MADDPG based Assembly Approach of Cooperative Robotic Arms
首发时间:2020-06-22
Abstract:Compared with the single robitic arm, a two-arm cooperative robot has a more complex problem of motion interference. In cooperative assembly problems, traditional control methods need much calculation to establish a mathematical model for motion planning, which can not meet the needs of modern industry. Multi-Agent Reinforcement Learning(MARL) provides a new way to solve the assembly problem. This paper proposes a assembly method of collaborative robotic arm based on MADDPG. A multi-agent system simulation environment is founded based on MuJoCo physical engine and then assembly tasks in the environment are solved with MADDPG model. Finally, the approach is proved to be effective by experiment.
keywords: multi-agent reinforcement learning robotic arm control multi-agent system
点击查看论文中文信息
基于MADDPG的协同机械臂装配方法
摘要:协作机械臂相比单机械臂存在更为复杂的运动干涉问题,在协作装配问题上,传统控制方法需要大量计算建立数学模型用于运动规划,无法满足灵活多变的现代工业需求。多智能体强化学习为解决协同机械臂的装配问题提供了新的解决思路,本文提出了一种基于MADDPG的协同机械臂装配方法,通过MuJoCo物理引擎建立了面向多智能体系统的协同机械臂装配仿真环境,并利用MADDPG模型实现了仿真环境下机械臂的孔轴装配任务。最后通过实验以验证该方法的有效性,为人工智能技术在制造业领域的应用提出了一种新的解决方案。
基金:
引用
No.****
同行评议
勘误表
基于MADDPG的协同机械臂装配方法
评论
全部评论0/1000