Scholarlink

To solve the obstacle avoidance problem of multi-autonomous mobile robot system, a distributed CBF-MATD3 algorithm is proposed based on Multi-Agent Twin-Delayed Deep Deterministic Policy Gradient (MATD3) algorithm combining control barrier function (CBFs). Compared with the environment of a single autonomous mobile robot, the environment faced by multi-autonomous mobile robots is more complex, which not only needs to deal with various obstacles, but also to maintain a safe distance from other autonomous mobile robots. The distributed CBF-MATD3 algorithm designs corresponding control barrier function constraints for each agent, which is easier to implement than designing centralized security constraints for the whole. At the same time, the distributed structure is scalable, and the security constraints of agents performing different tasks can be considered separately. Compared with the basic MATD3, the simulation results show that CBF-MATD3 can improve the training efficiency and safety, and the convergence of reward function is better.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-AMR Obstacle Avoidance Algorithm Based on Improved MATD3

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zongyu Li,
  • Yanxin Zhang,
  • Chenkun Yin,
  • Xiangbin Liu

摘要

To solve the obstacle avoidance problem of multi-autonomous mobile robot system, a distributed CBF-MATD3 algorithm is proposed based on Multi-Agent Twin-Delayed Deep Deterministic Policy Gradient (MATD3) algorithm combining control barrier function (CBFs). Compared with the environment of a single autonomous mobile robot, the environment faced by multi-autonomous mobile robots is more complex, which not only needs to deal with various obstacles, but also to maintain a safe distance from other autonomous mobile robots. The distributed CBF-MATD3 algorithm designs corresponding control barrier function constraints for each agent, which is easier to implement than designing centralized security constraints for the whole. At the same time, the distributed structure is scalable, and the security constraints of agents performing different tasks can be considered separately. Compared with the basic MATD3, the simulation results show that CBF-MATD3 can improve the training efficiency and safety, and the convergence of reward function is better.