基于 RN-MAPPO 的航天器集群协同围捕方法

To address the challenges of incomplete information，high training difficulty，and low coordination efficiency in spacecraft swarm’s cooperative round-up of non-cooperative targets，an improved multi-agent reinforcement learning-based cooperative round-up approach for spacecraft swarm is proposed. Firstly，considering the characteristics of the cooperative round-up task，the dynamic model，and the constraint conditions，a cooperative round-up model for spacecraft swarm is established. Secondly，to overcome the difficulty in converging complex policy networks，residual learning is introduced to optimize the policy network，and the residual network multi-agent proximal policy optimization （RN-MAPPO）algorithm is proposed. Furthermore，considering the unknown maneuvering capability and strategy of the target，a cooperative round-up approach for spacecraft swarm is designed based on the RN-MAPPO algorithm to achieve rapid strategy generation for multi-scenario cooperative round-up. Simulation results demonstrate that the proposed method exhibits favorable convergence and training efficiency，effectively reduces round-up time and fuel consumption，and ensures a high round-up success rate while fully leveraging the synergy of the swarm.