[3O3-GS-5] Agents: multi-agents

[3O3-GS-5-03] Autonomous Path Planning for UAV Formation Flight Control within Unknown Environment via Multi-agent Reinforcement Learning

〇Yuki Mori¹, Sachiyo Arai¹ (1. Chiba University)

Keywords:Multi-agent System, Formation Control, Obstacle Avoidance

In recent years, there has been a great demand for cooperative work by multiple autonomous mobile robots. This research deals with the task of maintaining the formation of multiple robots in an environment with obstacles. Existing research using deep reinforcement learning (DRL) deals only with the movement of the robot in charge of following in the formation (follower). However, if obstacles restrict the movement of the follower, it becomes temporarily difficult to maintain the formation. In this study, we propose a path planning method using multi-agent DRL for all robots in the formation, including the leading robot, which is responsible for determining the direction of the formation (leader). The performance of the proposed method is verified in an experimental environment, and it is confirmed that the autonomous action selection of each robot enabled formation maintenance and obstacle avoidance.