A Multi-agent Deep Reinforcement Learning Method for UAVs Cooperative Pursuit Problem

As an important form of intelligent warfare, UAV swarm is emerging. This paper designs a solution for UAV cooperative pursuit scenarios based on MADDPG. The clipping double Q network and policy delay update mechanism are proposed to solve the problems of overestimation of value function and wrong transmission in MADDPG algorithm. Due to the idea of centralized training and distributed execution of MADDPG algorithm and the architecture of constructing evaluation function for each agent, the method in this paper has good scalability and can be effectively applied to the environment of cooperative pursuit task.