Finding Guaranteed RL Control for Satellite Systems

Corvid, in cooperation with the University of North Carolina-Charlotte, will develop a novel reinforcement learning based control architecture that has guaranteed stability.  This architecture will help to create control schemes that can handle complicated inputs and maneuver vehicles, while ensuring a level of safety currently unavailable with contemporary reinforcement learning based control algorithms.  This will help lower risk and increase capability for various satellite operations, such as Rendezvous Proximity Operations (RPOs), docking operations, and Dynamic Space Operations (DSO).  Corvid and UNCC will develop and prove original theorems that demonstrate guaranteed stability of the novel architecture.  We will then validate the architecture on a set of simple, relevant simulated control systems.  The end result will be a proven, validated, guaranteed stable reinforcement learning based control architecture that is ready for experimental validation.