Pursuit-Evasion Games for Multi-agent Based on Reinforcement Learning with Obstacles

Abstract

Considering the problem of external interference and obstacle avoidance in multi-agent pursuit-evasion games, the deep deterministic policy gradient algorithm is used to train agents in continuous space. Obstacle and collision avoidance are realized by designing detailed reward function. Interference data are added to the original observations, and adversarial learning algorithm is used to eliminate the influence of interference and other agents. The evaluation function based on heading angle and relative distance is used to evalue evader’s escape strategy, which improves the robustness of the proposed algorithm. Simulation experiments are designed to verify the effectiveness of the algorithm.