A Multi-agent Reinforcement Learning Framework for Coordinated Multi-UAV Interception Strategies

The interception game between groups of unmanned aerial vehicles (UAVs) is crucial in the future intelligent warfare. In response to the collaborative interception gaming problem against aerial cluster attacks, a multi-agent deep reinforcement learning (DRL) framework based on the twin delayed deep deterministic policy gradient (TD3) method is proposed. The framework combines single-agent delayed policy gradient algorithms with a centralized evaluation and distributed execution algorithm architecture. In order to enhance the convergence of the algorithm, a generalized advantage function is designed. The simulation results show that the strategy enables UAVs to assign interception targets based on real-time battlefield conditions intelligently.