said validations, our FuSa flow is potentially an improved part of Electronic Design Automation flow for industry-standard space-grade IC designs.

In the next paradigm of New Space, commercial off-the-shelf (COTS) systems (vis-à-vis individual COTS ICs) are expected to be employed in space applications, offering high performance and functionality, possibly akin to that on-earth. Current-monitoring protection for Single-Event Latchup (SEL) and micro-SEL (µ-SEL) is necessary in COTS systems to distinguish abnormal latchup signatures from normal operating transients. Radiation (e.g., heavy-ion) tests remain essential for susceptibility and cross-section evaluation. However, stochastic latchup occurrence, limited beam time, and restricted rail access constrain validation of such protection, particularly in complex multi-rail COTS systems. To address this gap, we propose a radiation-informed deterministic SEL/µ-SEL emulation approach for protection validation – the design and construction of our ZM-PCG-01 programmable current generator and its application. By means of our ZM-PCG-01, a Hill-type sigmoidal current profile fitted to measured latchup signatures is reproduced as a controlled current draw from the rail supplying the sensitive COTS IC (i.e., where SEL/µ-SEL may occur). A two-dimensional ball grid array of abnormal-current increment and rise time defines 189 profiles per rail, each executed 20 times. Tests on Kintex-7 FPGA, Zynq UltraScale+ MPSoC, and NVIDIA Jetson Orin Nano platforms evaluate current profile reproduction and protection response. Across 10 representative rails, the mean absolute relative error in the current increment, mean absolute rise-time error, and run-to-run coefficient of variation are 3.7%, 2.2 µs, and 1.2%, respectively. For throughput, a complete rail sweep comprising 3,780 executions required only approximately 2.1 hours. In short, our proposed approach complements radiation tests by converting radiation-derived latchup signatures into repeatable test vectors.

This paper offers a path to accelerated consideration and adoption of Automotive Electronics Council Stress Test Qualification for Passive Components (AEC Q200) qualified commercial off-the-shelf (COTS) resistors and capacitors. Recommendations are provided for due diligence in part selection and technical mitigations for common concerns and tailored to program risk profiles. The high costs, long lead times, and technology maturity lag of MIL-SPEC components pose significant challenges for modern procurement needs. COTS components better align with today’s cost and schedule constrained satellite programs. However, COTS parts typically undergo extensive scrutiny and testing after procurement prior to use on U.S. Government satellites. These processes may be appropriate for some high-reliability long life missions, but they negate the cost and schedule benefits of using COTS and may unnecessarily stress the parts. By leveraging parts qualified to AEC Q200, space systems can take advantage of electronics that have been designed and tested to environments and reliability standards that exceed those for standard commercial electronics. The manufacturing and testing controls used by AEC Q200 part suppliers address many common concerns that historically hindered wide-scale adoption of COTS parts for space applications. Automotive parts must perform reliably over wide temperature ranges, endure frequent thermal cycling, and handle high vibration and mechanical stresses—conditions needed for satellite applications. AEC Q200 suppliers are compelled to deliver defect-free parts given the liability concerns associated with automotive applications. Their processes employ Design and Process Failure Mode and Effects Analysis (DFMEA/PFMEA), automated production, Statistical Process Controls (SPC), and real-time monitoring to achieve defect rates targeting less than one per billion parts. These processes result in quality levels that often exceed traditional MIL-SPEC manufacturing lines. Although discussions about COTS parts often focus on complex active components, this paper focuses on resistors and capacitors, which comprise up to 80% of a satellite’s bill of materials. These passive components have known qualification requirements documented in AEC Q200, and well-established design and manufacturing processes. Furthermore, resistors and capacitors are not subject to radiation concerns that affect active devices in most LEO applications. Common concerns regarding AEC Q200 resistors and capacitors in satellite applications include screening and inspection differences, configuration change management processes, life test differences, temperature range differences, pure tin finishes, other prohibited materials, and radiation. Each of these concerns are addressed individually with proposed technical mitigations appropriate for various program risk profiles. Specific recommendations for due diligence in part selection and higher level considerations that may warrant the use of AEC Q200 resistors and capacitors without component level testing are also provided. The robust design, testing, and process controls employed by AEC Q200 manufacturers demonstrate component quality suitable for space applications. For risk-tolerant programs, particularly those employing constellation architectures with on-orbit redundancy, AEC Q200 resistors and capacitors provide an impactful first step toward improving acquisition responsiveness and reducing costs while maintaining mission assurance.

Commercial off-the-shelf spacecraft processors can suffer data, control-flow, and reset faults during signature verification. A corrupted acceptance can cross the command trust boundary even when the underlying signature algorithm is sound. Fault-Contained ML-DSA Verification (FCMV) combines a sealed transaction record, independently rebuilt structural and challenge predicates, two randomized equation checks, ordered control tokens, and a transaction-bound coded latch. The design separates the cryptographic verdict from permission for a consumer to act and states the storage, challenge-source, and consumption assumptions on which containment depends. Evaluation separates functional cryptography from a finite decision model. Independent Python and C implementations jointly accept 144 Python-origin and 48 reverse-origin tuples across the three ML-DSA parameter sets and reject all 24,576 negative verification executions. A fully specified campaign enumerates 589,824 predicate-level injections. FCMV has zero unsafe acceptance in 73,728 one-site cases and 312 in 73,728 paired within-class two-site cases. The latter is 97.98 percent below duplication with a shared, unprotected final comparison in this model. The remaining cases expose a two-predicate hint cut. Scalar residual checks, explicit control-trace tests, and latch disturbances test distinct parts of the construction. The ideal two-projection bound is $q^{-2}$ only for a fixed nonzero residual and independent uniform challenges. Neither that bound nor the campaign predicts radiation rates. End-to-end FCMV overhead and target fault-domain independence remain unmeasured. The contribution is a concrete acceptance contract, a conditional containment argument, and an auditable model that identifies both protected paths and residual weaknesses.

Coming off a tremendously successful – in terms of both technical and scientific achievements and public engagement – Artemis II crewed lunar flyby mission, NASA and its commercial partners are working toward streamlined, executable systems, designs and missions for Artemis III and beyond. NASA’s Human Landing System (HLS) program, managed out of Marshall Space Flight Center in Huntsville, Ala., is charged with partnering with U. S. industry to provide demonstrations of lunar landers for delivering crew and large cargo to the Moon as part of Artemis. The HLS program is working with SpaceX and Blue Origin to develop these lunar landers for crewed missions and cargo landers that will deliver 15 metric tons of payload to the moon. The program and agency are focused on risk reduction activities prior to commencing crewed lunar landing missions, as well as reviewing mission designs and requirements to increase launch cadence for Artemis. To that end, NASA recently announced the intention to make Artemis III a mission in Earth orbit to test docking capabilities of the commercial landers with NASA’s Orion spacecraft for crew. To further reduce risk prior to the first crewed lunar landing, both SpaceX and Blue Origin are required to perform an uncrewed lunar landing demonstration mission at or near the Moon’s South Pole. These uncrewed missions will demonstrate the providers’ capabilities to coordinate launch operations; to perform a precision lunar landing; to send telemetry back to Earth for a minimum of two hours, testing NASA deep space communications assets with commercial assets; and to relight lander engines in the lunar environment. Following successful Artemis III docking and uncrewed lunar landing missions, NASA and its partners will land two crew members on the Moon, with two crew remaining in the Orion spacecraft, similar to the mission designs of the Apollo program. In addition to Artemis III and uncrewed lunar landing missions prior to the first crewed landing, the HLS program is partnering with SpaceX and Blue Origin on dozens of engineering collaborations in key risk reduction areas. Both providers also have the option to use NASA test facilities through Government Task Agreements (GTAs). Both SpaceX and Blue Origin have taken advantage of engineering collaborations, GTAs and NASA subject matter experts to advance systems and technology readiness. This paper will provide an update on the Artemis missions that will use the landers in development – both crewed and cargo – as well as the progress from both providers. In addition, lessons learned from Artemis I and II that are being implemented for HLS missions will be discussed.

Terrestrial mine planning is a mature discipline, with established methods for phased block and panel design, ventilation, ground control, and multi-stage permitting that trade production rate against geotechnical risk over a mine's operating life. No equivalent framework yet exists for lunar volatile extraction, where the geotechnical uncertainty of regolith at depth, the operational cost of liberated dust, and the near-total absence of in-situ repair capability make an uncontained failure, a cave-in or an uncontrolled dust release, far more consequential than its terrestrial analog. This paper adapts terrestrial mine-planning methodology to a phased architecture for a lunar volatile mine in which dust prevention and fault isolation are first-order design constraints from the outset, not mitigations layered onto a throughput-optimized plan. We propose a two-stage mine topology. Early development, spanning the first years of operation, proceeds as discrete excavation blocks, each independently sealed and dust-contained, sized to the production capacity of a small excavation platform such as the authors' TZ-II. As the mine matures, block-scale operations expand into a fractal, tree-structured network in which each branch can be isolated from its neighbors, so that a cave-in or dust-release event in one branch does not propagate operational risk to the rest of the site. We evaluate this topology against alternative mine architectures using a consistent set of criteria applied to every candidate, including production rate, resilience under a cave-in scenario, sensitivity to excavation fleet composition, dust containment, and, critically, the cost and feasibility of reviving a caved-in working area, a question with no direct terrestrial equivalent given the absence of on-site heavy repair capability on the Moon. Building on this topology, we develop a five-year mine development schedule and production estimate scaled to a small excavation platform, with an explicit scaling path to higher-throughput hardware as the fractal branches expand. We extend the terrestrial mine-planning workflow to the regulatory dimension unique to lunar operations: where terrestrial permitting proceeds through a single national authority, a lunar mine site requires a dual-track process spanning domestic government authorization and coordination under the UN Committee on the Peaceful Uses of Outer Space (COPUOS) framework, a dependency terrestrial mine planning has no need to model. We position this work against existing lunar ISRU mine-design literature, which has primarily optimized excavation throughput and system mass under nominal operating assumptions. This paper instead treats volatile extraction and severe dust mitigation as the governing design drivers from the outset, arguing that a mine architecture optimized only for nominal throughput is poorly suited to volatile-rich, geotechnically uncertain terrain, and that the fault-isolating, fractal approach developed here offers a more defensible basis for planning sustained lunar resource operations.

What was Apollo 11’s actual probability of success? Contemporary estimates spanned an extraordinary range: from extreme pessimism to extreme optimism. The question was never settled. To date, nobody has presented a well-supported estimate that rests on an end-to-end physical simulation of the flight. We present a physics-integrated Monte Carlo simulation that estimates Apollo 11’s mission-success and crew-survival probabilities by flying the entire mission — launch pad to splashdown — as one continuous numerically integrated trajectory. Every burn is a finite-thrust integration and every coast an n-body integration under Earth (point mass and J2, with the J3–J6 zonal harmonics near the Earth), Moon and Sun gravity, using the JPL DE440 lunar ephemeris and a degree- 8 GRAIL gravity field, with faithful renderings of Apollo’s guidance and operations: iterative-guidance ascent at a launch azimuth solved from the Moon-arrival geometry, a ground-tracked Manned Space Flight Network navigation layer that re-solves each midcourse cor- rection on a tracked orbit estimate, a two-burn lunar-orbit insertion, Apollo’s constant-drag entry guidance, and a physically modeled parachute descent and recovery. The vehicle is subjected to 1969-era engine reliabilities and a suite of failure modes sourced to the his- torical record — from Saturn V engine-out and an Apollo-13-class service-module failure to micrometeoroid strike, solar-particle-event exposure, and single-engine service-propulsion ignition risk. Across 10,000 trials we estimate 83.6% mission success and 94.2% crew survival at the model’s input failure rates (95% Monte Carlo sam- pling interval 82.9–84.3% on mission success), with a per-astronaut abort-survival model attributing the 10.6-point gap to Apollo’s abort architecture — missions lost to a survivable abort, most often a failed S-IVB relight for translunar injection that leaves the crew safely in Earth orbit. The larger uncertainty is in the input rates themselves. Three of them rest on a countable record — one service- module loss in 15 flights, no S-IVB restart failure in 10, about two docking anomalies in 21 — and propagating binomial posteriors on those three through a reweighting of the fleet gives a 95% credible interval (Jeffreys priors) of 63–92% on mission success and 87–96% on crew survival. The nominal trajectory reproduces Apollo 11’s flown burn times to within about a second, its trans-Earth injection ∆V to within 0.1%, its touchdown point to within eight kilometers of the planned landing site, and its entry deceleration and eight-day duration to within a couple of percent. Crucially, this nominal is derived without using any of Apollo’s flown trajectory values — the launch azimuth, the injection-guidance preset, and the descent- orbit timing are all solved from physics and geometry rather than commanded — so it predicts the flight rather than replaying it, and the same success rate holds across the program’s alternate launch opportunities. Our estimate sits above the crew’s own intuition, within a few points of the program’s eventual six-in-seven landing record, and far above the early pessimistic forecast.

Future lunar surface missions will require astronauts to traverse distances of up to 10 kilometers while maintaining accurate position and heading knowledge in the absence of a Global Navigation Satellite System (GNSS). Lunar surface navigation approaches that rely on dead reckoning and relative navigation accumulate heading error over distance, producing substantial cross-track position drift. This work investigates horizon matching as an absolute heading aid using monocular camera imagery and orbital Digital Elevation Model (DEM) data. The framework combines image-based skyline extraction, DEM-derived skyline generation, limited field of view (FOV) horizon matching, sequence-aware skyline feature tracking, and a terrain-dependent heading observability metric. Real Apollo 17 surface imagery is used to evaluate the horizon matching pipeline and characterize sensitivity to skyline extraction error, residual camera pitch and roll, and photometric degradation. These results define low, mid-range, and high-error heading scenarios that are propagated through unaided, thresholded aid, and continuously aided traverse models. The Apollo-derived characterization is then transferred to Nobile Rim 2 near the lunar South Pole using a DEM-derived observability map, without requiring synthetic surface imagery. For the Apollo 17 case, thresholded horizon aiding reduced final goal-position error from 50 meters to 21.7 meters, while root mean squared (RMS) cross-track error decreased from 76.4 meters to 73.4 meters because aid became available near the end of the traverse. At Nobile Rim 2, aid becomes available earlier, reducing RMS cross-track error to 59.8 meters and final goal-position error to 22.4 meters. Results demonstrate that horizon-based heading estimation reduces navigation error compared to unaided dead-reckoning. These results demonstrate that horizon geometry capable of supporting absolute heading corrections exists in both low-latitude and lunar South Pole environments, while also showing that navigation benefit depends jointly on the spatial distribution of terrain observability and the accuracy of the horizon derived heading measurement.

Crewed space missions generate substantial logistics waste. A crew of four aboard the International Space Station (ISS) produces roughly 2,500 kg of packaging and consumables annually, all of which must be stored or destroyed. As NASA pursues a sustained lunar presence, longer missions farther from Earth demand new approaches to trash management. Recycling waste into useful products for the Moon Base could reduce resupply needs, but the constraints of the space environment make traditional approaches impractical. To address this challenge, we present CERBERUZ (Composites for Extraterrestrial Recycling By Engineering the Reuse and Upcycling of Zotek), an end-to-end recycling and manufacturing architecture recognized as a Phase 1, Phase 2, and 1st place overall winner in NASA's LunaRecycle Challenge. Whereas traditional recycling approaches focus on sorted, pure waste streams, CERBERUZ exploits the heterogeneity of logistics waste by grinding and co-processing materials into a thermoplastic matrix-composite, with non-thermoplastic waste acting as reinforcing filler. The resulting feedstock is injection-molded into functional parts, with material combinations and fill fractions tunable to optimize properties for different end uses. Injection molding enables rapid production of precision parts, and the same system is adapted to extrude filament for 3D-printing. The architecture requires no water, chemical solvents, or hydraulic fluids. The same heterogeneous-feedstock approach generalizes to other waste streams and mission contexts, including Mars missions. Reflecting LunaRecycle's dual-track structure, we developed both a physical prototype and a digital twin of the process. Our prototype testing campaign demonstrated functional parts produced from five unique recycled feedstock formulations, including a screwdriver, net pot, structural short-pipe, and 3D-printing filament at an approximate energy cost of 3 kWh/kg of product. Our digital twin, combining physics-based models of system state with a machine learning model trained on sample products, accurately predicted novel composite material properties, flagged defect states early, and reliably simulated system behavior under lunar environmental conditions and off-nominal failures such as power loss. Beyond the positive test campaign results, CERBERUZ demonstrates a feasible avenue to rapidly mature space technology. In this two-year challenge, we paired digital twin development and validation with early-stage physical prototyping, advancing from concept generation to laboratory validation (TRL 2 to 4). The digital twin is now positioned to accelerate technology development where many promising systems stall (TRL 4 to 6). CERBERUZ's architecture extends beyond this initial demonstration; future work will qualify spare production for critical systems such as life support. Overall, CERBERUZ demonstrates a compelling, scalable approach to in-situ recycling of heterogeneous logistics waste into mission-critical products, relying on a digital twin framework to accelerate space technology development.

The establishment of a sustained human presence on the lunar surface necessitates new material science paradigms to address unprecedented extraterrestrial engineering challenges. High-performance polymers and polymer matrix composites offer a low-mass alternative to traditional aerospace metals, drastically reducing launch costs and payload constraints. These advanced synthetics also unlock the potential for lunar additive manufacturing, enabling 3D printers to fabricate structural components, tools, and habitats on the lunar surface. However, deploying these materials requires overcoming the novel engineering hurdles introduced by the space environment. This paper experimentally characterizes a newly developed space-rated polycarbonate under simulated lunar environmental conditions for vacuum, UV radiation, and extreme thermal gradients, and integrates those findings with finite element analysis and field tests on a lunar-simulant terrain bed to establish a comprehensive viability framework. The results demonstrate that this polymer class, when manufactured via material extrusion additive manufacturing, can satisfy structural, fatigue, and outgassing requirements for a compliant rover wheel intended for long-duration lunar exploration, as demonstrated by the recognition of a wheel designed with this material as a Phase 1 finalist in the Rock and Roll with NASA Challenge. The challenge required that a single wheel withstand loading of at least 256 N and absorb obstacles at speeds of up to 24 km/h, while having a mass of less then 2.3 kg and service life of 1000 km. While the wheel's operational speeds could only be maximized at 18.29 km/h while dealing with obstacles during simulations, the nominal operation speed was well above the specified rate at 38.62 km/h, and test coupons were able to survive over two times the number of load cycles that they needed to.

Maintaining a safe and habitable indoor environment is a critical challenge for crewed lunar habitats, where reduced-pressure atmospheres, limited ventilation resources, particulate contamination, and degradation of air-cleaning systems may affect crew health and mission sustainability. The degradation of filtering system negatively impacts the logistic for future missions. Thus, understanding how declining filtration efficiency affects indoor air quality, particle accumulation, and crew exposure is essential for developing reliable maintenance, replacement, and resupply strategies for long-duration lunar habitats. The influence of filtration degradation on airborne particle transport and accumulation is assessed by examining filtration efficiencies of 100%, 80%, 60%, and 40% and their effects on particle dispersion using computational fluid dynamics (CFD) within a inflatable crewed lunar habitat, with atmospheric conditions defined in consideration of NASA-STD-3001 Volume 2 requirements. The model evaluates the spatial distributions and partial pressures of oxygen (O₂) and nitrogen (N₂), while accounting for crew oxygen consumption. The particles were modeled at their highest concentrations in space, including particles generated by crew activities in addition to lunar dust entered into the inflatable habitat following extravehicular activities. A discrete phase modeling approach was used to characterize particle transport, accumulation, and removal under the prescribed ventilation conditions. The corresponding effects on airborne particle concentrations and spatial distributions were quantified. Changes in the oxygen supply condition are also investigated to represent the progressive reduction in available O₂ associated with crew consumption and prolonged occupancy. Steady-state CFD simulations are used to identify regions of inadequate gas mixing, elevated particle concentration, and potential crew exposure under different filtration and atmospheric scenarios. The proposed framework provides a basis for assessing the combined effects internally generated particles, lunar dust intrusion, and declining filtration performance on lunar habitat indoor environmental quality and may support the development of ventilation and air-revitalization strategies for sustained lunar missions.

The permanently shadowed regions at the lunar poles are among the most valuable places for a lander, and among the hardest to reach. The crater walls are steep and loosely packed, and with no sunlight on the floor, the cameras a lander leans on for terrain-relative navigation have little to work with. Landing on thrusters alone is risky, and blowing ejecta and few abort options add to the danger. Sending one large lander to the crater floor is the obvious alternative, but it commits the entire vehicle to treacherous ground and burns up heavy propellant, risking a landing it may not survive. Splitting the task into two, a stable host on safe terrain and a light lander on a tether, keeps the costly hardware out of danger, lets a far smaller craft reach places a full lander could not safely touch, and turns the tether into a lifeline for power, communication, and retrieval. But the tether that makes this possible also reshapes the control problem. It can pull but never push, so if the descent dynamics go wrong, the line can go slack and then snap taut again. Those snap loads can damage the tether, set the lander swinging, and ruin a careful touchdown. This paper introduces a constrained model predictive guidance scheme that replans a tension-feasible descent path on board each control cycle. We treat the daughter lander as a variable-length pendulum on an elastic tether with a compliant anchor, capturing how line stretch couples to swing and reel motion. At each step, the guidance coordinates the reel rate and a small reaction control system to follow a descent corridor, while keeping tension positive, under its limit, within reel and thruster limits, and clear of the crater wall. The core difficulty is the slack condition, which we formulate as a complementarity constraint linking tension to the cable's stretch. This makes the optimization non-smooth and nonconvex, so we relax it into a smoother form and solve it with a warm-started sequential convex program quickly enough to run on board. The cost function trades off tracking accuracy, propellant consumption, and tension spikes, and switches to a tension-only mode if the thrusters are unavailable. We evaluate the method in a six-degree-of-freedom simulation with a multi-segment elastic tether, a compliant anchor, and crater terrain, sweeping Monte Carlo cases across tether stiffness, initial swing, anchor slip, and wall slope. Because tension feasibility is built into the formulation, the guidance rules out slack and snap events by design, so the study focuses on how tightly it settles residual swing and how accurately it places the lander at contact, benchmarked against a fixed-reel PD baseline. This approach charts a practical path to the precise, repeatable access that Artemis-era prospecting will need.

In this work we present a novel application of neural model predictive control (MPC) to the rocket engine control problem of thrust and mixture ratio regulation. Based on an EcosimPro model of a bi-propellant rocket engine, neural network based representations of the engine’s dynamics are learned. Three variations are investigated in detail. One learn- ing the linear transfer function, one learning the full nonlinear representation and one linearizing the nonlinear representation online. All three are implemented in a neural model predictive controller using the CasADi framework. They are evaluated regarding their thrust and mixture ratio tracking performance and compared against a classical PID approach and existing analytical MPC approaches in literature. The investigations show that learned dynamics can be used for accurate real time (nSLMPC with 11 ± 1 ms and lin. MPC with 12 ± 2 ms per solve at 20 Hz) thrust and mixture ratio tracking. While the PID tracks the mixture ratio with lower cumulative error, the neural MPC implementations achieve more than 75 % reduction in thrust tracking error and a 65 % reduction in settling time for step changes in thrust and reduce mixture ratio overshoot and control effort. In contrast to the classical approach, the neural MPC implementations adhere to the admissible set, while the classical approach violates the state constraints. The neural MPC implementations compare well to their analytical counterparts applied in literature with similar settling time while reducing overshoot.

Modern space activities increasingly include commercial actors in partnership with government agencies. NASA’s Commercial Lunar Payload Services (CLPS) exemplifies this public-private partnership (PPP) model by contracting with commercial vendors to deliver NASA-sponsored science and technology demonstration payloads to the lunar surface. While PPPs offer cost efficiency, agility, and innovation, they also introduce challenges for NASA program managers coordinating manifest integration, schedule dependencies, and risk posture across vendors with differing technical maturity and business incentives. Commercial actors, however, are not inherently motivated by scientific objectives or concerned with environmental stewardship, creating a potential "tragedy of the commons,” where individual actors optimize for near-term gain while eroding a shared resource. Recent studies have begun characterizing the impact of surface activities on the lunar environment, including changes to the lunar exosphere, regolith, and permanently shadowed regions. For NASA program managers, these effects represent an emerging cross-mission risk that falls outside traditional single-mission risk frameworks. Exosphere alterations from exhaust plumes could compromise the baseline conditions scientific payloads are designed to measure, lander-ejected regolith could damage nearby hardware and instruments, and contamination of permanently shadowed regions could degrade the volatile deposits future ISRU and Moon Base missions depend on. This work characterizes environmental effects and then considers them from a risk perspective to identify institutional and programmatic levers available to NASA decision makers for management within the CLPS model. First, we characterize historical and projected impacts of CLPS missions through anthropogenic mass quantity, temporal, and spatial contributions. Results show that the proportion of lunar anthropogenic mass deposition attributed to CLPS missions has grown from 32 percent in 2024 to 88 percent in 2025. If all contracted 2026 missions are conducted, CLPS contributions alone will exceed three times those of 2025, the highest annual level since 1972. As cadence accelerates to establish the Moon Base with 10 proposed CLPS landings in 2027 and 12 in 2028, up from 2 in 2025, NASA program managers have limited time to anticipate environmental changes before cumulative effects constrain where future missions can land and what surface operations they can perform. To address this gap, we develop a network model of the CLPS initiative with human (H), technological (T), and environmental (E) nodes interacting through institutional (I) and knowledge (K) components. Quantitative network analysis generates possible interaction pathways, enabling analysis of socio-technical relationships among NASA, commercial providers, and the scientific community, and analysis of the influence of institutional mechanisms such as contracting structures and payload manifesting policies. The results identify high-leverage intervention points for NASA program managers, distinguishing actions that can be implemented through existing contracting and oversight authority from those requiring broader policy change. Key institutional changes are proposed, with downstream effects on commercial provider behavior evaluated using the model. Overall, this work provides program managers and commercial providers with a practical framework for anticipating and managing environmental risk created by an increasing cadence of lunar surface activity, supporting mission planning and oversight as the CLPS model scales to support a sustained lunar presence over the coming decade.

Future Moon base and lunar exploration activities will require more than isolated payload delivery. Sustained operations will depend on commercial service architectures that can acquire, condition, store, meter, and distribute locally available water and volatiles into useful operational functions. This paper presents a water-first lunar surface services architecture for supporting Moon base logistics, reusable mobility, sample return support, power and oxygen interfaces, and early utility infrastructure within emerging commercial lunar exploration models. The proposed architecture centers on a staged resource handling chain. A sorbent-based capture and release module, OASIS, is integrated within a broader WATERIG volatile handling system to support capture, holding, controlled release, purification, metering, and feedwater interface functions. Downstream storage and conditioning elements, represented by FIRN and TARN, provide intermediate buffering between intermittent resource acquisition and operational users. These users may include water-based propulsion nodes, hopping or landing mobility systems, oxygen generation interfaces, regenerative power systems, sample handling systems, and surface logistics services. Rather than treating lunar water only as a scientific target or commodity, the architecture treats water as an enabling service medium. This framing allows early lunar missions to be evaluated not only by extraction yield, but by their ability to convert uncertain, low-rate, or geographically distributed volatile availability into mission-relevant commercial services. Potential service categories include feedwater delivery, resource-status verification, payload relocation, mobility support, surface power support, and utility-style supply for crewed or robotic Moon base operations. A concrete use case is sample return support. Local resource handling and mobility assets can support sample pickup, preservation, provenance tracking, contamination control, transfer to an ascent or orbital handoff node, and eventual return as verified downmass. Rather than treating sample return as a standalone mission, the architecture treats it as one recurring service that can share infrastructure with feedwater delivery, mobility, power support, and Moon base logistics. This creates a pathway for commercial providers to offer smaller, more frequent, and better-characterized sample return services while reducing the need for each mission to build a bespoke surface logistics chain. The paper discusses service interfaces, functional decomposition, staged ground-test pathways, and the role of measurement in reducing technical, operational, and financial uncertainty. Early tests can focus on sorbent cartridge behavior, release control, fluid conditioning, contamination management, metering accuracy, and feedwater compatibility. Later demonstrations can connect these functions to mobility, power, logistics, or sample handling subsystems, creating a pathway from component validation toward integrated lunar surface utility services. This work is intended as a systems and service architecture paper rather than a claim of mature flight hardware. By organizing volatile capture, storage, distribution, mobility, Moon base support, and sample return logistics into a coherent commercial service stack, the proposed framework supports a transition from one-off lunar missions toward reusable, utility-grade surface infrastructure.

Recently stated NASA goals for establishing a Moon Base by 2035 require unprecedented cargo transportation capabilities. This endeavor will necessitate an estimated 81 total launches to deliver 214 metric tons of payload to the lunar surface. Currently available delivery capabilities have yet to provide cargo masses beyond 500 kg and would require upwards of 1,000 missions to enable NASA’s goals. The combination of these goals and market gaps provides vendors and stakeholders an opportunity to explore new lunar delivery architectures and approaches. This work will investigate a variety of segmented lunar cargo delivery approaches and associated capability ceilings through the lens of space vehicle design and launch limitations. Space vehicles of both tug and landing varieties will be modeled using a validated and verified integrated multidisciplinary design, analysis, and optimization (MDAO) framework. This MDAO framework enables the downstream impacts of vehicle staging, cargo masses, ΔV requirements, transit duration, and the space environment to be captured at the system, subsystem, and cost levels. Vehicle design landscapes will be generated by animating the MDAO framework using strategic designs of experiments (DOE). Characteristic performance and sensitivity relationships will then be extracted via the creation of multivariate surrogate models. This investigation will further explore multi-destination segmented architectures in both lunar and cislunar space. Such a multi-destination mission may include one of, or both, drop-off and pick-up services for various destinations including return to Earth concepts. Exploring segmented architectures from the perspective of vehicle design enables stakeholders and engineers alike an ability to assess the delivery capabilities holistically. Mission design guidance for segmented architectures rendered from the standpoint of design feasibility and cost thus offers practitioners and decision makers realistic expectations for lunar cargo delivery capabilities.

Serious games have demonstrated learning benefits across a variety of contexts and have been applied in aerospace education and professional training. However, usability evaluations of these games often rely on a single assessment and may not capture how users’ perceptions change with repeated experience. Renewed interest in lunar operations is extending the operational theater beyond Earth-centric orbits into cislunar space, a regime governed by chaotic dynamics that remain unfamiliar to many operators trained in space traffic management. The present study evaluated the usability of Project Meneforge, a tabletop serious game developed for the United States Space Force to support understanding of cislunar space dynamics and mission decision making, across repeated gameplay sessions. Thirty-one undergraduate and graduate students from three universities participated in one training session, six gameplay sessions, and a final interview over approximately three and a half weeks. Perceived usability was assessed using the System Usability Scale before gameplay and after each gameplay session, and motivation was assessed concurrently using the Intrinsic Motivation Inventory. In-game performance was operationalized as the total number of Success Points, a game-specific performance metric, earned during each game. Linear mixed-effects modeling examined longitudinal changes in usability, followed by exploratory correlational and mixed-effects path analyses examining relationships among usability, motivation, and performance. Usability varied significantly across sessions, with ratings lowest after the first gameplay session, recovering in the second gameplay session, and then stabilizing. These results are consistent with initial game complexity contributing to lower usability ratings during first-time gameplay, followed by increased familiarity with the game across subsequent sessions. Despite this improvement, the overall mean System Usability Scale score was 57.59, below the commonly cited benchmark of 68. At every session, average usability ratings remained below this benchmark, indicating consistently below-average perceived usability throughout the study. At both the between-person and within-person levels, higher usability was associated with greater perceived competence and greater interest and enjoyment during gameplay. Exploratory path analyses showed that in-game performance positively predicted perceived competence, which in turn positively predicted usability. The direct relationship between performance and usability was not significant after accounting for motivational variables. These findings suggest that stronger in-game performance may contribute to more positive usability perceptions by increasing players’ feelings of competence during gameplay, rather than through a direct effect on usability. Participant interviews also indicated increasing comfort with gameplay over time while identifying specific sources of confusion, such as ambiguous card wording, that were reported consistently across sites. These findings, together with participant and external playtester feedback, informed substantial revisions to the game, including changes to card wording, game-board design, components, rules, and operational realism. In general, these results suggest that longitudinal usability evaluation can help distinguish initial gameplay challenges from persistent design limitations and highlight how motivational experiences relate to usability, thereby informing the development of game-based training systems that better support operator competence and effective decision making in cislunar and other emerging operational environments.

The Environmental Control and Life Support System (ECLSS) is a mission-critical subsystem of any crewed spacecraft, responsible for maintaining a habitable environment during spaceflight. For India's first human spaceflight program, Gaganyaan, rigorous ground-level validation of the ECLSS is imperative prior to crew integration. This paper presents the design, development, and testing of a Human Metabolic Simulator (HMS) — a ground support system engineered to replicate the metabolic activity of a two-member crew within the Gaganyaan crew module, enabling comprehensive functional validation of the ECLSS under realistic operational conditions in the absence of actual crew. The HMS replicates four primary physiological outputs of human metabolism: oxygen consumption, carbon dioxide production, moisture generation, and metabolic heat dissipation, realized through the Oxygen Removal System (ORS), CO₂ Addition System (CAS), Humidity Addition System (HAS), and Heat Addition System (HeAS) respectively. Together, these subsystems drive the module's atmospheric composition and thermal environment to conditions representative of a two-person crew engaged in standard mission activity profiles. This paper focuses on the CO₂ Addition System, which injects controlled quantities of carbon dioxide into the crew module atmosphere to simulate respiratory output of the two-member crew at a combined metabolic CO₂ generation rate of 1.92 kg/day (0.96 kg/person/day), equivalent to approximately 22.2 g/hr per crew member. The CAS architecture incorporates a regulated CO₂ supply, precision flow control valves, pressure and temperature measurements. Key design challenges addressed include safe abort interlocks, redundancy of the components and full compatibility with the module's structural and electrical interfaces. Experimental results from testing campaigns demonstrate that the CAS successfully maintains CO₂ partial pressure profiles within defined operational bands. The HMS as a whole enabled end-to-end functional verification of the ECLSS, providing high-fidelity ground test data for flight model correlation and significantly reducing crew risk during early orbital flight campaigns.

This study develops a multimodal on-body sensing framework for assessing lower-extremity movement and pressure distribution during functional tasks relevant to spacesuit mobility. The synchronized acquisition of segment-level motion and joint-interface pressure enables characterization of the relationship between range of motion and pressure distribution, providing a basis for identifying mobility constraints and informing targeted spacesuit design requirements and individualized countermeasures. The framework integrates APDM OPAL inertial measurement units (IMUs) secured in the thigh, shank, and foot; flexible pressure-sensing matrices positioned over the knees, and an Intel RealSense D435i for kinematic evaluation. By integrating movement characteristics with localized joint-pressure measurements across complete and partial motion conditions, this approach enables quantitative characterization of how variations in movement influence range of motion. This pilot study aims to advance human performance countermeasure development by establishing a wearable, multimodal sensing framework for objectively assessing movement biomechanics and informing spacesuit fit and mobility considerations for future long-duration missions.

Long-duration spaceflight causes musculoskeletal deconditioning, motivating wearable countermeasures that can provide loading beyond scheduled exercise. The Gravity Loading Countermeasure Skinsuit (GLCS) applies distributed passive loading through garment tension, while neuromuscular electrical stimulation (NMES) can selectively activate muscles. Integrating the two requires placing an NMES electrode and rigid stimulator beneath a tensioned garment, where suit pressure and body curvature can change electrode–skin contact and therefore electrical delivery. This paper presents a coupled electromechanical model of this interaction. A garment mechanics model predicts suit tension and skin pressure, a finite-element contact model predicts the portion of the electrode in contact with the thigh, and a three-dimensional volume-conductor model translates the resulting contact state into current distribution and predicted neural activation. The framework is evaluated across garment configuration, contact pressure, electrode geometry, tissue thickness, and anatomical alignment. The results show a tradeoff between electrode contact and local skin pressure. With a garment pocket accommodating the stimulator, only 28–31% of the electrode width is predicted to contact the skin at the suit tension required for skeletal loading. With an unmodified garment, contact increases to 50–54%, but mean pressure beneath the stimulator rises from 13–15 to 24–26 mmHg. Contact loss also affects stimulation differently depending on electrical regulation. Under voltage regulation, it decreases delivered dose, whereas under current regulation it concentrates current within the remaining contact area. The sensitivity analysis identifies regulation mode as the largest source of variability in predicted activation. These results show that NMES electrodes integrated beneath a gravity-loading garment cannot be modeled as having fixed, complete skin contact. The model provides a way to evaluate integrated NMES–GLCS systems in their mechanically loaded, worn state and to guide future prototype testing.

Exposure to a large decrease in ambient pressure, for example when going from a habitat to a low-pressure spacesuit, can result in nitrogen coming out of tissues forming venous gas emboli (VGE) and possibly resulting in decompression sickness (DCS), a known health and safety risk for astronauts when embarking on an extravehicular activity (EVA). VGE can be visually detected in cardiac ultrasound images and appear as bright artifacts within the right atria and ventricle. VGE scores are numerically quantified by subjective human rater assessments of bubble severity on the Eftedal and Brubakk (EB) scale, an 8-point Likert scale from 0 to 7. Typically, assessments are made by 2+ independent graders, and grades are consolidated to a final score via the median of the supplied scores. However, the level of uncertainty between the EB scale assessments per image must be quantified to ensure that this measurable endpoint is reliable. Reliability in these measurements is paramount; VGE severity scores are used in DCS research to investigate the efficacy of prebreathe protocols, designed to mitigate DCS risks for future Artemis astronaut crews. This work characterizes grader agreement in VGE data, leveraging the NASA Exploration Atmospheres dataset as an example. The dataset was down-selected to 1) only include datapoints where 4 graders assessed each image, and 2) remove images where all graders agree no VGE are present (EB score of 0). Ultimately, 4,108 images were assessed for agreement, wherein Fleiss’ kappa indexes reliability and the stratification of scores per EB level was visualized. The score distribution analysis reveals how an image given a particular EB score by at least one grader is rated by the remaining graders. This distribution visually shows discrepancies in EB scale interpretation per EB score. The inter-rater reliability within the example data is κ = 0.272, interpreted as “Minimal agreement”. Generally, agreement was greater in the higher end of the EB scale compared to the lower EB scale categories. Notably, an image with at least one EB 2 score is frequently interpreted by others as EB 0, 1, or 3. The same analysis for EB 4 scores showed improved agreement, where the majority of the given scores were also EB 4, followed by EB 3 scores; however, the range of the given scores covered nearly the entire EB scale (EB 0 to 6). These results demonstrate the importance of quantifying VGE grade uncertainty. Notably, the interpretation of these quantifications is nuanced, as no standards for interpretation can be developed without first understanding the baseline levels of agreement from a range of raters. This work motivates ongoing work to perform a dedicated study to establish such baselines. The eventual understanding of VGE agreement will enable researchers to make robust conclusions regarding their data, ensuring the reliable use of VGE measurements to investigate prebreathe efficacy for mitigating astronaut DCS risks during Artemis EVAs on the Moon and beyond.

Exposure to microgravity eliminates the hydrostatic pressure gradients present on Earth, resulting in a redistribution of body fluids and associated cardiovascular and venous adaptations. Lower body negative pressure (LBNP) has been proposed as a countermeasure to these fluid shifts by generating a pressure gradient that draws blood and body fluids toward the lower extremities. Although LBNP has demonstrated beneficial effects in terrestrial and ground-based analog studies, its physiological effects and optimal dose during true microgravity remain incompletely characterized. Additionally, female-specific cardiovascular responses to graded LBNP in microgravity have received limited investigation. This study evaluates the acute cardiovascular, autonomic, cerebral, and venous responses of female participants to graded levels of LBNP during parabolic flight. Seven female participants completed testing during the 87th and 89th European Space Agency parabolic flight campaigns. Participants underwent baseline measurements under 1 g conditions in a seated posture (without LBNP), and in a supine posture with LBNP levels of 0, –20, –30, and –40 mmHg. The same supine protocol was completed during microgravity exposure achieved through parabolic flight with each LBNP level held continuously during 5 parabolas. A parabola provides approximately 22 seconds of microgravity, preceded and followed by 20 second hypergravity periods of approximately 1.8 g. Cardiovascular variables were continuously measured using a Finapres NOVA, including heart rate, electrocardiography, arterial blood pressure, and cardiac output, which was calibrated using the Innocor CO. Near-infrared spectroscopy and bioimpedance measurements were used to further assess changes in fluid distribution. Ultrasound measurements using the Butterfly iQ3 quantified internal jugular vein cross-sectional area and blood flow. Internal jugular venous pressure was assessed via compression sonography using the Compremium CPMX1 device. Preliminary analyses demonstrate measurable differences in physiological responses to LBNP between 1 g and microgravity conditions. In particular, microgravity produces cephalad vascular changes, while increasing levels of LBNP progressively modify these responses by redistributing blood toward the lower body. These findings provide female-specific data regarding the LBNP dose-response across multiple cardiovascular and cephalad outcomes. Characterizing this response is important for determining whether LBNP can be optimized as a countermeasure to microgravity-induced fluid redistribution and associated cardiovascular changes. These results contribute to the development of evidence-based countermeasure strategies for future long-duration human spaceflight and provide insight into female cardiovascular regulation during acute microgravity exposure.

NASA has identified the sensorimotor/vestibular challenges that astronauts face following a g-transition as a critical risk for upcoming lunar surface missions. The impacts of a partial gravitational environment on sensorimotor function (especially during mission-critical tasks such as landing and EVA) are largely unknown. Crew-based studies offer a very limited sample size, insufficient to fully characterize these risks and develop/test countermeasures, driving the need for effective ground-based analogs to supplement flight studies. Galvanic vestibular stimulation (GVS) has long been used as an effective analog for the vestibular disruption experienced by astronauts. GVS is a technique in which a small (<5mA) electrical current is passed through electrodes placed behind the ear, creating an electric field and altering the excitability of the afferent vestibular neurons. This “random noise” makes the vestibular cues delivered to the brain unreliable, impacting both postural control and manual control tasks. Recent efforts to further improve GVS with head-coupling algorithms and a recent push to integrate GVS with high-fidelity simulations (e.g., lunar lander, manual docking, etc.) offer the promise of a highly effective research and training tool. However, in-house NASA GVS devices are not easily accessible to external researchers, and the few commercially available options are currently inadequate for many research needs, limited by proprietary software interfaces, the lack of a user-facing API, and limited control modes. To address these limitations, we developed a galvanic vestibular stimulation device with a versatile set of control modes and a user-facing API that allows tight integration into existing systems. The device uses a low-side current control architecture and is capable of delivering current up to +/-5mA, with three redundant over-current protection systems to ensure the safety of users. The user can upload a predetermined profile to the onboard SD card or can command the current in real time using the USB serial or wireless UDP interface. Current and voltage measurements from the onboard sensors can be accessed through this API, allowing real-time recording of stimulation data, synchronized with other elements of the system (e.g., joystick inputs, IMUs, etc.). This real-time interface allows researchers to fully integrate GVS into their experiments, improving the quality of data and reducing the risk of operator error during experiments (e.g., forgetting to turn on the device, forgetting to save data, etc.). It also creates the capability for novel applications of GVS for human-in-the-loop simulations, such as adaptive GVS training, where the intensity of the stimulation changes in real time based on user performance, vehicle state, and vestibular modeling. This work was supported by a NASA Space Technology Graduate Research Opportunity.

Suborbital and orbital payloads have strict material restrictions imposed by the harsh environment of space, driving designers to exclusively utilize metallic components for structural assemblies. Metallic components inherently have low mechanical damping, which results in a high gain at the component resonant frequencies. Random vibration during launch and the low damping of electronics enclosures can easily damage sensitive components, significantly increasing risk to all payloads. Small, fully contained pockets of loose, unfused powder in metallic components have been demonstrated to produce internal, dynamic damping structures. Heat treatment is regularly employed to reduce fatigue in AM parts, but the additional treatment may cause the unfused powder to sinter, drastically reducing its performance. This paper investigates the effects of heat treatment on the damping properties of metallic parts with unfused powder cavity geometries.

NASA’s Dragonfly mission aims to land and operate a robotic, car sized rotorcraft on the surface of Saturn’s largest moon, Titan. The surface of Titan poses an extreme cryogenic environment that results in significant materials and engineering challenges. To maintain key internal temperatures and allow for long term operation, Dragonfly required the development of an advanced thermal insulation system. This paper presents a portion of the development of the thermal insulation subsystem, with an emphasis on the material selection process, and the advancement of the design from initial concept to flight ready solution. This development effort began with an assessment of heritage materials used in similar missions such as the Huygens Probe to identify potential insulation materials. Shortly after establishing an initial baseline of ROHACELL® foam, issues such as cracking and deformation began to arise in component level testing, which resulted in the need for a large-scale trade study of thermal insulation materials specifically suited for the Titan environment. Materials considered included polymethacrylimide, melamine, polyimide, and polystyrene based foams, aerogels, vacuum insulated panels, and more. Each candidate material was assessed for its ability to meet mission requirements including thermal, mechanical, and structural performance, mass, contamination, and integration constraints. Various criteria such as thermal conductivity, mechanical properties at cryogenic temperatures, density, manufacturability, ability to withstand changes in pressure, and susceptibility to internal convection were tested to give a direct comparison between candidates. Ultimately, SOLIMIDE® polyimide foam was selected as the preferred insulation material due to the combination of its thermal performance, mechanical properties, lack of internal convection, and relatively straight forward manufacturability. After selecting SOLIMIDE® foam, the development effort progressed from prototype evaluation to establishing the design of a complete flight ready thermal insulation subsystem. This was done through the evaluation of various encapsulation strategies, attachment methods, and box-level thermal and mechanical tests. Design evolution was driven by the need to develop a subsystem that would conform to the external structure of the lander, reduce or eliminate free airflow, maintain resiliency to localized failures, support quick reworks, remain removable throughout spacecraft integration and test, reduce contamination, and provide a uniform aerodynamic outer mold line for the lander. In addition to surviving the surface of Titan, the foam design needed to be robust against all phases of the mission, including launch loads and depressurization, extended vacuum cruise duration, high temperature resistance during entry, descent, and landing, and vibrational and aerodynamic loading introduced during periods of powered flight. The resulting Dragonfly foam subsystem demonstrates the successful development of a wide spread material selection trade study into a flight qualified thermal insulation subsystem that is capable of withstanding the harsh cryogenic environment of Titan. The results presented in this paper provide an example of a materials development campaign and subsequent component level testing designed to overcome the engineering challenges present in operating in extreme environments.

This paper presents the results of single-lap shear tests conducted to evaluate the strength of photopolymer-bonded joints for in situ repair and maintenance applications in space. The primary objective was to compare the shear strength of specimens manufactured under laboratory conditions with those produced in a high-vacuum, space-representative environment. Large Space Structures (LSS) are exposed to a wide range of mechanical and environmental loads throughout their lifecycle. Damage may occur during transportation and launch due to vibration and dynamic loading, during on-orbit assembly through handling or misalignment, and throughout operational service as a result of thermal cycling, fatigue, impacts from micrometeoroids and orbital debris, or accidental contact with servicing spacecraft and robotic systems. Such damage can range from local cracks and delamination to joint degradation or structural failure. In severe cases, individual components or entire structures may become unusable, resulting in costly replacement missions, reduced mission lifetime, and an increased contribution to space debris. Consequently, reliable in situ repair and maintenance technologies represent a flexible and cost-effective alternative to replacing damaged components. Adhesive bonding offers significant potential for the repair, reinforcement, and assembly of LSS. Bonded repair techniques can be employed to restore cracked or damaged structural elements, apply reinforcement patches, strengthen existing joints, or connect structural subcomponents into larger assemblies. Compared with welding, adhesive bonding is less dependent on material compatibility and requires considerably lower thermal and electrical power input, making it particularly attractive for orbital applications where energy resources are limited. Among adhesive technologies, photopolymer adhesives are especially promising for in-space use. Their cationic polymerization can be initiated by ultraviolet radiation and accelerated by heating, enabling rapid curing without the need for complex processing equipment. As liquid, single-component systems, these adhesives can be dispensed and cured either manually by astronauts or autonomously by robotic systems, making them well suited for future in-space manufacturing, repair, and assembly operations. In this study, a high-viscosity photopolymer adhesive was used to manufacture single-lap shear specimens under both laboratory and high-vacuum conditions. To demonstrate the feasibility of automated bonding processes for space applications, specimen fabrication was carried out using a Cartesian robotic system inside a vacuum chamber. Specimen geometry and mechanical testing were performed in accordance with ISO 4587, enabling a direct comparison of the shear strength achieved under terrestrial and space-representative manufacturing conditions.

Series Elastic Actuation (SEA) is a technique for creating highly sensitive actuators in which an elastic force-sensing element is placed in series with the effort output by the actuator. Current state of the art in-space manipulators use strain gauge-based torque sensing for SEA. This incurs various drawbacks: strain gauge-based sensing is highly susceptible to temperature fluctuations, which are inherent and frequent in the space environment, and they lack the increased mechanical compliance that increases the safety, energy absorption, and force sensing resolution of spring-based sensors. A common alternative widely used in terrestrial manipulators and other robotic systems like legged robots and prosthesis is a planar rotary spring (PRS) whose deflection is sensed by a high-resolution rotary encoder. A variety of flexure profiles exist in literature for this class of spring, with a variety of advantages and drawbacks. For each of this class of springs, flexures are constrained to an annulus within the outer diameter of the actuator and without the output shaft. Overall, existing profiles tend to rely upon wide annuli to maintain high amounts of compliance. Moreover, single-part (monolithic) planar rotary springs inherently struggle with energy absorption due to their boundary conditions, and two-part (bilithic) springs require high-tolerance machining and lack axial stiffness. Regardless of profile choice, two aspects of spring design remain consistent: A) longer flexures and B) wider annuli produce springs with higher amounts of compliance and energy absorption. For in-space applications, where mass and space are at a premium, maximally long flexures need to be fit within minimally large annuli. This allows for radially smaller actuators with larger thru-bores for harness. To this end, this work proposes a novel PRS profile which attempts to tightly pack the longest flexure possible within a given annular width by utilizing a space-filling, Hamiltonian path. The paper explores the relevant design parameters and considerations for springs of this type, and demonstrates the benefits of this strategy by designing and manufacturing a representative example PRS. The example spring presented in this paper was designed to meet certain load capacity and sensing resolution requirements through iterative design in a validated FEM in Abaqus. PRS in application are generally manufactured from specialized heat-treated alloys in order to optimize the energy absorption characteristics. The example in this work was manufactured out of grade 5 Titanium, a resilient yet widely available material that does not require heat treatment, more cost and schedule-effective and suited for research and development. Nevertheless, the proposed flexure geometry was able to satisfy the performance requirements. Ultimately, the performance of the spring was verified on a dynamometer. The spring reached 4.81 degrees of deflection at a load of 124Nm, providing a least count of 0.018Nm as installed, or .016% of maximum load.

The Dragonfly Mission is a NASA New Frontiers mission sending a robotic rotorcraft to the surface of Saturn’s largest moon, Titan, to investigate and characterize the moon’s habitability and prebiotic chemistry. Dragonfly is being designed and built under the direction of the Johns Hopkins Applied Physics Laboratory (APL) in Laurel, Maryland. Composite foam tiles made from SOLIMIDE® was selected as the thermal insulation system for the mission, comprising approximately 800 unique encapsulated tiles with a concurrent design and manufacturing approach. Since Dragonfly’s objective is to investigate Titan’s surface chemistry, the mission imposed stringent contamination requirements. Since SOLIMIDE® is an open cell polyimide foam comprised of organic compounds, the machined foam must be encapsulated to prevent contamination of surface samples. With the large-scale production of flight foam, the APL foam team has created and implemented new processes and specifications to support delivery and installation of the foam subsystem to Lander Integration and Testing. With Titan’s harsh cryogenic environment and the variety of materials across the lander including aluminum, magnesium, titanium, composites, etc., one of the main challenges was determining strategies for assembling foam and then how to scale and implement assembly for flight. Assemblies of SOLIMIDE® foam went through a campaign of subsystem level thermal and structural testing to determine the method of attachment to survive Titan’s cryogenic environment. To close out the seams between tiles, Kapton tape is used to prevent any cold external Titan air from reaching the main lander structure. To mitigate the peeling of Kapton seam tape at cryogenic conditions, testing was performed to qualify epoxy to stake the tape in place. As a result of this development and implementation effort, the Dragonfly foam subsystem will be able to meet its hardware delivery to the mission. With foam as a new “soft good technology”, the biggest challenges for Dragonfly and within the industry is implementation for flight due to unknowns and ambiguous development strategies required. Lessons learned from the Dragonfly foam effort can be applied to future missions for scaling up production of emerging technologies, especially within “soft good” technology.

Regolith thermal energy storage (RTES) is a promising long-duration energy storage approach for lunar surface power systems because it utilizes locally available material as the primary sensible-heat storage medium. However, successful implementation depends on understanding how thermal energy propagates through regolith during charging and discharge. This study presents results from AUREVO-PR1, a benchtop regolith-analog experiment developed to investigate transient heat-transfer behavior, validate thermal models, and identify design requirements for future RTES systems. The AUREVO-PR1 test article employed a 100 W cartridge heater, an aluminum core-fin heat spreader, silica sand as a regolith analog, and distributed thermocouples positioned throughout the storage bed. Temperatures were recorded during active heating and passive cooling to characterize thermal propagation, thermal lag, and energy redistribution within the storage medium. Rather than relying solely on experimental measurements, the measured temperature histories are compared with predictions from transient heat-conduction theory using Fourier's Law and finite-difference thermal modeling. This combined experimental and numerical approach provides insight into effective thermal conductivity, thermal diffusivity, and the mechanisms governing heat transport through packed granular media. The outer regolith-analog region reached its maximum temperature approximately 25 minutes after heater shutdown, demonstrating that thermal energy continued propagating through the storage bed even after direct heating ceased. This delayed thermal response highlights the distributed nature of sensible-heat storage and provides experimental evidence of thermal lag within packed regolith. The agreement between experimental observations and thermal modeling establishes a foundation for predicting the performance of larger RTES systems while identifying the importance of embedded heat-transfer pathways, thermal gradients, and insulation in improving storage efficiency. Beyond validating the current prototype, the results establish AUREVO-PR1 as an experimental platform for investigating heat transfer in regolith and validating thermal models for future RTES systems. The results inform the next phase of RTES testing, including improved manifold coupling, sealed helium-assisted heat-transfer experiments, enhanced insulation, output-interface development, and thermal-vacuum testing. Future studies will also investigate alternative fin geometries, embedded heat pipes, and different regolith analog materials while comparing experimental results with transient heat-transfer models. AUREVO-PR1 therefore provides an experimental basis for advancing RTES from a conceptual lunar-storage architecture toward a more testable, predictable, and quantifiable thermal subsystem.

The rapid growth of small satellite constellations has placed increasing emphasis on improving manufacturing throughput, reducing production cost, and shortening development and production turnaround time. Consequently, Assembly, Integration, and Testing (AIT) has become an increasingly important aspect of satellite development, as inefficient AIT activities directly impact production efficiency and program cost. Achieving high-efficiency AIT requires the co-design of AIT procedures, Mechanical/Electrical Ground Support Equipment (MGSE/EGSE), satellite architecture, subsystem placement, structural interfaces, and harness routing from the earliest stages of development. Conventional small satellite architectures often provide limited accessibility to internal subsystems after structural integration. Although permanently integrated hinged panels improve accessibility, they introduce additional mass and structural complexity that are undesirable for mass-constrained missions. Furthermore, inspection, troubleshooting, or minor rework following environmental qualification testing typically requires the removal of structural panels supporting antennas, sensors, connectors, and cable harnesses, increasing dismantling effort while introducing risks of cable pinching, connector damage, repeated harness handling, and sensor misalignment requiring additional verification or calibration. This paper proposes an integration-oriented satellite architecture, demonstrated using an in-house developed 60 kg engineering model satellite. The proposed approach co-designs the structural layout, subsystem placement, harness routing, and ground support equipment through four complementary design framework. First, critical debug interfaces and charging ports for both the satellite bus and payload are centralized on a single satellite face to simplify EGSE connections, reduce setup time, and minimize human errors. Second, dedicated harness routing and tie points ensure consistent cable routing, improve assembly repeatability, and facilitate modular harness installation. Third, a skeletal support interface retains externally mounted antennas, sensors, connectors, and cable harnesses during panel removal, preserving harness routing and sensor alignment while reducing dismantling effort and assembly-induced damage. In the demonstrated engineering model, no panel-mounted sensors or cable harnesses required removal to access internal subsystems, unlike conventional architectures. Although this approach introduces additional structural complexity and a modest mass penalty, it can be selectively implemented in areas requiring frequent access to balance structural mass against integration efficiency. Finally, a removable double-hinge MGSE separates the satellite into top and bottom decks, providing direct access to internal subsystems while minimizing harness disconnections, maintaining satellite operability through EGSE harness extensions, reducing harness bending stress, and eliminating any flight mass penalty. The engineering model demonstrated complete satellite integration by a single operator without additional hoisting equipment, a capability that is difficult to achieve using conventional approaches. The double-hinge mechanism enabled a 180° opening angle, exceeding the practical opening angle of conventional single-hinge arrangement. The paper further discusses how these design principles establish a foundation for both manual and automated assembly and integration. A motorized double-hinge MGSE, together with standardized harness routing, dedicated tie points, and modular harness assemblies, enables repeatable workflows suitable for both human operators and robotic manipulators, providing a practical pathway toward automated assembly and integration for the scalable and cost-effective production of future small satellite constellations.

CubeSats up to 3U typically generate between 10 and 30 W of power through solar panels, depending on the body mounted or deployed solar panel configuration which is employed. Thanks to high-capacity batteries, some satellite platform developers provide up to 200 W of payload power. However, given that the external surface area of a 3U CubeSat remains constant, the satellites’ thermal regulation capacity may not accommodate the increased heat dissipation requirement. In this paper, we explore the impact of high-power payloads on a 3U CubeSat from a thermal perspective. We first investigate the influence of payload power on the payload temperature profile. The results demonstrate that even short durations of payload activity can significantly heat up the satellite beyond its operational limits. We then analyse the impact of common passive thermal management strategies on a satellite’s thermal behaviour, namely: (i) thermal couplings across components; (ii) external faces’ optical surface properties; and (iii) integration of heat storages. The results demonstrate the wide range of thermal behaviours which can be expected from the integration of various passive technologies.

The NSF Center for Space, High-Performance, and Resilient Computing (SHREC) at the University of Pittsburgh has a growing history of deploying space computing experiments to the International Space Station. Graduate student researchers in SHREC have built systems that were deployed on the DoD Space Test Program (STP) Houston's H5, H6, and H7 pallets, namely, STP-H5-CSP, STP-H6-SSIVP, and STP-H7-CASPR. After four successful years operating STP-H7-CASPR, SHREC students have developed another experiment for the STP-H12 pallet, called the Visual And Neuromorphic Tracking and Geosensing Experiment (STP-H12-VANTAGE). The H12 pallet will sit at the COL-EPF-SDX (Columbus External Payload Facility Starboard Deck X-Direction) location, which gives VANTAGE a view of Earth's surface. The experiment's primary objective is to demonstrate sensor fusion between a traditional RGB imager and a neuromorphic event-based sensor. In addition, a set of radiation-tolerant computing resources is featured, including the space-qualified SHREC Space Processor (SSP) subsystem as head node, and two new accelerator subsystems. These accelerators include a radiation-tolerant Nvidia Jetson Orin NX GPU subsystem and a radiation-tolerant Microchip PolarFire FPGA SoC subsystem. In all three, a novel mix of COTS and RadHard components plus methods from dependability theory work together to achieve high performance and reliability with low size, weight, power, and cost. This paper documents both the design of the spacecraft and the verification procedures employed to ensure positive failure margins for the bolted joints and the chassis structural panels. This research takes multiple approaches to modeling the bolted joints using a variety of element types that capture different behaviors more accurately. The predictions for the system's modal characteristics from each of these variations are compared against real data obtained during workmanship vibe testing. The structural results are also compared to each other for stress analysis in the panels and the bolted joints themselves, following procedures compliant with NASA-STD-5020B. Finally, this work documents the thermal management design and compares simulations with data from the workmanship thermal cycling test program. Simulations of on-orbit thermal conditions are not presented due to a lack of detail about the overall system boundary conditions from the STP-H12 pallet and the surrounding ISS structure. This research can help guide future developers of similar small spacecraft experiments, especially students in a similar position to those in SHREC. Unlike companies and agencies like NASA whose expertise grows with career employees, student-centered institutions like SHREC are burdened by the constant turnover of its researchers, which inhibits, to some extent, the growth of institutional expertise for such complex projects as the development of spacecraft. Documentation like this research serves to lower the burden on new spacecraft developers by sharing the procedures that fulfilled the rigorous verification requirements for STP-H12-VANTAGE.

As space agencies and commercial partners transition towards long-duration lunar operations, ensuring the survivability of surface infrastructure, components, and hardware is essential to supporting a sustained human presence. During landing and ascent, rocket exhaust plumes interact with planetary regolith, entraining abrasive and erosive particulate matter into high-velocity, multi-phase flows. This phenomenon, known as Plume–Surface Interaction (PSI), subjects nearby landers, habitats, and scientific instruments to extreme kinetic and thermal loads imparted by high-velocity regolith particles. To effectively reduce mission risk and validate structural materials for these harsh extraterrestrial environments, comprehensive ground-based testing is required. This paper details the utilization of the Integrated Material–PSI Assessment and Component Testing (IMPACT) facility to experimentally simulate and characterize aerospace material degradation under high-fidelity, PSI-like erosive particle conditions. The IMPACT facility provides a controlled environment designed to assess material durability under plume-driven particle erosion. Utilizing a low-pressure inductively coupled plasma spray facility with a custom supersonic convergent-divergent nozzle, the system accelerates heated regolith simulant (sieved to a 100–300 μm nominal diameter) to particle velocities exceeding 340 m/s. To evaluate material survivability, degradation, and the effects of parametric changes in operational variables, a series of aluminum, titanium, and copper foil specimens were tested to establish baseline damage responses and develop post-exposure characterization workflows. Surface characterization of specimens after IMPACT testing revealed the rapid onset of material degradation caused by these simulated environments. Even 5–10 seconds of direct exposure to the supersonic plume produced topographical changes relative to the pristine baselines with both erosive damage and adhered regolith. Post-test surface diagnostics, including Backscattered Electron (BSE) microscopy, Energy Dispersive Spectroscopy (EDS), optical microscopy, and surface profilometry, revealed rapid surface roughening, cratering, pitting, and measurable volumetric erosion directly correlated with the simulated particle size envelope. In more ductile baseline materials such as copper foil, through-thickness penetration was observed. Characterizing these degradation modes is a critical step in the mission validation pipeline for sustained operations in extreme environments like the lunar surface. By identifying the mechanisms by which baseline aerospace materials wear, yield, and degrade under simulated lunar PSI conditions, engineers can develop and qualify robust mitigation strategies. Material coatings, thermal protection systems (TPS), and flight hardware can be evaluated under representative PSI conditions to adapt designs to the challenges posed by high-energy regolith transport, thereby increasing the likelihood of long-duration mission success. In parallel, integration of new diagnostic chambers and sample frameworks will enable in-situ measurements of plume geometries, particle speeds, and particle temperatures, both average and particle-based. The upgraded test frame incorporates a cooled sample plate with adjustable angle of incidence and supports a wide range of coupon geometries and components with minimal chamber modification. Ultimately, the high-fidelity data generated by the IMPACT facility provide critical insights required to select and validate materials for surface operations, thereby ensuring the longevity, safety, and resilience of critical infrastructure for sustained planetary exploration.

The rapid growth of high-performance computing and artificial intelligence payloads in low Earth orbit introduces a thermal-management challenge that is difficult to address using conventional steady-state radiator sizing. This work investigates an orbit-synchronized thermal management architecture that combines phase-change material (PCM) thermal energy storage with time-varying heat rejection through both a dedicated radiator and the backside of spacecraft solar arrays. During periods of high computational load or unfavorable radiative conditions, excess GPU waste heat is stored within a PCM thermal battery, reducing the instantaneous heat-rejection requirement imposed on the primary radiator. During orbital eclipse, when direct solar heating is absent and photovoltaic array temperatures decrease, a switchable thermal transport path transfers stored heat to the backside of the solar arrays, allowing existing spacecraft surface area to function as an auxiliary radiator. A transient lumped-parameter thermal model is developed to capture GPU heat generation, PCM melting and solidification, radiator heat rejection, solar-array thermal response, Earth infrared loading, albedo, and orbital illumination conditions. Parametric simulations evaluate the influence of PCM mass, phase-transition temperature, effective thermal conductivity, radiator area, solar-array area, computational duty cycle, and payload heat generation on maximum GPU temperature, PCM state of charge, recharge time, and sustainable computing duty cycle. The resulting design space is used to quantify the trade between dedicated radiator area, thermal-storage mass, and reusable spacecraft surface area. This approach reframes high-power spacecraft thermal management from a peak heat-rejection problem toward an orbit-averaged thermal-energy management problem and provides a framework for evaluating thermal batteries for future space-based computing platforms.

Cultivating crops in microgravity is a critical requirement for sustaining human life during long-duration space travel and eventual planetary colonization. However, managing fluid distribution and anchoring biological specimens in the absence of gravity presents significant engineering hurdles. In microgravity environments, traditional gravitational drainage is absent, causing fluids and small particles to behave unpredictably under the dominant influence of surface tension and capillary forces. This paper evaluates the engineering principles and structural efficacy of using nylon mesh configurations to encapsulate and stabilize seeds during transit, handling, and germination phases. Nylon mesh offers a unique combination of material properties that directly address spaceflight constraints. Mechanically, its porous architecture provides a robust, lightweight passive restraint system that secures seeds against vibrational torque, launch G-loads, and human handling without adding prohibitive mass to the payload. From a fluid dynamics perspective, nylon facilitates controlled capillary action. This allows for uniform, passive moisture transport across the seed matrix while preserving open pore networks necessary for root zone aeration, thereby preventing fluid stagnation and root rot. Furthermore, the material's structural flexibility and strength-to-weight ratio allows for easy compression into small volumes during transport and later, smooth transition to usage in orbit. While nylon mesh could not be actively tested in microgravity during the Student Spaceflight Experiments Program (SSEP) International Space Station (ISS) experiment, rigorous evaluation reveals critical integration risks that must be addressed for spacecraft deployment. These safety hazards include electrostatic discharge (ESD) caused by nylon’s electrical insulation properties, high flammability and toxic outgassing in oxygen-rich cabin environments, volatile organic compound (VOC) generation, and life-support filtration clogging from microplastic fiber shedding. This study analyzes these risks alongside passive mitigation strategies to outline a standardized flight-qualification framework for polymer-based seed handling systems.

Oxygen-containing propellants derived from in-situ resources are attractive for future electric propulsion systems because they offer a pathway toward reducing dependence on Earth-supplied propellant. Oxygen-containing plasmas are also relevant to emerging air-breathing electric propulsion concepts operating in very low Earth orbit. These environments, however, are chemically reactive and can modify plasma-facing metallic surfaces through oxidation, material removal, and ion-enhanced surface reactions, potentially altering both component integrity and electrical performance. This work presents a preliminary investigation of SS316 and Inconel 625 during oxygen-plasma exposures lasting up to 24 h in a low-pressure oxygen inductively coupled plasmas. A first-order electrical model was developed in which the evolving oxide layer is represented by an effective resistance determined by oxide resistivity and thickness. The model couples this resistance to the plasma-facing surface potential, ion impact energy, and collected current, providing a framework for interpreting electrical degradation caused by surface oxidation. Samples were exposed under two discharge conditions and multiple applied biases to vary nominal ion impact energy. Plasma properties were characterized using a single Langmuir probe, while time-resolved current measurements tracked electrical evolution during exposure. Pre- and post-exposure mass measurements quantified net material loss or gain, and X-ray photoelectron spectroscopy depth profiles were used to assess the persistence of oxidation beneath the exposed surface. SS316 remained electrically stable despite oxidation extending throughout the measured XPS depth range, while net material removal emerged as the more significant surface-integrity concern. Inconel 625 exhibited a stronger dependence on discharge condition, transitioning from removal-dominated behavior with little current degradation to net mass accumulation, greater oxide depth, and substantial current reduction. Across both materials, oxidation and material removal occurred concurrently, and neither ion energy nor oxide presence alone predicted electrical degradation. These results identify discharge condition as an important parameter governing long-duration material response, and should be considered explicitly when evaluating plasma-facing materials for oxygen-based electric propulsion systems.

Spacecraft propulsion plays essential role for all space-based missions, from insertion of satellites in the desired orbits to interplanetary and the deep space exploration missions. Hypergolic Chemical Propulsion is preferred, due to its high energy characteristics and simpler system configurations. However, most chemical propellants used in propulsion are toxic and carcinogenic and hence put a negative footprint on environment. Green Propellants are propellants, which can be used as replacement for the conventionally used chemical propellants, while being less hazardous, less toxic, non-carcinogenic, overall providing safer handling. The hypergolic green propellant combination is favoured in spacecraft propulsion owing to its repeatable usage and comparable performance characteristics. A thruster design was taken up for green bi-propellant combination developed in ISRO. The design studies took into account the unique properties and requirements associated with these green propellants, while trying to optimize the combustion efficiency, thruster performance and thermal characteristics of the system. The study delved into various injector designs, starting from impinging type to assess the preliminary combustion characteristics and effect of the propellant's varied properties on the same. After establishing a preliminary reference, swirl type injector design was adopted to handle the intricacies of the thruster performance. The swirl type injector provided better atomization properties, which led to refined combustion. The various components of the swirl injector were designed to optimize the atomization and achieve the required flow-rate at the rated pressure drop. Multiple swirl type injectors with varied component design were realized to study their variations as well. The thruster was subjected to sea level developmental tests with both types of injectors. With the impinging type injector, while hypergolicity could be established, the performance achieved was of the order 70% C* efficiency. The introduction of swirl type injector led to a betterment of C* efficiency with value up to 85% being achieved. Based on the preliminary analysis of the tests, further HAT level developmental tests were carried out, and over a series of tests the operation of the thruster with the propellant combination was optimised. The effective mixture ratio to optimise efficiency was found based on a series of tests investigating the performance of the thruster over a range spanning 2.8 to 6. The efficiency was optimized to 92% with conduction of the HAT tests at various mixture ratio conditions. The propellant combination also required specific operating environment and highly controlled ambient conditions for consistent operability. The requirement of high vacuum levels and its effect on the performance as well as the repeatability were studied over a vacuum level up to 40mbar. The paper deals with the design as well as the testing aspects of the green propellant thrusters, from its conception to the characterisation of the performance and establishing ideal operating regime and environment for optimum operation of the thruster.

This paper describes the design, development, and qualification testing of a tri-modal throttleable Reaction Control System (RCS) utilizing high-test hydrogen peroxide (H2O2) and refined kerosene (RP1) propellants for spacecraft attitude control and maneuvering. The VRP110-B RCS design, employs a pressure-fed propulsion architecture with three operational modes delivering 42, 65, and 113 pounds of thrust respectively. The system incorporates proprietary catalyst technology enabling 15+ mission cycles and patented VORTEX™ combustion technology that enhances thermal management and operational longevity. A comprehensive qualification test campaign was executed following MIL-STD-810 and SMC-S-016 standards to validate performance and reliability under simulated spaceflight conditions. The test regime included acceptance hot-fire testing, pyrotechnic shock exposure (100-10,000 Hz), random vibration (20-2,000 Hz), thermal cycling (-44°C to +92°C), environmental exposure, and extended service life evaluation. Qualification hot-fire testing encompassed post-environmental performance verification, run-box testing, blowdown testing, continuous operation (20+ minutes), hot restart, and saturated propellant testing validation. Service life testing accumulated over 45,000 pulses and 9,300 seconds of run time over 30 days, consuming 3,300 pounds of propellant. All qualification tests were completed successfully with no failures followed and thruster performance stayed within allowable limits. This RCS test campaign has demonstrated flight qualification readiness. Additionally, the use of H2O2 as a green, storable propellant offers significant safety and environmental advantages over traditional hypergolic systems while maintaining high reliability for reusable spacecraft missions.

Fluid hammer, a transient phenomenon caused by sudden changes in fluid flow, has been a long-standing concern in the design and operation of spacecraft propulsion systems due to its catastrophic effects. These systems face unique challenges due to high velocity flows, and rapid pressure variations. Understanding and mitigating fluid hammer is crucial for ensuring the reliability and safety of propulsion systems in the harsh environment of space. This paper explores the mathematical modelling of fluid hammer in spacecraft and discusses its implications on the propulsion system Fluid hammer in a propulsion system arises from abrupt valve operations, engine start, and shutdowns, as well as the interaction between liquid propellants and gases. The resulting pressure waves can exert considerable stresses on the spacecraft's structural components and disrupt the stability of the propulsion system, potentially compromising mission success and safety. Modeling such critical phenomenon is as important as qualifying the system for flight usage. The mathematical modelling of fluid hammer involves solving the fluid dynamics equations for compressible flows in closed systems with complex geometries and varying boundary conditions. This paper presents a comprehensive overview of the numerical methods and computational tools used to simulate and predict fluid hammer events in spacecraft propulsion. It covers the development of high-fidelity models that consider factors such as fluid properties, system geometry, and transient behavior of dynamic components. A conventional bi-propellant system with N2O4/MMH as propellants is being used as the standard spacecraft propulsion. A model was generated at the Liquid Propulsion Systems Centre, ISRO to simulate the pressure surge transience of a propulsion System with 5 Main Engines and 16 Reaction Control thrusters. A simulation model to capture the transient surge behavior was generated. Numerous tests were done at ISRO to demonstrate the performance of the propulsion system and the predictions made by the model were validated using these test results. Mainly, the peak fluid surge pressure and its frequency during different combinations of valve closures various mission scenarios of flight were validated. The peak surge pressures were 44bar during the initial configuration. Various strategies for fluid hammer prevention and mitigation were studied and implemented like staggered engine shutoff and modifying the feed line diameter helped in limiting the surge in pressure to 32bar which is within acceptable limits of the system. The paper explores the potential for design modifications, and operational procedures to enhance the resilience of spacecraft propulsion systems against fluid hammer events.

The global satellite market is projected to expand sevenfold by 2035 [1]. Such rapid expansion demands scalable, high-performance power systems. The photovoltaic (PV) industry is well-positioned to continue supplying reliable power to space missions and to scale as needed. This paper presents a techno-economic analysis of PV for space, including projected demand, manufacturing costs, and performance of the leading space PV technologies. It also reviews global government support for space PV research and space-based solar power programs, supply chains, and U.S. production of PV cells and power assemblies for the space market. The primary solar cell technologies and candidates in this study include multijunction III-V cells, silicon, perovskites, and silicon/perovskite tandems. This paper reviews the key performance requirements for PV cells in space and traces the value chain for these technologies from cell fabrication to coverglass/interconnect/cell (CIC) assembly. III-V multijunction cells remain the industry standard for Geosynchronous Equatorial Orbit (GEO) satellites, defense, and deep-space missions due to their high efficiency, industry-leading mass-specific power (W/kg), and radiation resilience. Nonetheless, widespread adoption of III-Vs in emerging space markets is limited by material supply constraints and the highest manufacturing costs among space PV options. Current space-grade III-V cell costs often exceed $150/W [2]. Current global manufacturing of III-V space cells is estimated at only about 2 MW annually [3]. This paper discusses III-V scaling bottlenecks, quantifies current III-V performance and cost benchmarks, and provides cost model results for technology advances, including substrate reuse, improvements in epitaxy, and cell metallization. Silicon cells leverage established supply chains for cost optimization and serve Low Earth Orbit (LEO) satellites. However, silicon cells have lower mass-specific power than III-Vs (primarily due to lower cell efficiency and thicker absorber layers) and suffer rapid radiation degradation that requires heavy protective shielding. Emerging silicon/perovskite tandem cells offer potential for flexible array architectures, leverage scalable manufacturing solutions from silicon and thin film PV, and could lead to improved mass-specific power relative to silicon or perovskites alone —yet these cells also face deployment and reliability challenges in long-term vacuum and under extreme thermal cycling [4]. This paper quantifies the current performance, reliability, and cost benchmarks for silicon, perovskites, and silicon/perovskite tandems, as well as cost model results for technology advances, including precision grinding of silicon bottom cells (which helps reduce weight) and robotic CIC assembly. The paper concludes by summarizing the analysis of PV technologies in terms of their relative performance, reliability in the harsh operating conditions of space, and costs. It also summarizes leading opportunities to further advance PV satellite power systems. [1] Space Power, https://peraspera.us/space-power/ (2026) - citing Rocket Lab: “global satellite market is projected to expand sevenfold by 2035” [2] A Techno-Economic Analysis and Cost Reduction Roadmap for III-V Solar Cells, https://docs.nlr.gov/docs/fy19osti/72103.pdf (2018) [3] Prepare for launch: Solar powers the $600 billion space industry, https://pv-magazine-usa.com/2026/04/24/prepare-for-launch-solar-powers-the-600-billion-space-industry/ (2026) [4] Perovskite Solar Cells for Space Applications: Progress and Challenges, https://pubmed.ncbi.nlm.nih.gov/33861877/ (2021)

Developing liquid rocket engines at a collegiate level is a task that requires finding cost-effective and creative ways to solve complex problems . Many of the vital components used in professional engines are either incredibly expensive or difficult to acquire. To address one of the most critical challenges in liquid rocket propulsion, thermal management, the Arkansas State Space Program developed a reusable liquid rocket engine that incorporates an actively cooled, water-based cooling jacket. Unlike most university-scale engines, the cooling jacket was produced using metal casting, allowing for a repeatable manufacturing approach suitable for iterative development. At max capacity, the system was designed to produce 360 lbf of thrust at a chamber pressure of 150 psi utilizing gaseous oxygen and ethanol. A specialized injector was developed to work towards maximizing mixing between multiphase propellants, combining both axial and impinging injections. ANSYS Fluent simulations were performed to quantify the discharge coefficients, pressure drops, and exit velocities across the injector. For the fuel line, the simulated discharge coefficient was 0.500, with a pressure drop of 28.469% and an exit velocity of 62.08 ft/s. For the oxygen line, the simulated discharge coefficient was 0.284, with a pressure drop of 47.551% and an exit velocity of 355.28 ft/s. These results closely matched the theoretical calculations, providing confidence in the injector design prior to testing. Cold-flow tests were then conducted to characterize the injector spray by measuring the breakup distance, cone angle, and penetration length. A complete avionics system was designed and built in-house around an ESP32 microcontroller. The system controls engine actuators through relays while acquiring sensor data using HX711 analog-to-digital converters. A software was developed by the students to control all the systems as well as visualize the sensor data. A routine module was added to automatically control multiple valves along with the ignition system. An electronically controlled ignition system was implemented utilizing a squib as the primary component. Additionally, a portable test stand was also designed and manufactured to test different types of rocket engines. Open flame injector tests were conducted, validating proper flame formation for a full hot fire test. Cold flow tests were successfully completed with some avionic complications identified along the way. Although a full hot-fire test was not completed due to avionics challenges and project time constraints, the overall development effort established a functional and reusable liquid rocket engine platform. This paper describes the overall design, manufacturing, testing, and integration process of the propulsion system while documenting the main challenges along the way. The lessons learned provide a framework which other universities can adapt and improve upon to develop their custom reusable rocket engines. It also provides insight to what changes could improve the design and what different decisions could be taken to facilitate integration.

Radioisotope power systems (RPS) have enabled spacecraft missions in environments where conventional power sources are impractical or insufficient. Integrating an RPS with a spacecraft requires more than demonstrating power-system performance; engineers must establish that available evidence is applicable and sufficient for the specific hardware, its lifecycle history, and the integration decision being supported. This paper examines historical RPS hardware assessment practices and the engineering evidence used to support technical concurrence for spacecraft integration. It organizes that evidence into five domains: operational readiness, anomaly resolution, traceable flight configuration, performance and predictive capability, and nuclear safety. Representative assessment practices illustrate how multidisciplinary evidence is evaluated collectively to establish a technically defensible basis for readiness decisions. The paper also examines the plan for assessment of the refurbished General-Purpose Heat Source Radioisotope Thermoelectric Generator (GPHS-RTG) Flight Unit 5 (F-5R) as a contemporary example of tailoring an assessment to hardware with an extended and atypical lifecycle history before assignment to a receiving mission. The analysis shows that assessment implementation can be tailored to the hardware and decision context while preserving the underlying engineering need for sufficient applicable evidence, appropriate technical insight, and multidisciplinary engineering judgment.

For 65 years, the United States has used Radioisotope Power Systems (RPS) as an enabling technology to execute deep-space science and exploration missions. In 2027, the Voyager-1 spacecraft surpasses 1 light day of distance and completes 50 years of continuous flight operations, demonstrating the resilience of Plutonium-238 (Pu-238) fueled Radioisotope Thermoelectric Generators (RTGs). Managed by NASA in partnership with the Department of Energy (DOE) Office of Nuclear Energy, the RPS Program sustains national space nuclear capabilities while advancing technologies required for extreme planetary environments. Recent accomplishments The RPS Program has fully transitioned from a mission-driven procurement model to a Constant Rate Production (CRP) framework. DOE manages CRP for ongoing production of Pu-238 heat source material and subsequent manufacturing into its final flight form. The program is on track to achieve its target production metrics: manufacturing an average of 10 to 15 fuel clads annually and scaling the production of heat-source Pu-238 oxide to an average of 1.5 kg per year in 2026. The program is actively facilitating system delivery and integration for two major upcoming exploration initiatives slated for launch in 2028: • Dragonfly Mission (Titan): Engineering integration of a single Multi-Mission Radioisotope Thermoelectric Generator (MMRTG) and 24 Light Weight Radioisotope Heater Units (LWRHUs). • Rosalind Franklin Mission (Mars): facilitating the delivery of 24 LWRHUs and a provisional UK developed Americium-241 Radioisotope Heater Unit. Future RPS Program Direction and Organization. In March 2026, the NASA Administrator began the Moon Base initiative to increase deliveries to the lunar surface. Moon Base is a long-term lunar exploration and infrastructure initiative designed to enable sustained human presence and expanded scientific and commercial activity on the lunar surface. Technologies baselined for Moon Base include a Fission Surface reactor for power and deployment of RHUs to help systems survive the extreme conditions of the lunar south pole. In May 2026, the Administrator announced his intent to reorganize the directorates by combining Aeronautics Research (ARMD) and Space Technology Mission Directorate (STMD) into a single Research and Technology Mission Directorate (RTMD), to better align NASA’s aeronautics, space technology, and nuclear power and propulsion capabilities into a single organization. In June 2026, the RPS Program received direction from NASA’s Planetary Science Division to revector the future generator development due to priorities within NASA. This revector included direction to close out the Next Gen RTG task at DOE, initiate the build of two copies of a Multi-Mission Radioisotope Thermoelectric Generator (MMRTG), and initiate pre-formulation of a new High Efficiency Generator. The main goal of the revector is to better utilize the Pu-238 production rates which were established based on the assumption of having high efficiency Stirling radioisotope generators in the RPS Program. Recent advances in Stirling conversion technology as well as interest by both U.S. and international industry and government agencies, is pioneering a path forward to integrate Stirling converters into radioisotope power system use in space.

When the first flight unit of the multi-mission radioisotope thermoelectric generator (MMRTG) launched on Curiosity in November 2011, it was a brand-new design. Because of its unknown future performance, it sparked interesting conversations during the early phases of the mission. With a design life of 14 years, predictions of end-of-design-life (EODL) varied wildly. Today, F1 has exceeded EODL, and EODL power is known to be 73.2 We. MMRTG power has strong dependencies on spacecraft use and local environment. So, this measured EODL power is representative of conditions similar to the Curiosity rover on Mars, with other conditions being predictable if the environment is known. Of all the methods used to predict MMRTG power, the rate law analysis proved to be, far and away, the most accurate. When it was first introduced, the rate law method predicted power of 75.2 We which was more than 25% higher than other reported methods. Based purely on empirical data, the rate law method has been used to monitor F1 for almost a decade. Today, we now know that MMRTG experiences a change in the thermoelectric degradation mechanism, which changes the rate of power loss experienced by the generator. The slight change in performance (~3% at EODL) is likely trivial for mission planners and operations but speaks strongly to the accuracy and potency of the rate law method. Analysis of F1 today shows that extrapolations performed over 4 years ago, after the mechanism change, are still accurate. With the caveat that a future, unexpected mechanism change could alter the results. Fortunately, Pioneer 10 and 11 used the same thermoelectric system as MMRTG, and they only experienced 1 mechanism change in their 30 and 22 years of operations, respectively. This suggests that only one mechanism change in MMRTG is anticipated. Analysis also shows that F1 is expected to reach 50% power after 21 years and 1 month, with some variation for seasonal influence on power. Telemetry from F2 is also assessed. A detailed extrapolation is not performed here because there is no evidence that F2 has experienced the mechanism change observed in both F1 and Pioneer. This is likely because F2 experienced less intense thermal and storage conditions. A mechanism change is expected in F2, but when it will occur is difficult to predict. Despite not having a mathematical prediction of F2 performance, comparative analysis with F1 shows very similar behaviors, including the effects of Martian seasons on power. This suggests that F2 EODL will be similar to F1 EODL power.

Planning for the Mars 2020 mission began in 2014. The Radioisotope Power Systems (RPS) Program worked with National Aeronautics and Space Administration’s Jet Propulsion Laboratory, the mission lead, and the Department of Energy’s (DOE) Nuclear Energy’s Infrastructure Programs and DOE’s Idaho National Laboratories. From 2014 until the mission launch in July of 2020, there were many changes to the budget that resulted in significant savings to the government. This paper describes why the RPS Program tracks the aeras of cost, what those areas represent, and the changes that resulted in approximately 20% savings.

Surviving the lunar night remains one of the primary challenges for sustained surface operations, driven by extended darkness, extreme cold, and limited energy availability. This paper compares key survive-the-night technologies that can enable spacecraft, landers, rovers, and surface payloads to endure long-duration eclipse or nighttime conditions on the Moon and other planetary bodies. The comparison includes approaches based on batteries, fuel cells, radioisotope power systems, and hybrid architectures, with attention to their relative benefits, limitations, and mission applicability. Particular emphasis is placed on evaluating these technologies in terms of power availability, thermal performance, mass and volume impacts, operational complexity, system lifetime, and integration with different mission architectures. The paper also considers environmental and design drivers such as lunar thermal extremes, dust exposure, eclipse duration, latitude-dependent lighting conditions, and operations in polar or permanently shadowed regions. By assessing the trade space across multiple survival strategies, this work identifies key considerations for selecting robust and efficient technologies to support future long-duration lunar surface missions.

Colonization of Mars will require aircraft for various missions. Raymer et al. designed the Raymer Manned Mars Airplane. This concept airplane is optionally manned and can take off and land vertically. It uses electric motors and propellers for thrust in forward flight. As the RMMP was developed by a volunteer consortium, several aspects of the design require further detailed study. Propeller design is one such aspect, as the RMMP operates in a low-density, low-Reynolds-number environment where aerodynamic performance is significantly degraded, strongly affecting propeller efficiency. Compared to airfoils optimized to operate at higher Reynolds numbers, low Reynolds number airfoils have degraded performance due to laminar separation bubble formation. Hence, airfoils operating at low Reynolds numbers must be specifically optimized for such conditions. In this study, the RMMP propellers are optimized using Blade Element Momentum Theory. The impact of propeller optimization on range and mission capability will be assessed. This study aims to provide guidelines for future propeller design in low-density atmospheres such as Mars. For optimization, a set of initial design parameters was selected. In the final study, optimized design parameters will be presented along with improved capabilities of the RMMP. BEMT analysis was conducted for one propeller. The initial design parameters are selected for a simple propeller design. In the optimization process, taper, twist, and custom airfoil designs will be considered. Four airfoils were considered in the initial study: NACA 0012, Selig S1223, E387, and SD7037. The aerodynamic properties of the airfoils were obtained from XFOIL. The XFOIL analysis covers various flow conditions along the blade by considering flow conditions at various Reynolds numbers and angles of attack. Lift and drag coefficients obtained from XFOIL were then imported to the BEMT code. The results obtained from BEMT analysis indicate that beyond an advance ratio of 1.5, propellers with NACA 0012 airfoil start to produce negative thrust. On the other hand, airfoils designed specifically for low Reynolds numbers performed better than NACA 0012, as expected. Among low Reynolds number airfoils, S1223 performs the best in the analyzed advance ratio range (J=1.25-1.8). S1223 and other low Reynolds number airfoils maintain positive thrust at higher advance ratios, proving the superiority of these airfoils in the Martian atmosphere compared to NACA 0012. Propeller efficiencies provide further insight into airfoil performance. S1223 sustains positive thrust across a wider advance ratio range, in which it maintains higher efficiency throughout. Conversely, NACA 0012 consistently shows the lowest efficiency in positive thrust scenarios, further emphasizing the necessity of specialized airfoils for efficient thrust production at low Reynolds numbers. So far, BEMT code was developed to analyze RMMP propellers. The initial analysis indicates the intrinsic low Reynolds number flow challenges that need to be overcome for the propellers, and the results indicate that conventional airfoils such as NACA 0012 are not suitable for Martian conditions. Work is currently underway to improve the design by taper, twist, and custom airfoil designs.

The AGARD 445.6 wing is a widely established benchmark for aeroelastic analysis and flutter prediction; however, its transient response following short-duration aerodynamic disturbances has received comparatively less attention. This study investigates the transient aeroelastic behaviour of the orthotropic AGARD 445.6 weakened Model 3 wing using two-way CFD–FSI coupling in ANSYS Workbench. Following validation of the structural model against published modal data, coupled simulations were performed at M=0.001 and M=0.96, both at 0^(∘ )angle of attack, under an identical short-duration pressure disturbance. The transonic case exhibited substantially greater deformation and a pronounced time-dependent mean bending and twist response, with the bending displacement reaching approximately -63.6 mm. Despite this strong transient response, the late-time oscillations became progressively decaying and were dominated by a frequency of approximately 9.77 Hz, closely corresponding to the first structural mode. Damping analysis confirmed a positively damped late-time response at M=0.96. The results demonstrate that transonic aerodynamic loading can strongly alter the magnitude and mean evolution of the transient aeroelastic response without necessarily producing sustained aeroelastic instability. No sustained or growing oscillations indicative of flutter were identified at M=0.96 within the simulated interval. The study provides a basis for extending transient two-way FSI analysis toward systematic aeroelastic stability and flutter-boundary assessment.

Steady Reynolds-averaged Navier–Stokes (RANS) runs often terminate once integrated forces vary little over an iteration interval. Stable forces, however, do not show that local interaction outputs are stationary. This study examines how much those outputs vary after a run meets the force gate (the 0.5% force-variation gate on total C_D over the final 1,000 iterations) and how that variation compares with wall-normal grid and turbulence-closure differences. Twenty-four families of slender axisymmetric biconics (18 geometry families and six Reynolds-number-pair runs, in three pairs), varying in forecone angle, aft-cone turn, Mach number, and Reynolds number, were run in SU2, the 18 geometry families with corresponding inviscid (Euler) runs. In a prior study, two of these families (hereafter the lead families) were stopped after 10,000 second-order iterations at Courant–Friedrichs–Lewy (CFL) number 0.1 once the force gate was met. Here, those runs were extended on the same meshes, and every family was continued and scored with a drift-rate stationarity test. Selected families were also run on coarse and fine wall-normal meshes and with the Spalart–Allmaras (SA) turbulence model. In shear-stress transport (SST) runs, total drag varied by less than 1% after the prior study's stop, yet normalized wall-shear margin rose by 0.32 in both lead families. That rise exceeds the SA–SST difference (0.07–0.08) and the largest wall-normal grid difference (0.17). Both fields also moved from class I (an isolated or too-short region of negative wall shear) to class S (separation spanning at least three contiguous wall nodes in successive snapshots). In one lead family, the pressure-rise change exceeds the grid and closure differences. The Euler–RANS pressure-drag difference changed by 0.5–1.2 percentage points after that stop, comparable to the grid differences. For the lead families, the prior study's SA runs, also stopped after 10,000 second-order iterations, were compared against new SA runs started from scratch and run to stationarity. Under SA, the field class remained the same, and the wall-shear margin moved much less than under SST: about 2% and 20% of the SST change in the two families, and in the opposite direction in both. Neither SA field contained a separation region spanning at least three contiguous wall nodes. Solver residuals proved unreliable indicators of settling. In one fine-mesh run, the residual leveled off while the wall-shear margin was still evolving; the margin failed 119 of 154 stationarity checks, made every 1,000 iterations. In one SA run, the outputs were stationary while the residual was about 25 times above its lowest value. Once outputs were stationary, Euler and RANS pressure drag differed by −1.5% to +0.9% across 17 of the 18 geometry families (one failed a sanity check). Each discrepancy is of the same order as the grid and closure differences, but peak pressure differed by 12–15% in the lead families, and an Euler run has no wall shear. A profile-sampling defect left boundary-layer-thickness-based quantities unusable for 18 of the 24 families; those values are not reported. The results support applying stationarity tests to each reported output, not to forces or residuals alone.

Mass flow distribution in single-inlet, multi-branch supersonic inlets directly affects airflow matching in multi-engine propulsion systems. However, the backpressure response and interbranch coupling associated with a shared upstream flow field require further clarification. This study employs three-dimensional steady Reynolds-averaged Navier–Stokes (RANS) simulations to investigate a single-inlet, three-branch, two-dimensional mixed-compression supersonic inlet at its design Mach number of 2.9. The results show that the equal-backpressure condition examined amplifies the initial flow distribution imbalance, while reducing the backpressure in branch C causes branch A to replace branch C as the branch with the lowest mass flow share. Throughflow, equal-backpressure, and asymmetric-backpressure conditions involving backpressure reduction only in branch C are compared. Mach number distributions, secondary flow structures, and static pressure fields are analyzed to examine mass flow redistribution and the corresponding internal flow adjustments. When all three outlet backpressure ratios are 13, the mass flow share of branch C decreases from 36.69% under the throughflow condition to 16.20%, while the difference between the largest and smallest branch shares increases from 7.35 to 27.58 percentage points. With the backpressure ratios of branches A and B maintained at 13, reducing the backpressure ratio of branch C to 11.7, 10.4, and 9.1 increases its flow share to 41.61%–43.46%. Meanwhile, the share of branch A remains below 20% in all three cases, making it the branch with the lowest share. Across the three asymmetric cases, further reductions in the backpressure of branch C increase its flow share from 41.61% to 43.46%. The difference between the largest and smallest branch shares increases from 21.96 to 25.70 percentage points, remaining below the equal-backpressure baseline of 27.58 percentage points but above the throughflow value of 7.35 percentage points. Furthermore, the mass flow ratio between branches A and B changes markedly despite their unchanged outlet backpressures, demonstrating steady aerodynamic coupling among the three branches. This redistribution is accompanied by adjustments in the pressure distribution within the common flow-splitting region, internal wave structures, and three-dimensional cross-sectional flow structures. These findings indicate that backpressure regulation in multi-branch inlet systems should consider both the response of the target branch and the air supply conditions in the other branches, and can inform overall inlet airflow matching in multi-engine propulsion systems.

Integrated Design and Analysis of a Thrust Vector Control System for Solid Rocket Motors This study provides an explanatory description of a series of activities related to the development of components for a solid rocket motor (SRM) and a test stand for investigating the thrust vector control system for this type of propulsion system. The objective of the project is to develop a control technology for rockets operating in the suborbital regime, where conventional systems based on aerodynamic control surfaces are unable to provide effective maneuvering due to the reduced atmospheric density. A three-dimensional numerical investigation of thrust-vector control by secondary hot-gas injection into the divergent section of a conical solid-rocket-motor nozzle is presented. The study was intended to map the design space before prototyping by quantifying the effects of secondary mass-flow ratio, injector location and inclination, and chamber pressure on thrust-vector angle and propulsive losses. Combustion-product properties for an AP/HTPB/Al formulation were obtained with NASA CEA; structured hexahedral meshes were produced in ANSYS ICEM CFD; and steady RANS equations were solved in ANSYS Fluent using the realizable k-ε model with scalable wall functions. A preliminary concept for the geometry and control of the exhaust-gas flow vectoring system downstream of the engine nozzle was developed. The adopted concept is based on gas-actuated electromechanical systems for controlling the exhaust-gas flow downstream of the rocket engine nozzle. The geometry of the mechanical systems was designed, while analytical calculations were performed in parallel to size the rocket engine elements and control-system components to ensure an adequate safety margin. The components were analyzed in terms of mechanical strength, thermal resistance, and erosion resistance under conditions of high gas temperatures and high flow velocities. Based on the results of numerical simulations coupling computational fluid dynamics and structural analyses (ANSYS Fluent, ANSYS Mechanical, Nastran), the predicted system performance, capabilities, dynamics, and efficiency were determined. Based on the numerical analyses, structural and material solutions were selected for components exposed to extreme temperatures and erosive conditions. For the purposes of dynamic testing, RDTT-324 rocket motors recovered from H-25MP missiles were disassembled. The technical and chemical condition of the motor propellant grain was examined at the WITU laboratory. A complete structural design of a test stand was developed for investigating the effects of thrust-vector control, with the capability of measuring force and moments in two planes.

Traditional flight-simulation models often build global aircraft behavior by connecting local models identified at individual flight conditions. Building on FlyNet, a physics-informed neural flight-dynamics architecture that instead learns continuous aircraft force and moment maps while preserving rigid-body equations of motion, this work evaluates whether learned rotorcraft dynamics can serve not only as trajectory-matching simulators, but also as controller-facing differentiable surrogates. Existing learned flight-dynamics models are often evaluated through supervised prediction error, trajectory matching, or rollout accuracy. A model may reproduce trajectories while still producing poor local linear behavior, artificial unstable modes, or misleading feedback-tracking performance. Using an AH-1S helicopter model generated in an open-source flight-dynamics simulation environment, learned force and moment maps are combined with rigid-body equations of motion and evaluated through a model-in-the-loop validation framework. The learned dynamics are rolled out on held-out cases and locally linearized at held-out operating points by computing the Jacobian of the learned state-update model. Local stability is assessed from the eigenvalues of this linearization, where unstable modes indicate directions in which small perturbations grow in the learned surrogate. The same learned models are used to design local linear quadratic regulator feedback controllers, which balance state-tracking error against control effort. These controllers are used as diagnostic probes of model usefulness, not as certified flight-control laws. The primary evaluation is paired tracking error under matched perturbations. Across 27 held-out AH-1S cases, a group distributionally robust optimization training objective, which emphasizes the worst-performing speed regimes during training rather than only average error, achieved a median normalized tracking-error ratio of 0.914 relative to open-loop simulation. The same group-robust model reduced the mean fraction of locally unstable modes from 0.771 for the fresh baseline model to 0.302, a 60.8 percent reduction. This does not imply that the physical aircraft became more stable; it indicates that the learned surrogate produced fewer locally unstable linearized modes around the tested operating points. A separate temporal-history model using 12 past samples reduced held-out rollout geometric-mean root-mean-square error from 0.06498 to 0.02293 and achieved a tracking-error ratio of 0.909 relative to the fresh-baseline controller. Under first-pass nonideal simulation conditions including sensor noise, actuator delay, actuator rate limits, and deterministic disturbances, no divergence was observed in the tested cases. As an external real-flight check, a National Aeronautics and Space Administration Dashlink tail-numbered commercial-flight dataset is also audited. The dataset provides synchronized flight-recorded time histories, including sensor measurements, sampling rates, units, parameter descriptions, and control-related channels, making it useful for testing whether one-step validation behavior and rollout behavior agree on real aircraft data. This audit provides supporting evidence for the need to examine rollout behavior beyond supervised prediction loss, but the full force and moment learning, local-linearization, and controller-facing validation framework is implemented on the AH-1S case. These results do not constitute certified flight-control validation or hardware flight testing. Instead, they demonstrate a practical evaluation path for learned flight-dynamics models: after trajectory fitting, inspect local stability, reduce learned local instability through regime-aware training, and test paired feedback-tracking behavior before claiming controller-facing usefulness.

Stop-rotor vertical take-off and landing aircraft offer a promising solution to achieving both efficient vertical lift and high-speed forward flight by enabling rotor blades to transition into fixed wings. However, this configuration requires a lifting surface to operate in both a rotor-wing and fixed-wing state, which imposes conflicting requirements on airfoil orientation, spanwise loading, and spanwise incidence. Previous stop-rotor morphing wing studies addressed airfoil orientation through reversal, but addressing spanwise loading and incidence remains an open topic of research. This paper presents a two-wing morphing architecture consisting of a reverse-twist wing that reverses its leading and trailing edge orientation while twisting, and a twist-only wing that twists without airfoil reversal. The mechanism extends a constrained Watt's six-bar reversal linkage with a gear-train actuation system that produces different prescribed rotations at the root and tip ribs. A twist-transition test measured approximately $-7.5^\circ$ of washout in the rotor-wing configuration against a $-7^\circ$ design target. Rigid aerodynamic shells representing the candidate wings were tested in a low-speed wind tunnel and on a rotor thrust stand. The fixed-wing tests show that the untwisted fixed-wing morph shell increased lift capacity by 7.3\% relative to the twisted morph shell over their shared lift region and by 54.5\% relative to a NACA~0014 shell at the cost of minor reductions in aerodynamic efficiency, 1.8\% and 6.9\% respectively. The rotor-wing tests show that the twisted rotor-wing morph shell increased hover figure of merit by an average of 3.9\% relative to the untwisted morph shell over their shared thrust region and by 5\% relative to a NACA~0014 shell. These results establish the combined reversal-and-twist mechanism for stop-rotor wings and suggest that spanwise twist morphing, in addition to airfoil reversal, is a mechanically feasible and valuable degree-of-freedom that could enhance the future of efficient VTOL aircraft.

Missile aerodynamic characteristics depend strongly on flight conditions and interactions between the body, stabilizers, and control surfaces. Reliable estimation of aerodynamic coefficients over a representative flight envelope is therefore essential for flight-dynamics analysis. The present study is focused on a three-stage 300 mm rocket covering subsonic, transonic, and supersonic flight regimes. The objective of this paper is to present a complete approach to the aerodynamic analysis of a missile configuration. The study includes the identification of representative flight and control conditions, the execution of CFD calculations for the complete missile geometry, and the analysis of the final aerodynamic results in terms of parameter sensitivity. The investigated flight and control ranges are defined based on requirements provided by the avionics and flight-control team. These requirements are derived from available flight data for related rocket configurations and from numerical flight simulations and are used in the present study as the basis for defining the aerodynamic calculation domain. The considered variables include Mach number, angle of attack, sideslip angle, and generalized control-mode variables corresponding to roll, pitch, and yaw control-surface deflections. The specified parameter ranges are used to define representative aerodynamic cases covering the relevant operational regions of the missile flight envelope. The CFD procedure includes geometry preparation, computational-domain definition, mesh generation, selection of boundary conditions, solver setup, and extraction of aerodynamic coefficients. All calculations are performed for the complete missile geometry in order to account for aerodynamic interactions between individual components of the configuration. In order to perform the sensitivity study, a set of representative aerodynamic calculation cases is defined and CFD calculations are conducted for the complete missile geometry. The obtained aerodynamic coefficients are then analyzed to assess the influence of the main flight and control parameters and the effectiveness of roll, pitch, and yaw control commands. The analysis includes the baseline aerodynamic characteristics, Mach-number dependence, control-effectiveness derivatives, and coupling between the generalized control channels. A sensitivity analysis based on standardized regression coefficients (SRC) was performed to quantify the relative influence of the flight-state and control parameters on the aerodynamic force and moment coefficients. The presented approach provides a practical workflow for aerodynamic analysis of a missile configuration, starting from the identification of representative flight and control conditions, through complete-geometry CFD calculations, to the interpretation of the final aerodynamic coefficients. The methodology makes it possible to determine which flight parameters and control commands have the strongest influence on the aerodynamic response. As a result, the study helps to identify the key factors that should be considered in further aerodynamic database development, flight-dynamics modelling, and control-effectiveness assessment.

The aerodynamic performance of non-slender delta wings at high angles of attack is strongly governed by the formation and stability of leading-edge vortices (LEVs). While passive bleed techniques have been widely investigated for flow control, but existing studies have primarily focused on parameters like opening size and porosity, with limited attention to geometric orientation. This study presents a combined experimental and numerical investigation into the effect of bleed opening orientation on vortex dynamics and aerodynamic performance. Water tunnel experiments were conducted at a Reynolds number of 12,565 on a 60° swept delta wing with a fixed bleed opening size of 0.9 mm. Two configurations were examined: a forward-oriented opening (+20°) positioned closer to the leading edge, and a rearward-oriented opening (−20°). Aerodynamic force measurements were complemented by unsteady Reynolds-averaged Navier–Stokes simulations using the SST k–ω turbulence model. The results demonstrate that the opening orientation plays a very decisive role in flow control effectiveness. The +20° configuration significantly enhances lift performance in the medium-to-high angle-of-attack range (15°–45°) and achieve a higher maximum lift coefficient as compared to the −20° configuration. Flow field analysis reveals that the forward-oriented opening promotes earlier interaction with the separating shear layer and stabilize the leading-edge vortex closer to the wing surface and increasing vortex-induced suction. In contrast, the rearward-oriented opening leads to increased vortex detachment and reduced control authority. These findings establish bleed opening orientation as a critical design parameter for passive flow control systems and provide new insights into the geometric optimization of vortex-dominated aerodynamic configurations.

Turbine-based combined-cycle propulsion is a promising architecture for hypersonic vehicles operating over a broad Mach-number range. During the transition from turbine-based propulsion to ramjet propulsion, rapid changes in captured airflow, fuel flow, net thrust, and propulsion-induced pitching moment may occur. These transients interact directly with the longitudinal dynamics of the airframe and may cause trajectory deviations, attitude disturbances, and temporary degradation of flight stability. This paper develops a longitudinal propulsion–airframe coupled modeling and simulation framework for a hypersonic vehicle undergoing turbine–ramjet mode transition. The framework integrates a nonlinear longitudinal flight-dynamics model, an aerodynamic model incorporating propulsion–airframe interaction effects, and a component-level dynamic model of the turbine–ramjet combined-cycle propulsion system. The vehicle employs an externally parallel inlet configuration, in which the turbine and ramjet flowpaths are supplied through separate external-compression inlet passages while being influenced by the same forebody compression field. The turbine flowpath is represented using nonlinear performance characteristics of the compressor, combustor, turbine, and nozzle, whereas the ramjet flowpath is modeled using quasi-one-dimensional compressible-flow relations and combustion-efficiency characteristics. Separate inlet models determine the captured mass flow, total-pressure recovery, and inlet-exit conditions of the two flowpaths. The effects of splitter-plate deflection and effective capture-area variation are included. During mode transition, these variable-geometry parameters are coordinated with turbine throttling, ramjet fuel scheduling, airflow redistribution, and nozzle-area modulation. A bidirectional interface transfers Mach number, altitude, angle of attack, forebody flow conditions, and geometry commands to the propulsion model. The calculated mass-flow rates, thrust contributions, fuel-flow rates, inlet-induced aerodynamic effects, and propulsion-induced pitching moment are returned to the longitudinal equations of motion. Numerical simulations will be conducted under a representative nominal constant-altitude acceleration condition covering the turbine–ramjet transition interval. The coupled model will be used to investigate the effects of asynchronous inlet-geometry adjustment, airflow redistribution, and fuel scheduling on total thrust, pitching moment, and longitudinal flight-state responses. Particular attention will be given to transient variations in velocity, altitude, angle of attack, pitch rate, and flight-path angle. Comparative simulations against a decoupled model using prescribed or quasi-steady thrust profiles will be performed to quantify the importance of bidirectional propulsion–airframe interaction. Parametric studies will further examine the influence of transition duration, fuel-scheduling overlap, inlet-geometry scheduling, thrust-line offset, and center-of-gravity location. The proposed framework is intended to support mode-transition performance assessment and the integrated design of inlet geometry, propulsion operation, and longitudinal flight-control systems.

Modern advanced aircraft increasingly rely on converter-dense electrical power systems to support sensing, computation, actuation, communications, survivability, thermal-management electrical loads, and flight-critical control. This growth in electrical demand increases coupling among generation, conversion, distribution, storage, and regulated mission loads, creating a need for models that preserve dominant converter and bus dynamics without the computational cost of full switch-level simulation. This work develops a compact, reproducible, mission-scheduled reduced-order model for a representative more-electric aircraft power network using generalized state-space averaging. The framework includes dual generator/source-converter channels, split 270 VDC buses, dual power distribution converters supplying a shared 28 VDC bus, a four-leg 115 V/400 Hz AC inverter branch, and high- and low-voltage backup storage. Mission demand is imposed through scheduled operating phases spanning ground, takeoff, climb, cruise, descent, and landing, enabling evaluation of bus regulation, source sharing, storage participation, converter coupling, and retained harmonic behavior under changing load conditions. A limited ten-run Latin Hypercube ensemble is used to demonstrate repeated mission-level evaluation under grouped parameter variation, with screening-level correlations used to identify candidate factor-response associations. The proposed model provides a tractable foundation for mission-level analysis of converter-dense aircraft electrical power networks and establishes a baseline for future integration of prime-mover limits, thermal dynamics, EMI indicators, degradation, expanded uncertainty ensembles, and prognostics-and-health-management observables.

Electric vertical take-off and landing (eVTOL) aircraft will fly low-altitude approach and departure profiles into vertiports where wind conditions are shaped by discrete gusts, building wake, and short-duration turbulence rather than the steady open-atmosphere flow assumed by many legacy rotorcraft handling-qualities studies. Regulatory and industry guidance for urban air mobility increasingly calls for gust and turbulence loading to be assessed explicitly during vertiport terminal-area operations, yet many early-stage eVTOL simulation studies still evaluate vehicle response only under nominal, gust-free conditions. There is a need for an accessible, reproducible simulation methodology that quantifies how standard, well-established gust models affect eVTOL flight response and control margin during this flight phase, without requiring the computational cost of high-fidelity CFD. This paper presents a mid-fidelity simulation study of a representative multirotor eVTOL configuration flying a vertiport approach-to-hover and departure profile under two standard, widely used gust representations: a discrete 1-cosine gust and a continuous Dryden turbulence field, applied at several intensities and relative headings. Rotor aerodynamic loads are computed with a blade-element momentum theory (BEMT) model including a dynamic-inflow correction, which is coupled to a six-degree-of-freedom (6-DOF) rigid-body flight dynamics simulation with a baseline attitude/altitude-hold controller representative of published eVTOL control architectures. Vehicle geometry and mass/rotor properties are drawn from a publicly documented reference eVTOL configuration so that the study is fully reproducible from open data. The simulation campaign compares vehicle attitude excursions, rotor thrust and power margins, and controller command saturation across the gust cases, benchmarked against the nominal (gust-free) approach profile. Results characterize which combinations of gust type, intensity, and heading most erode control margin during the approach-to-hover transition — the flight segment considered most gust-sensitive for multirotor eVTOL configurations — and quantify the difference between discrete and continuous gust representations for this vehicle class and flight phase. The expected contributions are: (1) a fully reproducible, open-data mid-fidelity simulation methodology for evaluating eVTOL gust response during vertiport terminal operations, implementable with standard BEMT and 6-DOF flight-dynamics tools; (2) a quantitative comparison of discrete-gust and Dryden-turbulence vehicle response for a representative multirotor eVTOL, identifying which gust representation and flight condition produce the tightest control margins; and (3) a case-study baseline that lower-cost simulation teams and academic groups can extend or benchmark against, without requiring CFD infrastructure. The work supports near-term modeling, simulation, and verification needs for urban air mobility vertiport operations and complements higher-fidelity aerodynamic studies in this session.

Titan’s extremely cold, dense, nitrogen-dominated atmosphere contains methane as a potential in situ fuel resource. To enable high-speed, long-range exploration within this atmosphere, this study proposes a ramjet that ingests ambient gas, uses onboard liquid oxygen as the oxidizer, and injects supplementary methane. The temperature required for combustion in nitrogen-diluted flow motivates an initial design Mach number of 7, while the cruise dynamic-pressure constraint leads to a design altitude of 70 km. A characteristic-station-based quasi-one-dimensional model was used to generate the training dataset for an artificial neural network surrogate. The surrogate screened propellant ratios, inlet-exit Mach numbers, and cross-sectional area ratios. The selected configuration, comprising an isolator, a diffuser, and a subsonic main combustion region, was then evaluated using the quasi-one-dimensional model. At the design point, the flow decelerates from approximately Mach 1.98 at the inlet exit to Mach 0.40 at the entrance to the main combustion region, where the static temperature and pressure are approximately 1043 K and 1.61 MPa, respectively. After combustion, the static temperature reaches approximately 1548 K, and the flow accelerates to Mach 1.89 at a downstream supersonic characteristic station. At the design point (freestream Mach 7, altitude 70 km, inlet total-pressure recovery of 0.24, methane augmentation ratio of 1.72, and overall oxidizer-to-fuel ratio of 3.59), the predicted thrust is 5.76 kN and the specific impulse based on onboard propellants is 493 s. This specific impulse exceeds a 350 s reference for a liquid-oxygen/methane rocket by more than 40%. At a fixed overall oxidizer-to-fuel ratio, increasing the methane augmentation ratio from 0 to 3.2 raises thrust from 1.76 to 8.90 kN, while specific impulse remains approximately 480–498 s. At fixed methane augmentation, increasing the overall oxidizer-to-fuel ratio from 2.5 to 4.0 raises specific impulse from 399 to 515 s; a further increase to 6.5 lowers it to 365 s. Thus, increasing onboard oxygen supply can increase thrust, but excessive oxygen addition reduces onboard-propellant specific impulse. An altitude–Mach sweep over freestream Mach numbers of 5–9 was conducted with the design-point geometry and propellant-control parameters held fixed. Lower-Mach conditions can yield higher thrust and specific impulse, but the flow entering the combustor may be colder. For example, at 56 km and Mach 5, the predicted thrust and specific impulse are approximately 12.4 kN and 603 s, respectively, while the inlet-exit static temperature is only about 299 K. Within the 20–100 kPa dynamic-pressure corridor, thrust ranges from 1.33 to 12.4 kN and specific impulse from 386 to 603 s. Increasing freestream Mach number raises the combustor-inlet temperature through ram heating but also increases inlet shock losses. Balancing propulsion performance against combustor-inlet temperature suggests 60–90 km and Mach 6–8 as a region warranting further investigation. The proposed configuration shows preliminary feasibility at the thermodynamic-cycle level and a specific-impulse advantage. Future work will match design parameters to individual flight Mach numbers over a wider speed range and assess inlet total-pressure retention, propellant injection and mixing, ignition, and flame-holding under nitrogen-rich conditions.

Micro-turbojet propulsion systems require accurate control-oriented models of the fuel-delivery subsystem because fuel-flow dynamics directly affect engine speed, thermal response, thrust generation, and closed-loop controller performance. This work presents an experimental system-identification methodology for the fuel pump and fuel-delivery path of an SW190B micro-turbojet propulsion system. The study combines gravimetrically referenced fuel-flow measurement, step-response characterization, pseudo-random binary sequence (PRBS) excitation, and randomized multilevel excitation to identify nominal linear dynamics and operating-point-dependent behavior. The brushless fuel pump was actuated through an electronic speed controller using pulse-width modulation, while fuel flow was obtained from the raw pulse output of a T3000 turbine flow sensor. The sensor was calibrated against timed gravimetric measurements, and an eight-pulse rolling frequency estimator was developed to reduce pulse-measurement ripple while preserving the transient response required for system identification. Preliminary step tests indicated dominant fuel-path time constants of approximately 0.2–0.45 s and were used to select the excitation time scales for subsequent experiments. A binary PRBS experiment with a 0.75 s bit interval and an effective sampling rate of approximately 19.6 Hz was first used for linear identification. A first-order-plus-dead-time model, \(G(s)=0.03118e^{-0.1528s}/(0.2218s+1)\), produced a validation RMSE of 0.238 g/s and an 82.6% fit. An ARX(1,4,1) model improved the RMSE to 0.217 g/s with an 84.1% fit. To assess wide-range behavior, a seven-level randomized excitation spanning 1150–1300 µs was subsequently applied. Linear ARX, quadratic and cubic Hammerstein, and first-order linear parameter-varying (LPV) structures were compared using held-out experimental data. The LPV model provided the best wide-range performance, achieving an RMSE of 0.291 g/s and an 80.9% fit, representing approximately a 20% reduction in RMSE relative to the linear ARX model on the multilevel dataset. The results show that a compact delayed first-order model is suitable for nominal control-oriented analysis, while an LPV representation better captures operating-point-dependent fuel-path dynamics. The identified models provide a foundation for subsequent micro-turbojet engine identification and RPM/thrust control development.

A swarm of cooperating intelligence, surveillance, and reconnaissance (ISR) platforms defending against an incursion of adversary small uncrewed aircraft systems (sUAS) must act collectively on what its members can individually sense and share: merging distributed observations into shared target custody, coordinating which platform attends to which threat, inferring adversary intent from observed motion, and rejecting infiltrators that imitate friendly flight to penetrate the formation—all under degraded sensing and navigation. We present and systematically evaluate a distributed-autonomy architecture that addresses these jointly. The swarm builds a single shared, identity-consistent picture of the airspace from its members' individual observations and acts on it collectively. Per-agent generalized labeled multi-Bernoulli (GLMB) tracking, generalized covariance intersection (GCI) fusion at a ship-hosted master station, and persistent label matching supply that picture—labeled and temporally stable, as coordination demands. On it, the agents self-organize: a stigmergic shadow-selection policy disperses surveillance across threats without central tasking, online inference classifies each adversary's behavior mode, and a friendly filter rejects infiltrators by ultra-wideband (UWB) ranging consistency rather than kinematics alone. Perception is centralized at the ship while action selection is distributed; fully decentralized fusion under contested communications is left to future work. We evaluate the swarm across a two-dimensional adversary taxonomy spanning tactical intelligence and kinematic energy, realized as a 30-cell parameter grid. The principal result is a counter-infiltration capability both complete and invariant to adversary energy: zero of 5,628 active-infiltration evaluations were classified friendly, with honest-versus-infiltrator separability of AUC 0.993–0.999 across all sustained-infiltration cells—kinematic mimicry cannot defeat a physical-channel consistency check, however aggressively the infiltrator maneuvers. We further find that collective tracking difficulty is non-monotonic in threat intensity—the most demanding regime is mid-intelligence standoff maneuvering, not maximum aggression—and we characterize a long-range track-cardinality pathology in the per-agent filter. Finally, a Latin-hypercube Monte Carlo campaign over eleven sensing, navigation, and engagement-geometry parameters quantifies the robustness of these results and the absolute-navigation quality the swarm needs to sustain them. The campaign exposed a failure mode peculiar to cooperative swarms and invisible to nominal-condition testing: when agents fold one another's shared position estimates back into their own navigation filters, they double-count correlated information—"data incest"—and grow falsely confident. Because the formation's horizontal centroid is unobservable from relative inter-agent measurements alone (a gauge freedom resolved only by periodic absolute-position fixes), this manufactured confidence collapses the gain on those fixes and re-opens the gauge freedom they were otherwise closing; the swarm then drifts coherently out of its operating area while its fixes arrive on schedule. The finding motivates confining cooperative ranging to identity verification and excluding it from state estimation.

This paper addresses the problem of autonomous emergency handling of critical events that require immediate landing, by non-human-piloted aircraft such as small-Unmanned Aircraft Systems or Advanced Air Mobility platforms, when these are operating away from civilian infrastructure and are surrounded by unstructured natural environments. In such cases, the human paradigm of scene understanding to identify nearby emergency landing zones has been field-proven historically. To implement this capability for autonomous aircraft, we propose a Learning Enabled Component that relies on visual appearance-based classification and pixel-level segmentation of unstructured natural environment classes of interest, and extracts their boundary and an interior point for emergency landing. A framework that progressively constructs a database of such contingency locations is used, allowing to autonomously select the nearest one and re-route the aircraft upon a critical fault. A field demonstration is shown, corresponding to an autonomous opportunistic emergency landing on water conducted in real-time, which took place during the NASA Autonomy Fest public event carried out in the presence of established academic and major industry stakeholders.

Autonomous air combat presents distinct challenges for multi-agent systems: agents must pursue dynamic adversaries across a continuous six degree-of-freedom flight envelope, manage limited resources under time pressure, and coordinate under partial observability. Existing approaches either rely on simplified flight dynamics that limit realism, or treat coordination as a byproduct of shared rewards without explicit mechanisms for inter-agent communication or opponent modeling. This paper addresses these gaps in 2v2 air combat, developing a multi-agent reinforcement learning (MARL) framework built on a custom F-16 flight dynamics model with JSBSim aerodynamic coefficients, a wing-level trim solver, and weapon engagement zone (WEZ) modeling. Multi-Agent Proximal Policy Optimization (MAPPO) serves as the baseline under the centralized training with decentralized execution (CTDE) paradigm, achieving a 46.4% win rate against a rule-based heuristic opponent over 500 deterministic evaluation episodes. A Graph Attention Network (GAT) communication module enabling distance- and threat-weighted inter-agent message passing raises this to 69.8% (+23.4 percentage points, non-overlapping 95% confidence intervals). An episode-level outcome decomposition shows this gain comes almost entirely from converting episodes in which no engagement was ever established, rather than from faster completion of engagements already begun. To address credit assignment, Factored Multi-Agent Centralised Policy Gradients (FACMAC) with Twin Delayed Deep Deterministic Policy Gradients (TD3) stabilization is evaluated, decomposing the joint value function while supporting continuous action spaces. FACMAC reaches 82.0% using a small fraction of the environment interactions required by MAPPO+GAT. A centralized opponent modeling module then classifies enemy intent from observation history using aspect-angle supervision, resolving a label-collapse failure mode seen in threshold-based alternatives, and a REINFORCE-trained module allocates non-overlapping tactical roles conditioned on inferred intent, reaching 84.2%. The central finding concerns why this role mechanism succeeds in one architecture but fails in the other. The identical module collapses to a single dominant role under MAPPO, yet produces complementary specialization under FACMAC. Controlled ablation isolates three causes. Resampling the role at every timestep prevents the learning signal from attributing an episode outcome to any particular assignment, so the assigner learns nothing until assignment is stabilized. Once assignment is stable the estimator drives the distribution to a single role pair regardless of the reward structure, and which pair wins varies with the random seed. Most consequentially, removing the learned role input erases the division of labor in FACMAC while leaving MAPPO untouched, indicating that per-agent credit from the monotonic mixing network is what makes role specialization learnable, and that a shared critic cannot supply it. Notably, this failure is invisible in aggregate win rate. Out-of-distribution generalization is assessed by pitting the two architectures against each other, having never met in training. FACMAC+OM+REINFORCE achieves a 68.0% win rate versus 17.2%, substantially wider than the margin observed against the fixed heuristic, with consistent results across both team color assignments. Five recurring failure modes are documented with the interventions that resolved each: policy collapse under sparse rewards, heading misalignment at long range, fire suppression via reward hacking, Q-value divergence in continuous actor-critic training, and replay buffer distribution shift under non-stationary self-play.

In disaster response environments, communication infrastructure is often degraded or completely unavailable, while human operators experience high cognitive load due to rapidly changing and uncertain conditions. These challenges limit the effectiveness of conventional remotely piloted Unmanned Aerial Vehicle (UAV) systems, especially when fast and reliable decision-making is required. Most existing autonomous UAV systems use fixed rules or separate models for perception and navigation. Some systems use machine learning for tasks, such as object detection or path planning, but they usually operate independently and do not support overall mission-level decision-making. As a result, these systems cannot effectively adapt to changing mission needs, sensor inputs, and resource limitations in a coordinated way. This paper presents a novel Large Language Model (LLM)-assisted autonomous UAV framework for disaster triage operations, in which mission-level decision-making is integrated into the system’s high-level control and planning process. The system performs perception, reasoning, and navigation with minimal human intervention under resource-constrained conditions. In this framework, the LLM performs sector prioritization, route selection, and waypoint ordering and makes high-level mission decisions, such as whether to continue flight, hold position, or return to base, based on battery status, victim detection confidence, wind conditions, and localization quality. The proposed system builds a simulation-based autonomous UAV framework implemented using PX4, ROS 2, and Gazebo, to form an integrated software architecture for flight control, communication, and simulation. To enhance system robustness and operational reliability, three capabilities are integrated into the framework. First, a victim detection module uses an onboard vision pipeline based on the YOLOv8 deep learning model to detect and localize humans in real time within designated search sectors. Second, a battery-aware mission planning module estimates remaining energy before waypoint selection and prioritizes sector coverage to prevent mission failure from battery depletion or forced return-to-base behavior. Third, a wind-aware decision layer incorporates environmental disturbances simulated in Gazebo to support adaptive route planning under varying wind conditions. To ensure operational safety, the system maintains a hierarchical architecture with deterministic override rules that bypass the Large Language Model (LLM) decision layer when safety thresholds are violated, such as low battery levels or degraded Extended Kalman Filter (EKF) state estimation quality. Additionally, a confidence-based LLM gating mechanism evaluates decision reliability using system state indicators and prevents low-confidence outputs from being executed by the flight controller. Simulation results show that the system reliably executes mission tasks under varying conditions. The autonomous UAV performs perception and navigation despite battery constraints and sensor uncertainty, while the LLM makes mission-level decisions for sector selection, route planning, and mission continuation. The system achieves an average decision latency of 4.16 seconds and maintains a decision confidence of 83.3%, demonstrating the effectiveness of the proposed decision-making framework. The results show that large language models can function effectively as a structured reasoning layer within autonomous UAV systems rather than replacing conventional flight control logic. The proposed framework may help enhance disaster response and surveillance operations by enabling faster victim detection, adaptive mission planning, and improved situational awareness in complex environments.

Reliable environmental sensing with unmanned aerial vehicles (UAVs) requires understanding how the aircraft’s relationship with its environment evolves throughout an encounter. This paper formulates UAV–volcanic-plume interaction recognition as a temporally structured perception problem with four states: approaching, entering, inside, and exiting. Using 12 fixed-wing flights collected at Volcan de Fuego, Guatemala, we construct 75 temporal interaction annotations and synchronise forward-facing video with aircraft telemetry over a common five-second observation horizon. For vision, frame-wise ViT-B/16 is compared with video-native MViTv2-S, while aircraft dynamics are modelled using a TimeMIL-inspired temporal Transformer. MViTv2-S achieves a validation Macro-F1 of 0.7500, with ambiguity occurring between inside and exiting. Aircraft telemetry achieves 0.4034 Macro-F1 on the same four-state task, but 0.7860 when distinguishing low and high-disturbance dynamics, indicating complementary information across modalities. Residual gated cross-attention is therefore used to condition the visual representation on aircraft dynamics, increasing validation Macro-F1 from 0.7500 to 0.7629 and accuracy from 0.7480 to 0.7570. Whole-flight evaluation further reveals prediction errors and implausible state transitions that are not apparent from aggregate window-level metrics. These results establish a multimodal framework for temporally aware UAV–environment interaction perception and motivate sequence-aware modelling for autonomous environmental sampling.

Advanced Air Mobility (AAM) is an emerging concept of low-altitude aviation operations conducted in dense urban environments and surrounding cities that require scalable traffic management methods. This work develops a novel multi-agent reinforcement learning (MARL) framework for autonomous, decentralized conflict resolution in a recently proposed network of structured air corridors. This air corridor network is an analog of a ground road network, in which vehicles travel along predefined two-directional lanes, and the principal coordination challenge arises at the intersections where corridors cross. At air corridor intersections, aircraft proceeding straight, turning left, or turning right may come into conflict with crossing, merging, or opposing traffic. Because each vehicle must stay on its assigned corridor, it cannot avoid conflicts by freely steering sideways; instead, it adjusts its speed or changes altitude layer. Thus, each vehicle resolves such conflicts by adjusting its speed when it enters a shared intersection, or by changing altitude layer to vertically separate from nearby traffic, or by combining both actions. The problem is formulated as a multi-agent Markov decision process and addressed using Proximal Policy Optimization (PPO) under a centralized-training, decentralized-execution (CTDE) paradigm. Two variants are implemented: Independent PPO (IPPO), with a parameter-shared actor and a decentralized local critic, and Multi-Agent PPO (MAPPO), whose centralized critic applies cross-agent attention over all active aircraft to evaluate joint behavior during training. At execution, each vehicle acts solely from local observations, including its own state and route progress together with its nearest neighboring aircraft, encoded by a multi-head attention network that incorporates relative geometry, encounter topology (same-direction, merging, crossing, or opposing), and the time and distance to the closest point of approach. The action is hybrid: a continuous speed command combined with a discrete altitude layer change (climb, stay, or descend). Since horizontal routing is fixed by the assigned corridor path, coordination is achieved purely through speed and altitude control. A shaped reward encourages route progress, conflict-free completion, and safe separation while penalizing losses of separation, unnecessary control effort, and layer changes. The framework is built on BlueSky, an open-source air traffic simulation platform, in which the structured corridor network, its intersections, and discrete altitude layers are represented and used for both training and evaluation. The learned policies are evaluated in a layered grid-corridor simulation environment against a random baseline across increasing traffic densities, assessing whether decentralized coordination remains safe and efficient as the number of active vehicles, and thus the rate of intersection encounters, grows. The framework provides a foundation for scalable traffic management in structured AAM corridor networks and can be extended to include vehicle priorities, weather disturbances, communication limitations, and vehicle heterogeneity.

Micro Aerial Vehicles (MAVs) are increasingly considered for time-sensitive work in underground spaces such as mine rescue, tunnel inspection, and subterranean mapping, where satellite positioning is unavailable, lighting is poor, and obstacles can appear with little warning. A vehicle operating in these conditions has to move quickly, respond to a changing environment, and avoid obstacles it has never encountered before. Classical path planners such as A*, the Rapidly exploring Random Tree (RRT), and its asymptotically optimal variant RRT* remain standard tools for aerial navigation, yet all three depend on a discrete cycle of sensing, planning, and execution. Once obstacles move faster than a fresh trajectory can be computed, this cycle becomes the limiting factor and mission success decreases. To identify the point at which this occurs, we implemented the three planners and evaluated them across four tunnel configurations of increasing difficulty and a range of target speeds in a 100 m dynamic tunnel, finding that none maintained performance beyond approximately 2 m/s when the obstacle field was both dense and moving. Against these baselines, we evaluated a learned controller trained in the same tunnel using the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm, selected because its continuous action output matches the smooth velocity commands required by the task, while its twin critics and delayed policy updates help reduce value overestimation that can lead to unstable motion near fast-moving obstacles. The policy receives range data, a body-frame inertial state estimate, and a goal-direction cue, and directly outputs velocity commands without an occupancy grid, map, or replanning loop. Trained in the dense dynamic tunnel, the converged policy achieved a 96.0% success rate and a 95.8% obstacle avoidance rate at a mean speed of 2.6 m/s across the full 5000-episode training run. In head-to-head evaluation, the learned policy performed better across the tested speeds and configurations, while in the hardest condition the classical planners achieved success rates of 29% for A*, 25% for RRT*, and 19% for RRT at a 3 m/s command. This advantage comes from the learned policy acting at every control step rather than waiting for a new plan to be generated. However, the classical planners retained an advantage in calm environments, where A* produced smooth and efficient paths through sparse, stationary clutter. The contribution of this paper is therefore not to replace classical planning, but to identify the crossover speed at which environmental motion makes a learned policy more suitable and to show that this crossover occurs within conditions relevant to real underground missions.

As militaries incorporate Unmanned Aerial Vehicles (UAVs) into the modern battlespace, their communication links must evolve to withstand adversarial countermeasures, such as jamming. While commercial-off-the-shelf (COTS) drones have gained wide utility due to their expendability, their communication systems lack anti-jamming capabilities. Modifying the hardware of COTS drones is costly, so algorithmic anti-jam solutions are preferable. In this paper, we propose Roving Autonomous Coverage for Electronic Warfare Resiliency (RACER), an algorithm that enables swarms of autonomous relay drones to provide continuous coverage to a ``mission drone" performing some task. This is accomplished using a 3-phase repeating process; 1) relays measure the electromagnetic environment to characterize the jammer, 2) relays identify optimal positions to support sensor data exfiltration, and 3) relays reposition while using connectivity maintenance methods to ensure connections are maintained throughout. We extend this method by applying it to a path dedicated to looking ahead of the current mission drone position, anticipating large changes in the optimal positions and preempting them. We will demonstrate in simulation the effectiveness of this method and show that without it, there exist scenarios where data flows could not otherwise be maintained.

Mars Regional Exploration via Dynamic Wind-Impelled Networked Gliders (REDWING) is a mission concept for persistent, regional-scale investigation of the Martian lower atmosphere using a coordinated swarm of unpowered autonomous aircraft. The concept addresses the observational gap between orbiters, which provide broad coverage but limited near-surface sampling, and landers or rovers, which provide detailed measurements over small areas. REDWING targets the southern mid-latitudes, where seasonal polar-vortex winds and terrain-induced lee waves may provide atmospheric energy for flight at approximately 500–1000 m altitude while the vehicles collect meteorological, magnetic-field, atmospheric-composition, and visible-spectrum imaging data. The mission concept targets operations lasting up to 30 h. The proposed architecture comprises six flexible-wing gliders organized as two redundant groups of three payload-specialized vehicles. Each group includes a meteorology and magnetometry glider, a tunable-laser-spectrometer glider, and a relay-communications glider; all vehicles share navigation imaging, inertial sensing, onboard computing, inter-glider communications, and moving-mass flight control. Environmental feasibility was assessed using Mars Climate Database wind and solar data, atmospheric models, Scorer-parameter analysis for lee-wave formation, and a candidate flight corridor between approximately 50°S and 70°S. Vehicle performance was evaluated through low-Reynolds-number airfoil and wing analyses in XFOIL/XFLR5, subsystem mass and power budgets, thermal-radiation models, communication link budgets, controller simulations, a preliminary moving-mass hardware experiment, and an integrated modular six-degree-of-freedom swarm simulation. The resulting design uses a 6 m-span, 9.45 m² Dacron flexible wing with a custom blended reflexed airfoil. For the heaviest, approximately 31 kg configuration, the aerodynamic model predicts a maximum lift coefficient of 0.55 and a sink speed of 3.3 m/s, below the 4.0 m/s system requirement and the 8.2 m/s lowest modeled updraft in favorable route segments. A preliminary pitch controller produced 0.36% overshoot and a 3.54 s settling time. Link analyses at 256 kbps yielded estimated margins of 30.1 dB for a 250 km glider-to-orbiter relay and 8.18 dB for a 15 km inter-glider crosslink. Integrated simulations demonstrated coordinated separation, formation maintenance, updraft-seeking behavior, and altitude recovery in representative Martian conditions. These results support REDWING as a technically plausible architecture for extending aerial operations beyond powered-rotorcraft endurance and enabling simultaneous, spatially distributed measurements within the Martian boundary layer.

Air traffic is increasing, and low-altitude airspace is becoming increasingly congested. Voice communication remains the primary means of coordination between pilots and controllers, potentially making automatic speech recognition (ASR) a powerful tool for reducing controller workload, detecting readback errors, and enabling human-machine teaming. Unfortunately, general-purpose ASR foundation models (FMs) perform poorly in aviation environments due to noisy radio channels, lexical shift due to specialized language—including callsigns, headings, and altitudes—and substantial variation in accents across different airspaces. Adapting these models typically requires human-transcribed audio, which is scarce and expensive to produce and must often be recollected for each new operating environment. We present a method for adapting general-purpose ASR FMs to new airports and airspaces without human transcripts. Our two-pronged approach first ensembles decorrelated ASR models to pseudo-label raw target-domain audio. We then augment these pseudo-labels with synthetic speech generated by style-transferring published phraseology to match the target acoustic environment. The resulting dataset is used for parameter-efficient fine-tuning. No stage of the adaptation pipeline requires human-labeled transcription of the audio data. We evaluate our approach on real-world aviation radio audio spanning multiple airports and regions. The method reduces word error relative to foundation ASR models while substantially improving recognition of operationally important content such as callsigns and commands. Despite lacking human-transcribed training data, it outperforms publicly available ATC-specific models and achieves accuracy comparable to systems trained on human-labeled audio. Since the same adaptation procedure transfers across airspaces, deploying ASR in a new operating environment requires only unlabeled radio recordings and computation. As airspace becomes more congested and increasingly automated, this capability provides a scalable voice-understanding layer for emerging technologies that support decision-making for both controllers and pilots, contributing to overall safety in flying operations.

The Ingenuity helicopter opened a new and exciting chapter of aerial exploration of Mars. Ingenuity demonstrated how a rotorcraft can quickly reach features of interest by flying over, rather than around, intervening terrain, and provide a unique aerial survey of those features. As a pioneering technology demonstration, Ingenuity’s scouting capabilities were relatively modest; Ingenuity carried a single sub- 5-gram smartphone-grade color camera and operated within an approximately 1 km radius of its radio relay on the Perseverance rover. Since Ingenuity’s early scouting flights, new Mars rotorcraft concepts have been developed that offer increased exploration and science capabilities. Examples include the Chopper hexacopter and the SkyFall helicopters, which can carry multi-instrument payloads for increased data return and direct-to-orbit radios for unconstrained communication range. These new capabilities, however, require adding mass and therefore generating increased thrust. One of the key constraints limiting achievable thrust for a given rotor design is the maximum allowable advancing blade tip Mach number during forward flight. The commanded rotor speed must leave headroom such that ground speed and winds will not take the advancing tip past this limit. Beyond the limit, the rotor may encounter drag divergence, where rotor efficiency declines and blade structural loads increase rapidly with tip air speed. While the risk from excessive tip speed is clear, it is not inherently obvious what the safe threshold for a given rotor is; compressibility effects are difficult to model accurately, and so the threshold must be explored experimentally. Recent work on the Sample Recovery Helicopters (SRH) and Chopper designs worked to a self-imposed advancing tip speed limit of Mach 0.95, based on successful tests completed up to that level in the JPL 25 foot Space Simulator during the SRH project. Tests at that level did not, however, show any adverse impact to rotor efficiency or blade structural integrity, suggesting that the tip Mach limit could be pushed further. Thrust increases as the square of rotor tip speed, so finding relaxation to the tip Mach constraint can provide significant added thrust capability. This paper will discuss the methodology and results of a newly completed Mars rotor test campaign that investigated operation with advancing blade tip speeds above Mach 1.0 in the JPL 25 foot Space Simulator. Two different rotors were tested in CO2 at Mars density. In both cases, the rotor under test was spun up to a constant rotor speed corresponding to a tip speed of Mach 0.98. Increasing levels of wind were then applied on the advancing side of the rotor disk, simulating air speed in forward flight, to push the advancing tip Mach number progressively higher. Wind speed was measured using a pitot-static probe, and the rotor under test was instrumented with a force-torque sensor and accelerometers. Both rotors were successfully pushed to advancing tip speeds of Mach 1.08 without encountering adverse effects on rotor efficiency or blade structural integrity. This result has provided direct benefit to the ongoing SkyFall helicopters development effort, as will be discussed in the paper.

The rapid expansion of experimental Unmanned Aerial Vehicles (UAVs) demands high-fidelity flight data acquisition, yet traditional Flight Test Instrumentation (FTI) ar- chitectures often introduce single points of failure or impose prohibitive size, weight, and power (SWAP) penalties. This paper proposes a distributed, fault-tolerant, and modular FTI architecture designed to minimize the payload footprint while maintaining rigorous data integrity. The system utilizes a “plug-and-stack” vertical topology with standardized 110 × 66 mm PCBs, significantly reducing the planar area required within the aircraft’s payload bay. Coordinated by a central Master Module, the architecture segregates critical sensing domains—Performance, Flight Instruments, and Electrical Systems—via a deter-ministic Controller Area Network (CAN) bus and a redundant I2C communication link.To bridge the gap between low-cost prototyping and formal airworthiness guidelines, the design incorporates galvanic isolation and hardware partitioning aligned with DO-178C and DO-254 safety-critical principles. Featuring a lightweight 212 g profile and a low 10 W power draw, the system eliminates active cooling requirements and prevents payload penalties. Functional bench tests, including Weight-on-Wheels (WoW) transitions and se-vere 90A current transient monitoring, validate the system’s robust fault containment and zero-packet-loss telemetry. This approach provides a scalable baseline for airworthiness aligned data acquisition in multi-mission scenarios.

Advances in high-fidelity, multidisciplinary simulations require high-frequency, accurately time-synchronized absolute surface pressure data for validation, in efforts towards the vision of virtual flight test as an approved process for certification. Classic measurement systems rely on remotely located pressure transducers connected via long tubing, introducing significant low-pass filtering, damping dynamic effects, and resulting in high installation costs. In the early 2000s, the feasibility of using Micro-Electro-MechanicalSystems (MEMS) pressure transducers for pressure belt applications was successfully demonstrated as a way to overcome these tubing-induced effects. Building upon this concept and extending it towards the current requirements of aerodynamic measurements, this paper introduces a modern, flexible, MEMS-based surface pressure measurement system developed by the German Aerospace Center (DLR). The system utilizes digital MEMS pressure sensors embedded in a 2.4 mm thin (1.9 mm in development), flexible Printed Circuit Board (PCB) assembly (“Belts”) that can be applied directly to unmodified aerodynamic surfaces, enabling extensive installations for Flight Test Instrumentation (FTI) and wind tunnel applications. While the general approach of using MEMS sensors on flexible substrates has been explored in the past, this new, fully digital system distinguishes itself from previous implementations through its unique combination of key features. It couples comparatively high data throughput with a thin footprint, dynamically adjustable pressure sampling rates up to 480 Hz (limited by the current sensor model to 200 Hz), support for different types of measurement nodes and sensors, and a high degree of configurability and scalability. The pressure belt segments can be cascaded into belts of 11 (plans for up to 16) elements, each featuring 32 pressure sensors or more, with a flexible spacing down to 8 mm. A range of additional sensors can be equipped. The belts are time-synchronized to microsecond accuracy using a proprietary protocol and the Precision Time Protocol (PTP) as far as possible, utilizing a in-house developed Switch-and-Timing Component (STC) that further handles connectivity with the belt segments trough a multi-drop Ethernet standard. The use of a single cable per segment and daisy-chained STC modules allows thousands of sensors to be connected via a single Power-over-Ethernet (PoE) cable, drastically simplifying installation. Furthermore, the underlying system architecture is designed for future expansion, with work already underway, including wireless communication and a further in-house developed approach for time synchronization using Ultra-Wideband (UWB) communication, with specific challenges in flexible antenna design and power supply yet to be overcome. This paper introduces this new system and its development, analyzing its capabilities and providing insights into its previous applications and those of its predecessors, to show the impact of this technology. Additionally, the improvements achieved by eliminating classical tubing are evaluated through the experimental determination of the pneumatic transfer function to the sensor. The paper details the experimental setup, outlines the methodology used and discusses the implications of inverted, surface-mount positioning of the sensors on high-frequency data accuracy. Initial results suggest a frequency response favorable for this approach of sensor placement.

Increasingly, aircraft operators evaluate fuel-saving modifications, such as finlets and other drag-reduction devices, through operational in-service evaluations rather than dedicated, expensive flight tests. Although their experimental design is not well documented, these evaluations follow a pattern: a representative subset of a fleet receives the modification (treatment), and a comparable subset remains in the baseline configuration (control), with data collected over many flights before and after installation. Raw data are conditioned for confounding factors, such as gross weight, airspeed, and altitude, using methods ranging from simple filtering on stabilized cruise to advanced statistical, physical, and machine-learning models. A difference-in-differences analysis of the conditioned data then estimates the treatment effect, using the control group to remove variation common to both groups. These studies are amenable to established experimental-design rules based on statistical power, anticipated effect size, and the irreducible between-airframe variability in conditioned performance. In principle, these rules can determine detectability prior to an evaluation. However, the open literature provides scant evidence of such prospective design practices. Further, the between-airframe performance variability that governs detectability has not been characterized as a design input. This paper develops an experimental-design framework that determines, in advance, whether a target effect is detectable using data from a baseline fleet. A closed-form detectability frontier relates the minimum detectable effect to two quantities: the number of airframes (“tails”) and the dispersion in per-tail performance change. The required fleet size scales with the square of this dispersion, making dispersion the dominant constraint, not data quantity. Thus, a study using a high-dispersion fleet cannot be rescued by collecting more data per tail. Because between-tail variance is irreducible under within-tail sampling, the individual tail is the unit of replication. This variance decomposition is supported by prior published analysis and corroborated across multiple operational narrowbody fleets. The operational range of between-tail variation is bracketed using independent reference data. Published data show that variation is minimal when aircraft are new and increases over time in service as airframes deteriorate at different rates. Combining the frontier with this range yields an iso-detectability map with contours of constant detectable effect over the plane of dispersion and fleet size. Feasibility is read directly from this map: improvements of approximately 1% are detectable across most fleets, sub-percent effects only on low-dispersion fleets, and effects below a quarter-percent are undetectable at any fleet size. After conditioning, detectability is governed by the true between-airframe heterogeneity in performance, a fixed but unobservable property of the fleet, independent of the method. Because the variation can only be estimated, the quality of conditioning is central to assessing feasibility: application of physical, statistical, or machine-learning models to accurately estimate the variation can improve the reliability of the assessment. This paper also identifies planning hazards that the idealized frontier ignores, such as the progressive attrition of tails from major maintenance events that disrupt the difference-in-differences design. These guidelines provide a quantitative basis for assessing the feasibility of an operational evaluation before committing resources, and can reveal when a credible evaluation is unachievable.

Hybrid vertical take-off and landing (VTOL) unmanned aerial vehicles (UAVs) combine the hover and maneuverability of multirotors with the efficient forward flight of fixed-wing aircraft. However, many research-level hybrid VTOLs remain difficult to manufacture, assemble, and maintain while supporting iterative prototype development. This paper presents the design, fabrication, integration, and ground testing of a 3D-printed tilt-rotor VTOL UAV with a 2 m wingspan, incorporating modularity features and developed as a research platform with potential for search-and-rescue applications. Aerodynamic, mass, and stability analyses were conducted using XFLR5, MachUp4, and OpenVSP/VSPAERO, while propulsion and power sizing were carried out in MATLAB based on momentum theory and aerodynamic inputs from XFLR5. CAD models were developed in Autodesk Fusion 360 and realized through multi-material fused-filament additive manufacturing using PLA+, PLA Aero, PC, and TPU, with carbon-fiber reinforcement incorporated for stiffness at low weight. Modularity features included a tab-and-socket wing–fuselage interface and a tail-boom splice connector, which ease assembly, transport, and maintenance. The airframe was integrated with a Pixhawk flight controller, power-distribution hardware, and a tilt-rotor mechanism, with an electrical layout arranged for component access and ease of replacement. In addition, YOLOv11 was fine-tuned using the VisDrone dataset for human detection in aerial imagery and evaluated on a Raspberry Pi as a proof of concept for future search-and-rescue applications. Ground testing covered actuator testing, motor characterization, static thrust evaluation, avionics calibration, and system-level integration. Static thrust testing measured a maximum thrust of 3.8 kg per motor, providing a total available thrust of 15.2 kg for a 6 kg UAV and a maximum thrust-to-weight ratio of about 2.5:1. Flight testing is planned as part of the ongoing validation process. This work demonstrates the practical use of multi-material additive manufacturing and modular design for developing manufacturable hybrid tilt-rotor VTOL UAVs, while documenting the build-and-test lessons obtained during prototype development.

High Intensity Radiated Fields (HIRF) from onboard and external RF sources are an escalating operational concern as airborne radars, SATCOM, high power datalinks, and terrestrial transmitters increase in power and spectral use. This paper presents a zone aware methodology to assess HIRF vulnerability of aircraft electronics by accounting for equipment mounting location, enclosure shielding performance, cable routing, and aperture/penetration characteristics. The approach maps systems by exposure—prioritizing components near cockpit glazing, antenna panels, and long exposed harness runs—and uses field to wire transfer concepts and enclosure insertion loss metrics to estimate likely induced voltages and internal field levels across critical frequency bands. We review principal radiated coupling mechanisms including aperture and penetration coupling, long conductor antenna effects, and nonlinear rectification at metal contacts and semiconductor junctions, and link these mechanisms to representative failure modes such as false sensor outputs, microprocessor resets or latch up, receiver desensitization, and thermal stress. A layered mitigation strategy is described that emphasizes enclosure shielding and aperture control, correct shield termination and filtered feedthroughs, broadband front end protection (filters, attenuators, RF limiters), and architectural measures such as redundancy and graceful degradation. Tradeoffs between added mass/complexity and immunity are discussed, with guidance for risk based allocation of protection to the most critical and exposed subsystems. Testing and certification guidance aligns with RTCA DO 160 Section 20 radiated susceptibility practices. We emphasize realistic test planning: selection of frequency bands and field levels driven by operational scenarios and zone mapping; representative mounting, harness routing, and interface terminations during tests; appropriate use of continuous wave, square wave, and pulse modulations; and both horizontal and vertical polarizations. A concrete radiated susceptibility case study for a Motor Converter Unit (MCU) illustrates DO 160 Category R execution, including calibration with an isotropic probe, antenna selection and positioning, instrumentation, dwell times, operational modes under simulated load, and acceptance criteria tied to functional monitoring and post test checks. Test findings are used to prioritize mitigations and verify effectiveness. The paper concludes with recommendations to integrate HIRF assessment early in the design cycle, perform staged verification from component to system and flight validation, and develop improved predictive field to wire transfer models for complex airframes. Updates to standards and test guidance are recommended to address emerging wideband and pulsed RF threats and to account for composite and hybrid structures. These practices enable robust, test grounded strategies to preserve avionics functionality and safety in evolving electromagnetic environments.

Counter-unmanned aircraft system (C-UAS) performance is often limited by sparse, noisy, delayed, or degraded sensor observations. Small UAS can operate in cluttered environments, produce weak or intermittent signatures, and perform abrupt maneuvers such as hovering, rapid acceleration, altitude changes, and sharp turns. Under these conditions, trackers based on generic constant-velocity or constant-acceleration assumptions may produce inaccurate state estimates, underestimate uncertainty, or lose track continuity. This work presents a feature-aided Unscented Kalman Filter–Interacting Multiple Model (UKF-IMM) tracker for robust UAS state estimation under challenging sensing conditions. The primary contribution is the integration of a probabilistic, digital-twin-based motion model with conventional analytical motion hypotheses within the UKF-IMM model bank. The learned model represents the UAS state using position and velocity in a common world frame. Given the current kinematic state and propagation interval, it estimates both the expected next state and the associated transition uncertainty. The digital twin motion model is learned directly from ground-truth UAS trajectories using a physics-informed residual architecture. A constant-velocity propagator provides a weak kinematic prior, while a neural network learns state-dependent deviations caused by acceleration, nonlinear maneuvers, and other unmodeled disturbances. The model parameterizes a heteroscedastic Gaussian transition distribution whose conditional mean combines the nominal propagation with the learned residual and whose covariance represents state-dependent process uncertainty. Incorporating this probabilistic transition model into the IMM enables the tracker to account for flight behaviors and uncertainty patterns that may not be adequately represented by fixed analytical models. Target features provide additional context for adapting motion-model probabilities and measurement uncertainty. These include observation-level characteristics, such as target shape and apparent size, and temporal behavioral characteristics, such as maneuver patterns and transitions among hovering, translational motion, turning, ascent, and descent. By combining kinematic estimates with appearance and behavioral context, the tracker is designed to respond more effectively to changes in target dynamics and observation quality. The tracker is developed and evaluated using the Multi-Modal Anti-UAV Dataset (MMAUD), which was used in the UAV tracking and pose-estimation track of the CVPR UG2+ Challenge. To extend the available test conditions, we also investigate the synthetic insertion and composition of UAS targets within MMAUD scenes. These controlled modifications support stress testing under variations in target number, trajectory, maneuver behavior, sensor visibility, observation degradation, and cross-modal consistency. The proposed method will be compared with conventional UKF and UKF-IMM baselines using state-estimation error, track continuity, uncertainty consistency, motion-model selection behavior, and robustness under degraded sensing. The resulting framework provides a physics-informed and uncertainty-aware approach for improving the adaptability and testability of C-UAS tracking systems.

An airborne relay can restore a radio path across an obstruction, but the aircraft must carry and power the payload long enough to provide useful service. This paper examines which placement and dwell targets warrant further development for a notional multirotor relay. The architecture assigns mission traffic to the payload and aircraft command, electrical power, and mechanical support to the carrier. That allocation defines two screens: a link check on both relay hops and a carrier check that closes the battery, mass, and hover-power loop. At 10 km separation and 120 m relay height, the reference 30-minute case closes at 4.77 kg with a 0.50 m rotor, while the 45-minute case exceeds both size limits. Moving the relay toward the ground endpoint raises the weaker-hop margin by about 1 dB without changing modeled hover demand, which makes it a provisional candidate for site assessment. Published hover data constrain one power coefficient. Fitting it to each test point alone moves the rotor-size boundary from 33 to 49 minutes, and pairing the least favorable point with a 15 percent hover-power allowance places that boundary near 28 minutes. The 30-minute target is therefore a candidate that installed hover-power measurements must confirm. Site-specific link measurements and a payload-fault check address the other risks to that decision. The assessment covers outbound traffic during hover. The contribution is a traceable path, organized by the NASA system design processes, from each failed screen to the design decision and interface test it calls for.

This paper evaluates a Positioning, Navigation, and Timing (PNT) integrity supervisor for unmanned aircraft systems (UAS) when only one navigation solution is available. The supervisor normalizes recorded observations, evaluates freshness, reported quality, kinematic consistency, operatingmode behavior, and receiver timing, and fuses the available evidence into GOOD, DEGRADED, INVALID, RECOVERING, or UNKNOWN states. Two ArduPilot multirotor datasets provided 45,510 retained primary-GNSS epochs. During nominal replay, more than 99% of epochs were classified as GOOD. Eight distinct fault runs were then applied to baseline-GOOD segments: a reported GNSS outage, GNSS navigation-solution freezes during hover and translation, a 75 m position step, a 100 m position ramp, +1000 ppm GNSS clock drift, GNSS-time freeze, and GNSS-time reversal. Each faulted run was paired with a nominal run under an identical fixed configuration. The reported outage and the time-reversal, timefreeze, and translation-freeze faults triggered alerts, whereas the hover freeze and the position-step, position-ramp, and clockdrift faults generated insufficient evidence to cross the supervisory thresholds. These results distinguish detector sensitivity from state-level alerting and show that a non-alert can arise either from limited observability or from configured fusion and dwell thresholds. While the deterministic replay did not estimate detection or false-alert probabilities, it established a reproducible baseline and identified the independent truth, raw-observable, and cross-source information needed for formal multi-source PNT integrity assessment.

This paper presents an adaptive flight-control framework based on an Explicit Model Following (EMF) direct adaptive control architecture for robust flight performance under actuator faults and aircraft model uncertainties. To support the proposed methodology, a customised nonlinear seven-degrees-of-freedom F-16 simulation model is developed in which all primary control surfaces can be commanded independently, enabling systematic representation of actuator failures, and mass-property variations. Building upon this reconfigurable modeling environment, an integrated control architecture combining Nonlinear Dynamic Inversion (NDI), model-reference tracking, optimization-based Control Allocation, and Multilayer Adaptive Neural Network compensation is formulated. Neural-network adaptation laws are derived using Lyapunov stability theory to ensure bounded closed-loop behavior in the presence of uncertainties and fault-induced dynamics. A further contribution of this work is the implementation of an automated MATLAB-based fault generation and evaluation framework, eliminating the need for manual fault injection and enabling repeatable large-scale simulation studies. The proposed approach is assessed through case study on representative fault scenarios, including mass-property deviations, and elevator failures, under both adaptive and non-adaptive controller configurations. Comparative simulation results demonstrate that the adaptive controller significantly improves command-tracking accuracy, transient response, and fault-recovery capability, with the greatest performance gains observed during severe multi-failure conditions. These results highlight the effectiveness of adaptive neural-network augmentation for enhancing the robustness and reconfigurability of nonlinear flight-control systems.

The rapid advancement in the application of unmanned aerial vehicles (UAVs) has expanded in logistics, disaster response, infrastructure inspection, and payload transportation. Among various multirotor configurations, hexarotors offer enhanced payload capacity and fault-tolerance compared to quadrotors, which makes them particularly suitable for carrying bulky external loads. The most common and mechanically simple method for payload transportation is the use of slung load configuration. The approach avoids the need for complex grippers or landing mechanisms. In addition to its simplicity, it faces significant challenges from the underactuated nature due to the translational-rotational coupling and the dynamic coupling between the hexarotor and the swinging load. The coupled system exhibits highly nonlinear, time-varying dynamics with additional degrees of freedom from the suspended load. These characteristics complicate attitude and position control, necessitating advanced strategies that account for the coupled dynamics, load swing suppression, and robustness against disturbances. Earlier studies primarily focused on quadrotors with suspended loads by developing the dynamic models that incorporate the coupling cable tension and load swing dynamics with multirotor dynamics. Various control strategies have been proposed, including PID-based methods, adaptive control, and model predictive control (MPC) for simultaneous UAV position tracking and load swing suppression. Notable contributions included geometric control approaches for attitude stabilization, and energy based swing damping techniques to mitigate the effects of load motion on the UAV. While significant progress has been made with quadrotors, research on hexarotor with slung load remains relatively limited, despite their advantages in payload capacity. To address the aforementioned challenges, the dynamical modeling of the coupled system is performed using Euler-Lagrange's equations, capturing both translational and rotational motion. A cascaded architecture is adopted, featuring Adaptive Sliding Mode Control (ASMC) in the outer loop and a backstepping technique in the inner loop. The sliding surface is designed to explicitly incorporate both the position errors and swing angle errors of the load. The outer-loop ASMC, featuring an adaptive reaching law gain, provides strong robustness against uncertainties induced by the swinging load, and external disturbances. The adaptive mechanism automatically tunes the gain for reaching law, making the transition to the sliding surface smooth while effectively mitigating chattering. This provides enhanced stability, particularly in operating regions where the system is most vulnerable to uncertainties and disturbances induced by slung load. Meanwhile, the backstepping controller in the attitude loop enables systematic, recursive design of virtual control naturally accommodating the nonlinearities for fast and precise stabilization. The combination effectively decouples the slower translational dynamics along with load dynamics from the faster rotational dynamics. The stability analysis of the closed loop is rigorously established through Lyapunov stability analysis for the outer loop as well as the inner loop. Furthermore, a comparative analysis have been conducted against existing methods in the literature, which reflects the superior performance of the proposed approach in terms of robustness, tracking accuracy, and disturbance mitigation for coupled system. The overall methodology thus provides a robust and effective solution tailored to hexarotor with slung load system, successfully addressing the inherent nonlinearities and disturbance mitigation requirements.

Modern fixed-wing autonomous unmanned aerial systems (UASs) rely on advanced guidance algorithms capable of balancing path-tracking accuracy, disturbance rejection, and computational efficiency. High performance of these algorithms is required to carry out operations in urban environments due to the complexity of the airspace, and is becoming increasingly important due as the development of urban air mobility progresses. Linear model predictive control (MPC) provides computationally efficient optimal guidance through quadratic programming (QP) formulations, making it desirable for real-time aircraft applications. However, using a linear dynamic system, either a linear time variant (LTV) or invariant (LTI) propagated unchanged across a long horizon may result in drifts from the true dynamics and predict a wrong-sign cross-track response, especially in presence of high wind and external disturbances. A nonlinear MPC can mitigate these challenges, however it has much higher computation demands compared to a linear MPC, requiring larger hardware and higher power where UASs suffer from stringent payload and power capacity. This paper introduces a method of predicting the dynamics of the guidance LTV system along the MPC horizon length using a QP-MPC receding trajectory manifold, re-linearizing the LTV guidance model along the predicted trajectory in a single-QP-step to improve the fidelity of the guidance model along the horizon including the high-bank maneuvers where the single-point linearization fails. Additionally, the functionality of the guidance algorithm is improved by augmenting the dynamics with a cross-track error integral and path-forward velocity. Further modifications include using the exact discretization for the linear system instead of the Euler discretization and adapting the cost of the states with varying cross-track error and course error. Severe wind disturbances are modeled using a Dryden turbulence framework with steady winds equal to 18\% of cruise velocity and gusts reaching 46\% of cruise velocity, the proposed MPC guidance algorithm reduced cross-track error by a factor of two compared to conventional $L_1$ and $L_n$ guidance both in simulation and flight test. The method also reduces the path-departure issues possible with standard MPC guidance due to the lack of cost on incorrect path direction, improving path-following reliability under severe disturbances when following complex paths, necessary for flying in urban environments. One solution this algorithm provides is the ability to be tuned simultaneously for any turning angles, as opposed to $L_1$ guidance which may be only optimized for a single turn angle. Although $L_1$ guidance remained approximately three times faster computationally, the proposed algorithm was only 36\% slower than conventional MPC, offering substantial performance gains with minimal additional computational cost. The proposed algorithm was successfully flight tested in September 2026, and also demonstrates the gust rejection as there was 17\% cruise steady wind and up to 48\% cruise gusts.

In aerospace engineering, control theory has enabled aircraft to stay around a safe default with little to no human intervention. However, if the aircraft encounters any unexpected changes, e.g., from environmental factors like wind turbulence, classical control algorithms may not be enough to help the aircraft correct its course. In this paper, we examine the use of different deep neural network (DNN) architectures to predict new states of a quadcopter in flight. We use a simulator to sample trajectories and use them to train three different DNN architectures: Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). After comparing their performance, we use our chosen DNN architecture to train an RL-based controller to predict the action that best keeps the quadcopter on course. We believe that training DNNs on more complex dynamic systems or applying trained controllers to more complex vehicles would merit further study, with implications for improvement in aircraft control and safety.

This paper presents the design, implementation and validation of a lightweight azimuth control system for a high- altitude balloon payload carrying a calibrated light source for ground-based telescope calibration. The ALTAIR (Airborne Laser for Telescopic Atmospheric Interference Reduction) payload requires the exit port of the light source to remain directed into the aperture of a telescope during multi-second exposures. The payload must also abide by strict mass, power and volume constraints. Additionally, due to the small size of the payload, wind disturbances and motor friction effects are relatively larger as compared to large balloons, rendering azimuth pointing of small balloon-borne payloads challenging. This paper presents a comprehensive pointing architecture that combines passive stabilization, innovative at- titude determination technologies, and active azimuth control using a momentum-wheel concept with active desaturation using a second motor. A dynamic model of the payload is simulated in a high-fidelity environment including motor dynamics and limits, sensor noise, and wind-disturbance modeling through both generated and experimentally measured wind profiles. The simulation is used to validate actuators and tune various control loops. Attitude determination is performed using onboard sensor fusion of rate gyros, accelerometers, magne- tometers, and carrier phase measurements from GNSS antennas. The payload avionics, flight software, power management, motor control, and real-time ground telemetry systems are developed by the authors. The system is evaluated through ground testing, simulation, and flight data. Passive stabilization reduces off- axis pendulating motion sufficiently for azimuth control to be treated independently. In flight, a pointing root-mean-square error of 5◦ is realized over an 8 minute window. These results exceed the mission requirements of maintaining pointing within ±8◦ for at least 30 seconds that would allow for multiple long exposure images with an uncertainty of approximately ±1 % in measured flux due to angular misalignment. The results demonstrate that precise pointing of a lightweight balloon payload can be achieved using a sim- ple and modular payload and flywheel design. A wheel desaturation method was also developed to address the main limitation for momentum-wheel based pointing which is saturation under constant and biased distur- bances. A motorized connection between the payload and flight train was designed and built to address this problem, dumping the accumulated momentum from the wheel to the balloon and allowing for extended pointing periods. Although this desaturation system has not been flight validated, simulation and ground-test results are promising. By en- abling precise and sustained pointing on mass and power-constrained balloon platforms, this work expands the capabilities of small high-altitude balloons as low-cost, rapidly deployable platforms for astronomy, atmospheric science, remote sensing, communications, and various scientific experiments.

Vehicles operating among obstacles that move over time require paths that are recomputed frequently and on a strict time budget. We present Model Predictive Planning (MPP), a real-time path planner that represents obstacles as smooth bumps on a continuous cost surface and routes the path through the low-cost valleys between them. Our earlier work applied this cost surface to a single static snapshot; MPP's contribution is to run it in real time - on every cycle it re-solves the complete path from the vehicle's current state to the goal, so the same snapshot formulation handles moving obstacles by re-solving fast enough to keep up. This solve - execute - re-solve loop applies the receding-horizon principle of model predictive control at the planning layer rather than the control layer: it produces a reference path, not control inputs, hence Model Predictive Planning. The path is a Bernstein polynomial whose control points bound the entire curve, so the whole path is kept clear of obstacles. MPP determines the geometric path to be tracked by the vehicle. The formulation is purely geometric - it optimizes the path's shape, trading off arc length against obstacle clearance and prescribes no temporal parameterization. Far from a limitation, this leaves the speed profile as an additional degree of freedom for the tracking controller, or a higher coordination layer, to assign. We exploit exactly this freedom in our demonstration: a single unchanged planner - same cost surface, solver, and parameters - fed to three different vehicle models, each tracked by its own controller that enforces that vehicle's motion limits. The three are (i) a double-integrator under a PD tracking law, bounded in speed and acceleration; (ii) a fixed-wing UAV whose tracking controller holds a nonzero minimum speed and respects a minimum turn radius; and (iii) a differential-flatness controller that recovers the required inputs by differentiating the path and throttles speed by curvature to stay within its acceleration bound. The planner models no vehicle. We evaluate this across all three controllers in scenes of static and moving obstacles, confirming that MPP re-plans within its real-time budget and that every flown trajectory stays clear of obstacles while respecting its controller's limits.

Perching is a biologically inspired maneuver in which a fixed-wing aircraft emulates avian landing by executing a rapid, high-angle-of-attack terminal trajectory to induce aerodynamic stall and achieve compact landing without a runway. This capability provides distinct advantages for unmanned aerial vehicles (UAVs) operating in urban, inspection, search-and-rescue, or emergency-response environments, where conventional landing infrastructure may be unavailable. Despite these operational advantages, generating feasible perching trajectories is notoriously difficult; the combination of nonlinear high-angle-of-attack aerodynamics, longitudinal coupling, and stringent terminal constraints creates a highly nonconvex optimization problem. Consequently, while deterministic optimization-based methods can produce feasible solutions in restricted settings, their convergence behavior, computational cost, and sensitivity to initial conditions limit their use for real-time autonomous planning. To address these limitations, this paper proposes a feasibility-first trajectory planning framework based on a Closed-Loop Rapidly-exploring Random Tree (CL-RRT) algorithm. While open-loop RRT frameworks are well-suited for high-dimensional spaces and real-time planning, open-loop reachability ignores the restrictive impact of the flight control system and fails to capture the true closed-loop dynamics necessary for real-world viability. CL-RRT mitigates this by embedding high-fidelity 6-DoF dynamics and an active control loop into the branching mechanism, shifting it from geometric steering to a dynamic forward simulation that approximates the aircraft's practical controller-induced reachable set. Each sampled state triggers a brief simulation of the aircraft's closed-loop tracking response, and the branch is accepted only if the propagation is collision-free and dynamically feasible. However, generic guidance and control architectures are fundamentally insufficient for managing the severe, highly constrained dynamics of fixed-wing perching, leading to poor assessed reachability and failed branching attempts in CL-RRT. Therefore, the proposed framework introduces tailored guidance and control formulations that redefine the CL-RRT update loop. A Total Energy Control System (TECS) architecture is embedded within the planner's propagation step to regulate altitude and airspeed through explicit energy management while simultaneously achieving high-angle-of-attack commands to induce stall when rapid energy dissipation is required. Because traditional linear look-ahead methods degrade across short, irregularly spaced propagation windows, a method of pure pursuit guidance is utilized during tree expansion. Furthermore, unlike geometric RRT methods, the planner utilizes a dynamics-aware distance metric, overcoming the structural shortcomings of Euclidean distance in accounting for the nonholonomic and constrained behavior of fixed-wing systems. The final paper will evaluate the proposed sampling-based method against deterministic trajectory-generation baselines (e.g., shooting and direct collocation) in perching simulations, with emphasis placed on success rate, convergence behavior, planning time, and sensitivity to initial conditions.

This paper presents a data-driven framework for Lyapunov stability characterization of nonlinear closed-loop systems directly from sampled trajectory data. Without requiring an explicit model of the closed-loop dynamics, the proposed method constructs quadratic Lyapunov candidates via a Cholesky-parameterized neural network and enforces a relaxed discrete-time decrease condition inspired by input-to-state stability, which incorporates a robustness margin to accommodate noise, model mismatch, and unmodeled dynamics. Building on this formulation, a bisection-based procedure is developed to jointly estimate the largest admissible exponential decay rate λ¯⋆ and the corresponding robustness margin ¯δ ⋆ from data. A regime-wise unified stability characterization framework is further proposed, in which trajectory-wise candidates are averaged to form a shared candidate that is validated on unseen hold-out trajectories. The framework is evaluated on a quadrotor UAV altitude control system in a PX4/Gazebo software-in-the-loop environment, treated as a black-box closedloop system. The proposed method achieves 100% satisfaction rate of the relaxed decrease condition on both nominal-takeoff and altitude-step trajectory sets with λ¯⋆ ≈ 1.0, compared to satisfaction rates below 58% and λ¯⋆ < 0.001 for strict neural Lyapunov and fixed quadratic baselines. Unified regime-wise Lyapunov candidates generalize successfully to unseen trajectories in both operating regimes, providing quantitative evidence of shared contraction structure across maneuver classes

This paper presents the development of a hardware-in-the-loop (HITL) simulation framework for the DJI M300 RTK and the implementation and validation of trajectory-tracking controllers, including proportional-integral-derivative (PID), linear-quadratic regulator (LQR), and model predictive control (MPC). The framework integrates the DJI Onboard SDK (OSDK) with ROS Kinetic on the DJI Manifold 2-G onboard computer, enabling real-time onboard execution of the control algorithms and closed-loop HITL evaluation using the DJI Assistant 2 simulator. Both LQR and MPC are formulated using the same linear models, and all three controllers share a common trajectory generation pipeline. Experimental validation on multiple trajectories demonstrates satisfactory tracking performance for all three controllers. Observed performance differences are primarily attributable to controller tuning rather than inherent algorithmic limitations. Wind-disturbance tests further indicate robust tracking under wind disturbances, with noticeable degradation occurring only when attitude saturation limits are reached. Overall, this work provides a validated and reproducible HITL framework for implementing and evaluating trajectory-planning and control algorithms for the DJI M300 RTK.

This paper presents a convex optimization-based trajectory-planning framework for unmanned aerial vehicles (UAVs) in obstacle-dense environments. Obstacles are decomposed into convex polygons, and a bidirectional Rapidly-exploring Random Tree (RRT) generates a coarse reference path. Along this path, overlapping convex regions are greedily constructed to form a safe connected corridor without full workspace decomposition. Dynamic feasibility is enforced via differential flatness, and convex-hull approximations of admissible phase-space sets are imposed as linear constraints on Bézier derivative control points. Combined with corridor-containment and continuity constraints in a single convex program, the formulation produces collision-free, dynamically feasible trajectories without nonlinear refinement. Compared to a Graph of Convex Sets (GCS) planner refined with Kinodynamic Trajectory Optimization (KTO), the proposed method achieves a 36% reduction in traversal time and a mean computational speedup of 17×. Monte Carlo experiments over 1,001 randomized start–goal configurations demonstrate a 99.4% success rate.

Tailless Unmanned Combat Air Vehicles (UCAVs) can introduce significant coupling between longitudinal and lateral/directional dynamics during the deployment of elevons. Effective control of such a coupling effect becomes significant, particularly when elevon deflections generate simultaneous pitching, rolling, and yawing moments. Because such cross-coupling behavior can degrade flight-path tracking, induce undesirable lateral-directional responses, and adversely affect the robustness of conventional single-channel controllers. This study attempts to bring out the coupling effects arising from elevon deployment in a tailless UCAV. This article presents a Model Predictive Control (MPC) framework for cross-coupling disturbance rejection while ensuring dynamic decoupling of the longitudinal and lateral/directional channels. For this purpose, a six-degree-of-freedom (6-DOF) nonlinear flight-dynamics model of the UCAV is formulated by incorporating aerodynamic force and moment derivatives and the control effects of the elevons. The coupling characteristics are then quantified by analyzing the response of longitudinal variables, such as angle of attack and pitch rate, to lateral/directional control inputs and also with the corresponding response of roll and yaw variables to longitudinal control inputs. The resulting cross-channel interactions are then injected into a state-space representation used for MPC controller synthesis. The MPC simultaneously predicts the evolution of coupled aircraft states over a finite prediction horizon and determines the optimal control inputs, considering the actuator and state constraints. The cross-coupling terms are explicitly considered in the prediction model. It enables the controller to compensate for unwanted interactions while maintaining the desired longitudinal and lateral/directional responses. Simulation studies demonstrate that elevon deployment produces appreciable cross-coupling between the longitudinal and lateral/directional modes and vice versa. Therefore, an independently designed single-input single-output (SISO) controller cannot effectively suppress the cross coupling of dynamics. The proposed MPC significantly attenuates cross-channel disturbances and reduces tracking errors and transient interactions. It improves the decoupling between longitudinal and lateral/directional responses. The controller also provides improved robustness under various operating conditions and actuator constraints compared to control strategies with a conventional decoupled nature. The results establish that MPC-based multivariable control provides an effective approach for managing the inherently coupled dynamics of tailless UCAVs. The methodology adopted is mainly significant for high-agility configurations employing elevons as primary control surfaces, where aerodynamic coupling and control-surface interactions are prominent. The work contributes toward robust, integrated flight-control architectures for advanced UCAVs by demonstrating simultaneous trajectory tracking, cross-coupling disturbance rejection, and dynamic decoupling as a cohesive control framework. Keywords: Model Predictive Control (MPC); Unmanned Combat Air Vehicle (UCAV); elevon; longitudinal dynamics; lateral/directional dynamics; cross-coupling; dynamic decoupling; multivariable control; disturbance rejection; six-degree-of-freedom (6-DOF) flight dynamics.

This paper examines the performance and accuracy of the inner control loop using an intelligent flight control that couples adaptive sliding mode control (ASMC) with a trained reinforcement-learning framework (RL). The RL framework tunes the switching gain and boundary layer online to overcome the limitations of traditional sliding mode control and proportional integrative derivative (PID) by providing a chattering-free control signal and maximizing characteristic performance, and it has been conducted under nominal conditions. Two different conditions are considered in this paper. Nominal conditions: mean error is reduced with the proposed framework as compared to SMC and PID controllers. Under the model mismatch condition: with a +50% increase in inertia, a reduced thrust coefficient by 15%, and variable speed of motion in the presence of a gust, the proposed framework reduces the mean tracking error by 44% relative to SMC and recovers the pitch axis from 0% to 100%, with improvements in rise time on every axis. The results demonstrate the improvements in transient response and accuracy increase in actuator total variation as the policy collapses the boundary layer and recovers near-discontinuous switching. The proposed SMC-RL framework represents a notable advance in nonlinear control, offering a robust solution for systems characterized by complex dynamics and disturbance. It shows improvement on conventional sliding mode controllers and PID by reducing steady-state error and chattering while retaining the capacity to handle unbounded external disturbances.

Iterative development and simulation are essential for advancing innovation in guidance, navigation, and control (GNC) for autonomous fixed-wing unmanned aircraft systems (UASs). However, assembling a complete software-in-the-loop (SiTL) environment requires compatible flight software, aircraft models, development tools, and simulation dependencies, creating barriers for users without specialized software-integration expertise. This paper presents CloudUAS, a cloud-based integrated development and simulation environment for autonomous fixed-wing UASs. Authenticated, browser-accessible workspaces combine code editing, software building and execution, closed-loop simulation, visualization, and persistent storage without requiring installation of the complete stack locally. Users develop and execute GNC software within the same Robot Operating System (ROS 2) workspace, supporting a continuous edit–build–run–inspect workflow. Built within the CogniPilot Autopilot environment, the simulator incorporates an aircraft-specific nonlinear six-degree-of-freedom model of the Hobbyzone Sport Cub S2 and uses consistent ROS 2 interfaces for simulation and subsequent physical testing. We evaluate the environment through a pylon racing competition involving students and industry participants, with virtual development followed by indoor mixed-reality flight testing. During the five-week competition, the median time from remote desktop login to a first running simulation was approximately three minutes, and 41 of 55 participants used the integrated VSCode environment to open ROS 2 projects. Participants generated 2,357 simulator sessions and 648 scored submissions, demonstrating sustained experimentation. In one documented example, a participant completed 59 scored runs within 9.2 hours, reducing the scored lap time from 19.95 to 12.84 seconds. The open-source simulator, a sample GNC solution, and flight data collected during the race are available via GitHub and upon request from the authors.

Complex robotic systems on Mars frequently interface with the external environment and auxiliary hardware under extreme diurnal thermal cycles. These severe temperature fluctuations induce structural distortions which, upon inter-vehicular docking, translate into significant mechanical loads that threaten structural integrity and mission longevity. This paper presents a methodology for modeling and computing these thermally induced loads by combining precomputed free-space thermal deflections from a full-scale finite element model (FEM) along a diurnal cycle with a computationally efficient reduced-order compliance model for the kinematic chain spanning both vehicles (six degrees of freedom per kinematic link). Unlike bulk thermal expansion models used on previous Mars missions, this approach captures spatially-varying thermal gradients and time-dependent interface loads while avoiding the computational cost of full FEM analysis. Validation against full FEM simulations demonstrates strong agreement in predicted interface loads. The method was integrated into a scalable robotics analysis pipeline, which analyzed rover parking and docking operations with a stationary lander across multi-million-pose configurations and identified safe parking envelopes under worst-case Martian diurnal cycles.

Mission-level studies of distributed space-object surveillance and identification (SOSI) architectures require thousands of sensor evaluations across hardware configurations, environmental conditions, target-observer geometries, and clutter regimes. Higher-fidelity three-dimensional image-formation simulations preserve the electro-optical phenomenology needed for calibration but are too slow for campaign-level architecture trades; simple probability-of-detection models are fast but omit the track formation, catalog association, ambiguity, and false-track mechanisms that drive mission performance. This paper presents an implemented Python framework that uses an image-producing, medium-fidelity electro-optical simulation to train a reduced-order model (ROM) for computationally accelerated SOSI mission analysis. The simulation represents sensor and optical hardware, atmosphere and sky state, target geometry and brightness, natural clutter, point/streak detection, temporal tracklet formation, catalog association, and truth-referenced scoring. The ROM predicts eight mission-response quantities spanning truth measurement and association, false-track burden, track quality, measurement and tracklet counts, uncertainty, and runtime. The ROM is an eight-target Ridge regression model trained on 166 cases from a 195-case blocked design-of-experiments campaign spanning 13 physical geometry groups and variations in sensor/optical parameters, atmosphere and sky state, target brightness and angular rate, natural clutter, and association context. 29 cases from a difficult geometry family were excluded as a holdout dataset. On that holdout, mean absolute errors were 0.411 for both measured-truth and correct-association fraction, 0.318 for track-quality score, and 0.258 for false-track rate, with an 11.8% truth-fraction MAE improvement over the pre-augmentation baseline. A separate 70-case validation campaign covering 12 independent source geometries produced in-domain MAEs of 0.151, 0.151, 0.096, and 0.137 for the same four responses. Uncertainty is retained as an explicit ROM response and assessed through residual coverage. The independent campaign achieved 91.4% coverage under a locked absolute-error criterion. Typical medium-fidelity cases required 51-106 seconds per scenario, with a median runtime of 64 seconds on an Intel Core i7 with 16 processor cores and 64 GB of RAM. In a representative Python batch benchmark, evaluation of the eight-target Ridge architecture was millisecond-scale, indicating approximately four orders of magnitude reduction in scenario-evaluation time. The present work quantifies this medium-fidelity-to-ROM acceleration; a controlled same-hardware comparison with a separate full-three-dimensional image model remains future work. Additionally, benchmarking the medium-fidelity model and ROM to real world data is currently being pursued. The implemented model is therefore reported as a reproducible, bounded research ROM rather than an operationally validated capability. This result provides a traceable computational bridge from sensor phenomenology to architecture analysis while explicitly identifying the calibration and domain-expansion work required for future operational use.

Airfoil Shape Optimization (ASO) is essential to modern aircraft design. However, the high computational cost of aerodynamic analysis, coupled with the high-dimensional design space intrinsic to such problems, renders ASO a challenging task. Metaheuristic optimization algorithms, particularly swarm- and evolutionary-based methods, have emerged as promising alternatives to conventional gradient-free optimization strategies because they are highly parallelizable and can be readily integrated with surrogate models. Despite their growing use, the relative performance of different metaheuristic algorithms, as well as their interaction with geometric parameterization methods, remains insufficiently understood. To address this gap, this study presents a systematic parametric comparison of five swarm- and evolutionary-based optimization algorithms across five geometric parameterization methods. Namely, PARSEC, Class/Shape function Transformation (CST), B-splines, Hicks-Henne bump functions and Free-Form Deformation (FFD) are used as parameterization methods coupled with Genetic Algorithms (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE), Covariance Matrix Adaptation Evolution Strategy (CMA-ES), and Artificial Bee Colony (ABC) as optimization algorithms. Furthermore, as a computationally efficient aerodynamic predictor, the study employs a previously developed and validated deep-learning surrogate trained on Reynolds-Averaged Navier-Stokes simulation data. The RAE2822 airfoil is optimized in transonic viscous flow in the drag-divergence Mach range at a Reynolds number of Re = 2.7×106 and under a lift coefficient constraint of CL = 0.733. This study evaluates single- and multi- point optimization scenarios, with the objective to minimize the drag coefficient CD. The one-point case is evaluated at a Mach number of M = 0.74, the two-point case at M = 0.68 and 0.74, and the four-point case at M = 0.68, 0.71, 0.74, and 0.76. In the single-point scenario, the B-spline & CMA-ES combination achieves the best performance, with a 74% drag reduction. In the two- and four-point case, the FFD & ABC combination was the best performing, achieving 75% and 77% improvements in their respective multi- point drag reduction optimization objectives. Among the optimizers, DE is the least sensitive to the choice of parameterization, exhibiting a standard deviation of only 12 drag counts, compared with 23, 21, 34, and 19 drag counts for GA, PSO, CMA-ES, and ABC, respectively. This paper provides a reproducible framework for evaluating metaheuristic optimization strategies for airfoil design and offers insight into how algorithm choice and shape parameterization jointly influence ASO performance.

On-orbit servicing, assembly, and cooperative robotics missions such as Northrop Grumman's Mission Extension Vehicle, Starfish Space's Otter, and Astroscale's servicers seek to operate multiple vehicles increasingly transitioning between free flight and docked operation. Across these phases, realistic spacecraft simulation and control design call for multibody models carrying internal joints and kinematic constraints, including manipulators, gimbals, and docking mechanisms. When two such vehicles dock, conventional approaches either assemble a single monolithic constrained system or restructure a recursive solver into one unified tree. Both demand global re-rooting of the multibody topology, redefinition of reference frames, and reassembly of the mass and constraint matrices, which discards the per-vehicle modularity that guidance, navigation, and control architectures are built around. This paper presents a modular constrained-dynamics formulation that preserves each vehicle's high-fidelity multibody model and couples the vehicles only at the docking interface. Beginning from the Karush-Kuhn-Tucker system for constrained rigid-body dynamics, this work partitions the global problem by vehicle and groups the constraints into intra-vehicle joints and inter-vehicle docking sets. A central contribution is the derivation of a nested Schur-complement reduction in which each vehicle resolves its internal constraints independently, producing an explicit mapping from the shared docking forces to its own accelerations and internal reactions. A single low-dimensional interface solve recovers the docking forces, which back-substitute to complete each vehicle's state. The same nesting extends to chains of three or more docked vehicles. The nested reduction is algebraically equivalent to solving the monolithic system for the fully assembled configuration, whereas it preserves the model and software modularity those approaches give up. Docking is handled without re-rooting or recentering the per-vehicle coordinate trees that recursive multibody solvers require, so existing independently maintained spacecraft models, and the control laws designed around them, couple with minimal refactoring. The computational structure separates the large internal solve of each vehicle from a small interface solve whose size is set by the number of docking constraints, six for a rigid connection, rather than the total system degrees of freedom. Each docking constraint must still propagate through a vehicle's internal dynamics, so the coupling cost grows with the product of the interface size and the internal degrees of freedom. For high-fidelity vehicles joined by a small rigid interface this cost stays small, the interface solve remains independent of total system size, and each vehicle's dynamics admit parallel evaluation. These claims are proven through a standalone reference implementation of the method. This work demonstrates numerical equivalence against monolithic solutions across accelerations, reactions, and constraint violation, and that equivalence forms the backbone of the paper's validation. Runtime measurements from the same implementation characterize how the method scales with internal degrees of freedom and with the number of docked vehicles, confirming the analytical cost structure. By preserving subsystem autonomy while holding exact equivalence to classical constrained formulations, this work delivers a scalable, implementation-ready foundation for high-fidelity simulation of docked spacecraft and on-orbit assembly of multiple vehicles.