Enhancing Trajectory Efficiency in 3D UAV Swarm Confrontation Using MAPPO Algorithm with Adaptive Iterative Resolution

In the context of UAV game-theoretic adversarial training, where large parameter spaces, uncertain information, and sparse rewards pose challenges to convergence, this paper proposes a UAV adversarial method based on the Adaptive Iterative Resolution Multi-Agent Proximal Policy Optimization (AIR-MAPPO) algorithm. Starting from low-resolution scenarios, the training adaptively increases the scene resolution based on improvements in training performance. The strategy learned from low-resolution scenes is used as an experience pool to assist in training higher-resolution scenes, ultimately leading to a strategy that defeats the adversary in a normal-resolution scenario. Additionally, simulation experiments are conducted to validate the method, and the results show that the AIR-MAPPO algorithm can quickly find an adversarial strategy to defeat the enemy. Compared to the standard Multi-Agent Proximal Policy Optimization (MAPPO) algorithm, AIR-MAPPO finds a strategy to defeat the enemy more quickly, avoids local optima, significantly improves training efficiency, and results in higher and more stable reward values.