Daniel Huterer Prats, University of Colorado Boulder; Hanspeter Schaub, University of Colorado Boulder; Chris Wheeler, Interactive Aptitude
Keywords: Space Situational Awareness, Space Domain Awareness, Reinforcement Learning, Autonomy, Scheduling, Space-to-Space Surveillance
Abstract:
The increasing number of resident space objects (RSO) in low Earth orbit poses significant challenges for autonomous Space Situational Awareness (SSA). Unlike Earth observation, space-based SSA requires agile imaging of fast-moving targets under stringent constraints on power, line-of-sight, and illumination. This work researches having a satellite taking images of space objects with known trajectories. The paper formulates the space-to-space RSO inspection problem as a partially observable Markov decision process and trains a reinforcement learning (RL) agent for simultaneous dynamic target selection and onboard resource management. Using the BSK-RL environment and the Basilisk high fidelity spacecraft simulator, an actor-critic RL agent learns to autonomously image RSOs while maximizing coverage and adhering to subsystem limitations. Results show that the learned policy generalizes across orbital configurations, exploits eclipse periods for proactive downlinking, maintains energy margins quasi-autonomously, and achieves more timely and useful image delivery than a myopic heuristic. These findings support the potential of RL-enabled autonomy for future scalable SSA missions.
Date of Conference: September 16-19, 2025
Track: Machine Learning for SDA Applications