Autonomous Spacecraft Collision Avoidance with Multiple Space Debris Based on Reinforcement Learning

Collisions with space debris can be devastating to spacecraft. In this paper, a new algorithm for collision avoidance maneuver optimization of autonomous spacecraft is proposed. A collision avoidance controller that satisfies realtime trajectory optimization based on proximal policy optimization (PPO) is designed. The PPO guidance controllers can extract hazard information from motion states through empirical learning and map decisions directly to impulse maneuver commands, which can divert the spacecraft away from hazardous areas. Meanwhile, the collision probability and energy consumption are considered as the optimization objectives, then the reward functions is determined in the learning process. Simulation results demonstrate that the proposed method enables the spacecraft to autonomously avoid multiple space debris and meets the realtime requirement.