Enhanced Multi-Agent Proximal Policy Optimization for Multi-UAV Target Offensive-Defensive Decision

Autonomous collaborative decision-making is the key technology to achieve large-scale unmanned combat. Focus on the problem of multiple unmanned aerial vehicles' cooperative decision in target offensive and defensive combat, a multi-agent deep reinforcement learning (MADRL) based decision framework is proposed in this paper. Firstly, the simulation environment with a high-fidelity fixed-wing motion model is built. Secondly, to address the issue of high-dimension state space and credit assignment under a multi-agent environment, an enhanced multi-agent proximal policy optimization with mean-field counterfactual advantage (MAPPO - MFCOA) is proposed. Finally, the results of simulation experiments will verify the performance of the proposed approach.