Multi-Pursuer Multi-Target Encirclement Strategy Based on Multi-Agent Deep Deterministic Policy Gradient

This paper proposes a multi-pursuer multi-target encirclement approach based on multi-agent deep deterministic policy gradient (MADDPG) algorithm, within the context of orbital game for satellite swarms. First, the multiconstraint impulsive orbital game model between the satellite swarm and non-cooperative targets is established using the CW equation and game theory. Then, through analyzing the orbital game process and integrating the Markov decision process (MDP), the MDP model between the multi-pursuer and multi-target encirclement game is developed. A corresponding reward function is designed and the training process of the network based on the MADDPG algorithm is examined for the orbital game mission. Finally, the MADDPG algorithm is applied to solve a typical multi-target orbital game problem, and comparisons with traditional numerical algorithms are performed, which demonstrates the effectiveness and feasibility in multi-target encirclement game for satellite swarms.