Cooperative Spacecraft Formation Flying Based on Reinforcement Learning

This article introduces an approach for spacecraft formation flying on elliptic orbit, combining multi-agent system theory with reinforcement learning algorithms. The contributions include a game-based strategy for local neighbor-to-neighbor information interaction and a novel reinforcement learning algorithm for optimal spacecraft formation control. The proposed multi-agent dual heuristic programming (MADHP) algorithm ensures stability by considering local interactions and is successfully applied to achieve spacecraft formation flying on elliptic orbit. Numerical simulations confirm the effectiveness of the approach. This study demonstrates potential applications of reinforcement learning in autonomous space missions and cooperative multi-agent systems.