关键词: 规避机动, 轨道威胁, 动作规划, 深度强化学习

Abstract: Spacecrafts face with multiple complex constraints during avoidance maneuvers. There are several problems in the traditional motion planning methods based on numerical optimization when processing corresponding models and constraints, such as the sensitive initial value and long calculation time, which makes it difficult to deal with close range orbital threats in time. To address this problem, a multi constrained avoidance motion planning method based on deep reinforcement learning (DRL) is proposed in this paper. First, the spacecraft six degree of freedom nonlinear dynamical model and related constraints for attitude orbit maneuvers are established. Then, the avoidance motion planning method based on twin delayed deep deterministic policy gradient (TD3) is proposed, and the multi constrained avoidance maneuvering actions can be online generated via the neural networks trained by TD3. Finally, the normative DRL training environment matched with the proposed planning method is constructed to ensure the effective interactions between agents and environments. Simulation results show that the proposed method can rapidly generate avoidance actions in real time when the expected rendezvous time is only in tens of seconds, and the planning period is less than 9 ms, which is much lower than the Gauss pseudo spectral method as a comparison item.

Key words: avoidance maneuver, orbital threat, motion planning, deep reinforcement learning