Periodic Reporting for period 4 - EXTREMA (Engineering Extremely Rare Events in Astrodynamics for Deep-Space Missions in Autonomy)
Periodo di rendicontazione: 2025-07-01 al 2025-12-31
Solar system exploration is carried out with probes guided from ground. During their cruise toward planets, asteroids, or comets, interplanetary probes are operated by dedicated teams that reconstruct the spacecraft position and elaborate a flight plan consistent with the mission requirements. This approach requires a significant effort in terms of human resources deployed and ground-based assets used. Given the time scale of an interplanetary cruise (months or even years), the current paradigm makes it considerably costly to operate spacecraft from ground.
In the last decade, miniaturization of electronics has enabled nanosatellites, or CubeSats: shoebox-sized systems that are able to carry out scientific investigations alike conventional spacecraft. CubeSats have reduced the entry-level cost in low-Earth orbit by one order of magnitude, and are nowadays the standard for NASA, ESA, and ASI, among others. Nevertheless, the current paradigm prevents their usage for deep-space exploration: operating CubeSats might require the same costs as developing them, so canceling their advantages. This is the context in which EXTREMA operates.
Engineering Extremely Rare Events in Astrodynamics for Deep-Space Missions in Autonomy, or EXTREMA, introduces “self-driving interplanetary CubeSats”: miniaturized probes able to drive themselves during the cruise, without requiring any contact with ground. The idea is challenging: nanosatellites must self-determine their position by sensing the environment, and must elaborate a guidance law to actuate by virtue of miniaturized thrusters. The project embeds elements of artificial intelligence and exploits ballistic capture, a delicate celestial mechanism to acquire an orbit about a planet.
EXTREMA achieved a validation of the self-driving spacecraft concept in laboratory environment by virtue of a hardware-in-the-loop simulation policy within an accelerated framework. Almost 100 successful simulations were completed, for a total runtime of more than 430 hours of HIL simulation. Thanks to the accelerated framework, this corresponds to about 40000 simulated days of interplanetary transfers. This is a remarkable result, which goes beyond the state of the art to a significant extent.
In the last decade, miniaturization of electronics has enabled nanosatellites, or CubeSats: shoebox-sized systems that are able to carry out scientific investigations alike conventional spacecraft. CubeSats have reduced the entry-level cost in low-Earth orbit by one order of magnitude, and are nowadays the standard for NASA, ESA, and ASI, among others. Nevertheless, the current paradigm prevents their usage for deep-space exploration: operating CubeSats might require the same costs as developing them, so canceling their advantages. This is the context in which EXTREMA operates.
Engineering Extremely Rare Events in Astrodynamics for Deep-Space Missions in Autonomy, or EXTREMA, introduces “self-driving interplanetary CubeSats”: miniaturized probes able to drive themselves during the cruise, without requiring any contact with ground. The idea is challenging: nanosatellites must self-determine their position by sensing the environment, and must elaborate a guidance law to actuate by virtue of miniaturized thrusters. The project embeds elements of artificial intelligence and exploits ballistic capture, a delicate celestial mechanism to acquire an orbit about a planet.
EXTREMA achieved a validation of the self-driving spacecraft concept in laboratory environment by virtue of a hardware-in-the-loop simulation policy within an accelerated framework. Almost 100 successful simulations were completed, for a total runtime of more than 430 hours of HIL simulation. Thanks to the accelerated framework, this corresponds to about 40000 simulated days of interplanetary transfers. This is a remarkable result, which goes beyond the state of the art to a significant extent.
EXTREMA builds on three Pillars. On top of the three pillars lies the EXTREMA Simulation Hub, where the outcome of the pillars is merged to perform integrated simulations.
In Pillar 1 (Autonomous Navigation), we developed and completed an end-to-end autonomous vision-based navigation pipeline for deep-space missions, spanning image processing, attitude determination, and full state estimation. The navigation chain was implemented and executed on representative space-grade processors. To support validation activities, a high-fidelity space rendering engine was developed for both software-in-the-loop and hardware-in-the-loop simulations, enabling accurate modeling of the camera and imaging chain in SIL and the stimulation of real sensors in HIL with geometric and radiometric consistency representative of on-orbit conditions.
In Pillar 2 (Autonomous Guidance), an autonomous closed-loop guidance approach is developed to achieve the mission target by repeatedly recalculating the optimal trajectory to compensate for thrust execution errors, navigation errors, and model approximations. In particular, the core of the guidance algorithm is based on successive convex programming. This algorithm has been developed by paying particular attention to the computational burden, in order to make it not only deployable but also executable aboard. It is integrated into the simulation framework in a plug-and-play fashion, enabling easy testing of alternative approaches, such as the indirect algorithm.
In Pillar 3 (Ballistic Capture), we characterized ballistic capture corridors, studying their peculiarities and understanding their potential exploitation as pathways guaranteeing temporary capture at major planets for autonomous, deep-space CubeSats. We devised an autonomous ballistic capture algorithm for the inexpensive synthesis of ballistic capture corridors. The algorithm was made compatible with limited-capability, autonomous, interplanetary CubeSats. Pre-computed ballistic capture sets at Mars at different epochs are publicly available on Zenodo.
For what concerns the ESH, we have developed the building blocks, namely SPESI, STASIS, and the flatsat OBC. SPESI is the Space Environment Simulator, which takes care of the spacecraft-environment simulation and represents the simulation authority. STASIS is the SpacecrafT Attitude Simulation System, which is used to simulate the spacecraft attitude dynamics and control. The platform’s attitude control is provided by a bespoke suite of actuators, consisting of four reaction wheels in a pyramidal configuration. Each unit integrates a BLDC motor with a custom-designed PCB for motor driving and logic control, paired with a prototyped flywheel engineered for specific torque profiles. The system is managed by a dedicated ADCU, which fuses optical and inertial data via a Kalman Filter to execute the control laws and synchronize the actuator response with the satellite state machine. The flatsat On-Board Computer (OBC) was developed in terms of both software and hardware; it controls the pillars experiments and represents the spacecraft authority.
In Pillar 1 (Autonomous Navigation), we developed and completed an end-to-end autonomous vision-based navigation pipeline for deep-space missions, spanning image processing, attitude determination, and full state estimation. The navigation chain was implemented and executed on representative space-grade processors. To support validation activities, a high-fidelity space rendering engine was developed for both software-in-the-loop and hardware-in-the-loop simulations, enabling accurate modeling of the camera and imaging chain in SIL and the stimulation of real sensors in HIL with geometric and radiometric consistency representative of on-orbit conditions.
In Pillar 2 (Autonomous Guidance), an autonomous closed-loop guidance approach is developed to achieve the mission target by repeatedly recalculating the optimal trajectory to compensate for thrust execution errors, navigation errors, and model approximations. In particular, the core of the guidance algorithm is based on successive convex programming. This algorithm has been developed by paying particular attention to the computational burden, in order to make it not only deployable but also executable aboard. It is integrated into the simulation framework in a plug-and-play fashion, enabling easy testing of alternative approaches, such as the indirect algorithm.
In Pillar 3 (Ballistic Capture), we characterized ballistic capture corridors, studying their peculiarities and understanding their potential exploitation as pathways guaranteeing temporary capture at major planets for autonomous, deep-space CubeSats. We devised an autonomous ballistic capture algorithm for the inexpensive synthesis of ballistic capture corridors. The algorithm was made compatible with limited-capability, autonomous, interplanetary CubeSats. Pre-computed ballistic capture sets at Mars at different epochs are publicly available on Zenodo.
For what concerns the ESH, we have developed the building blocks, namely SPESI, STASIS, and the flatsat OBC. SPESI is the Space Environment Simulator, which takes care of the spacecraft-environment simulation and represents the simulation authority. STASIS is the SpacecrafT Attitude Simulation System, which is used to simulate the spacecraft attitude dynamics and control. The platform’s attitude control is provided by a bespoke suite of actuators, consisting of four reaction wheels in a pyramidal configuration. Each unit integrates a BLDC motor with a custom-designed PCB for motor driving and logic control, paired with a prototyped flywheel engineered for specific torque profiles. The system is managed by a dedicated ADCU, which fuses optical and inertial data via a Kalman Filter to execute the control laws and synchronize the actuator response with the satellite state machine. The flatsat On-Board Computer (OBC) was developed in terms of both software and hardware; it controls the pillars experiments and represents the spacecraft authority.
In the autonomous navigation pillar, we have shown that the celestial triangulation is a viable option to navigate deep-space probes without human intervention. We demonstrated that autonomous vision-based navigation is a viable and effective solution for deep-space probes, reaching a maturity level of TRL 5. Extensive validation campaigns showed that the proposed approach achieves position estimation errors below 1025 km (3σ) and velocity errors below 0.42 m/s (3σ) in software-in-the-loop simulations, and below 5320 km (3σ) and 0.97 m/s (3σ), respectively, in hardware-in-the-loop simulations. The image-processing pipeline achieved an attitude determination accuracy of 15 arcsec (1σ) and a planet centroiding accuracy better than 0.3 pixels (3σ), and was successfully validated on real open-sky acquisitions. The optical stimulator calibrations have broken any record in accuracy, with radiometric calibration with errors below 100 electrons and geometric calibration better than 0.3 pixels.
In the autonomous guidance pillar, we have developed novel algorithms based on convex programming. These algorithms have been deployed on in-flight representative hardware, and a simulation campaign with a processor-in-the-loop approach has been performed in this pillar as well. The results demonstrate that the guidance approach enables autonomous interplanetary transfers for self-driving spacecraft, targeting moving asteroids, planets and, within some accuracy limits, ballistic capture corridors. Computational performance remains efficient across all tested scenarios; the time required to calculate optimal trajectories on the Zynq UltraScale+ MPSoC hardware averaged less than five minutes.
Extensive HIL simulations have been carried out, simulating autonomous interplanetary transfers from Earth to different celestial bodies of the solar system. Almost 100 successful simulations were completed, for a total runtime of more than 430 hours of HIL simulation. This corresponds to about 40’000 simulated days of interplanetary transfers and thousands of duty cycles.
In the autonomous guidance pillar, we have developed novel algorithms based on convex programming. These algorithms have been deployed on in-flight representative hardware, and a simulation campaign with a processor-in-the-loop approach has been performed in this pillar as well. The results demonstrate that the guidance approach enables autonomous interplanetary transfers for self-driving spacecraft, targeting moving asteroids, planets and, within some accuracy limits, ballistic capture corridors. Computational performance remains efficient across all tested scenarios; the time required to calculate optimal trajectories on the Zynq UltraScale+ MPSoC hardware averaged less than five minutes.
Extensive HIL simulations have been carried out, simulating autonomous interplanetary transfers from Earth to different celestial bodies of the solar system. Almost 100 successful simulations were completed, for a total runtime of more than 430 hours of HIL simulation. This corresponds to about 40’000 simulated days of interplanetary transfers and thousands of duty cycles.