Spacecraft stealth through orbit-perturbing maneuvers using reinforcement learning

Spacecraft maneuvers are planned with operational objectives in mind, ranging from making up for atmospheric drag to collision avoidance. Though these areas have been researched in depth, performing maneuvers to avoid detection by sensors hasn’t been explored until recently. Reinforcement learning has been shown to be an effective method for optimizing a maneuver strategy for the purpose of avoiding detection by a sensor with a pre-defined search strategy. This work expands on that further by incorporating the opposed sensor into the learning process as well and results in an optimal strategy for both opponents with respect to one another. It was found that, with an average maneuver magnitude of 19m/s and an average of 3.12 days between maneuvers, a controlled spacecraft is able to successfully avoid being tracked by a single sensor for 31.2% of the observation windows.