基于多智能体强化学习的大规模无人机集群对抗

UAV swarms operation with low attack cost, high survival rate, and fine-grained flexible capabilities will be the important form in future war. Efficient and adaptive fine-grained task planning of UAV is very important in enhancing the drone swarm operation effectiveness. Due to dimensional disaster and explosive combination, multi-agent reinforcement learning can only be applicable to the sequential decision-making tasks in small-scale scenarios. The swarm mission planning in an adversarial environment is modeled as a Markov game, and the mean field theory is adopted to simplify the complex interaction between large-scale UAVs, as the interaction model between a single UAV and average impact of the multiple nearby UAVs. The multi-agent reinforcement learning algorithm based on the mean field theory is used to solve the adaptive task planning problem of the drone swarm in the large-scale confrontation simulation environment, and its high efficiency and flexibility are verified.