Afrah Ghedira, University of Colorado Boulder; Hanspeter Schaub, University of Colorado Boulder
https://doi.org/10.64861/UUUA7655
Keywords: Autonomous spacecraft inspection, Spinning resident space object, Impulsive guidance; Proximity operations, Reinforcement learning, Semi-Markov decision process
Abstract:
Space object inspection is a critical capability for rendezvous and proximity operations, enabling applications such as damage assessment, docking port localization, and space asset characterization. However, inspection is inherently constrained by factors including illumination conditions, viewing geometry, relative distance, and collision avoidance requirements. These challenges are further compounded when the target is tumbling, as the accessibility of inspection regions depends not only on orbital geometry but also on the target’s attitude dynamics.
This paper presents a deep reinforcement learning approach for the autonomous inspection of spinning resident space objects, considering both general tumbling motion and stable rotation about the orbit-normal axis with constant spin rate. The problem is formulated as a semi-Markov decision process for sequential decision making. By leveraging natural orbital dynamics, the agent executes impulsive maneuvers followed by variable-duration coasting phases to efficiently complete the constrained inspection task under time and fuel budgets.
Unlike conventional planning methods that optimize a fixed observation sequence offline, the proposed framework continuously adapts its decisions online based on the observed environment. The learned policies achieve safe and efficient inspection across a wide range of Low Earth Orbit scenarios with varying target spin rates and lighting conditions.
To assess performance, reinforcement learning policies spanning different time-versus-fuel optimization objectives are compared with baseline heuristic strategies, demonstrating the benefits of adaptive decision making for autonomous on-orbit inspection.
Date of Conference: September 15-18, 2026
Track: Conjunction/RPO