Scholarlink

<p>This paper introduces a reinforcement learning-based control strategy for multi-redundant robotic manipulator systems, designed to address the challenge of collision avoidance while also maintaining distributed formation control. The strategy integrates both formation control and collision avoidance within an optimization problem, employing a potential energy penalty term. It proposes a meta-heuristic reinforcement learning algorithm that eliminates the need for a kinematic Jacobian matrix pseudo-inverse, incorporating a Q-learning framework, Particle Swarm Optimization, and the Beetle Antennal Olfaction algorithm. The meta-heuristic reinforcement learning algorithm has been enhanced with evolutionary and migratory actions to improve performance, and its computational complexity is linear relative to the number of joints of all robotic arms. Theoretical analysis confirms the strategy’s convergence and stability, and simulations with three 7-DOF manipulators substantiate its effectiveness.</p>

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

Collision-free distributed optimal formation control for multi-redundant manipulators based on reinforcement learning

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Xinyu Li,
  • Ruohan Mi,
  • Mengyang Wu,
  • Jinwei Yu

摘要

This paper introduces a reinforcement learning-based control strategy for multi-redundant robotic manipulator systems, designed to address the challenge of collision avoidance while also maintaining distributed formation control. The strategy integrates both formation control and collision avoidance within an optimization problem, employing a potential energy penalty term. It proposes a meta-heuristic reinforcement learning algorithm that eliminates the need for a kinematic Jacobian matrix pseudo-inverse, incorporating a Q-learning framework, Particle Swarm Optimization, and the Beetle Antennal Olfaction algorithm. The meta-heuristic reinforcement learning algorithm has been enhanced with evolutionary and migratory actions to improve performance, and its computational complexity is linear relative to the number of joints of all robotic arms. Theoretical analysis confirms the strategy’s convergence and stability, and simulations with three 7-DOF manipulators substantiate its effectiveness.