Reinforcement Learning-Based Method for Collaborative Target Assignment Against Heterogeneous UAV Swarms

To address the challenge posed by saturated attacks of drone swarms to air defense systems, and to achieve the winning goal of “using swarms to counter swarms”, a cooperative target assignment method based on proximal policy optimization (PPO) was proposed. The approach incorporated an attention mechanism to capture interaction features between intercepting and target clusters, enhancing the model’s situational awareness. A hierarchical masking mechanism was also introduced to handle variable-scale target clusters, dynamically screen available interceptors, and avoid fire overlap, thereby satisfying cooperative constraints. Experiments demonstrate that the method maintains good generalization and robustness in complex adversarial scenarios, offering a new solution for intelligent target assignment under dynamic threats.