Autonomous Rendezvous Guidance via Deep Reinforcement Learning

This paper aims developing a new guidance algorithm for autonomous rendezvous with small continuous thrust. In particular, the goal is to create an algorithm that can meet the requirements of real-time and trajectory optimization at the same time. To Achieve the goal, this paper designed a new guidance controller based on deep reinforcement learning (DRL). The DRL guidance controller can map the motion state to the acceleration command directly by learning from experience, so that it can meet the real-time requirement. At the same time, by setting an appropriate reward function during learning, the terminal constraints and fuel optimization can be realized. Simulation results show that the algorithm is feasible and robust even though there is measurement error or model error. Compared with traditional PD algorithm, time and fuel consumption are reduced by more than 30%.