Reinforcement Learning Based Spacecraft Autonomous Evasive Maneuvers Method Against Multi-interceptors

摘要

This paper proposed an autonomous intelligent decision-making method, which can be used to evade the attack of multi-interceptors. A self-learning model based on MADDPG algorithm is established with four engines of a spacecraft as multi-agent system. The model takes the relative distance and the total maneuvering time as variables to design the evaluation function. A simulation environment is constructed with four interceptors intercepting a spacecraft at the same time. The autonomous evasion impulse maneuvers of the spacecraft are realized by training. Compared with the random evasion maneuvers method, this autonomous maneuvers method improves the success probability of escape by nearly 30%. This research provides a valuable theoretical method for modern space operations.