A Multi-agent Deep Reinforcement Learning Framework for UAV Swarm

Unmanned aerial vehicle (UAV) swarm plays more and more important role in modern warfare, they can cooperate, communicate and share information with each other to enhance their survivability and combat ability in modern warfare. However, UAV swarm faces dynamic battlefield situation, making them hard to learning optimal cooperation and confrontation policy. To address the issues, a framework of UAV swarm cooperation and confrontation based on multi-agent deep reinforcement learning (MADRL) is proposed, and it can greatly improve training efficiency and model adaptability.