Hierarchical Multi-Robot Pursuit with Deep Reinforcement Learning and Navigation Planning
Wenzhang Chen, Yuanheng Zhu · 2024
Robotic systems have been used to solve multiple important problems, including pursuit. However, traditional methods in game theory and controlling make it difficult to resolve pursuit problems with high dimensions, continuous spaces, and complex non-convex obstacles. Recently, many researchers have proved that deep reinforcement learning (DRL) has strong feature extraction and decision-making ability when solving decision problems in high-dimensional, continuous space. Thus, applying DRL methods to multi-robot pursuit problems is reasonable. Firstly, in this paper, a multi-robot pursuit simulation environment with complex obstacles in the Unity simulation engine is built. Moreover, a hierarchical method combined with global decision by DRL, local navigation through A* and Navigation Mesh, and collision avoidance by Reciprocal Velocity Obstacles (RVO) is proposed. Finally, the experiment results show the effectiveness of our methods in solving the multi-robot pursuit problem.