基于对决深度Q网络的机器人自适应PID恒力跟踪研究  

Research on Robot Constant Force Tracking Based on an Adaptive PID Algorithm with Dueling Q network

作　　者：杜亮 梅雪川 DU Liang;MEI Xuechuan(School of Mechanical and Electrical Technology,Guangdong Industry Polytechnic University,Guangzhou Guangdong 510300,China;SINOMACH Intelligence Technology Co.,Ltd.,Guangzhou Guangdong 510700,China)

机构地区：[1]广东轻工职业技术大学机电技术学院,广东广州510300 [2]国机智能科技有限公司,广东广州510700

出　　处：《机床与液压》2024年第15期50-54,共5页Machine Tool & Hydraulics

基　　金：广州市科技计划项目(202002030243);国家市场监管重点实验室(智能机器人安全)开放课题(GQI-KFKT202304);广东省教育厅特色创新项目(2021KTSCX203)。

摘　　要：为确保机器人与环境接触时能保持稳定的接触力,基于对决深度Q网络设计一种自适应PID控制恒力跟踪算法。分析机器人与外界的接触过程,并构建基于PID算法的机器人力控制器;提出基于对决深度Q网络的自适应PID算法,以适应外界环境的变化,该算法利用对决深度Q网络自主学习、寻找最优的控制参数;最后,通过Coopeliasim与MATLAB软件平台展开机器人恒力跟踪实验。仿真结果表明:提出的基于对决深度Q网络的自适应PID算法能够获得较好的力跟踪效果,验证了算法的可行性;相比于深度Q网络算法,力误差绝对值的平均值减少了51.6%,且收敛速度得到提升,使机器人能够更好地跟踪外界环境。To ensure that a robot can maintain a stable contact force when it contacts the environment,an adaptive PID control for robot constant force tracking was designed based on a dueling deep Q-network.The contact process between a robot and the environment was analyzed,and a robot force controller based on a PID algorithm was constructed.The adaptive PID algorithm based on a dueling deep Q-network was proposed to adapt to changes in the external environment.In this algorithm,the dueling deep Q network was used to learn and find optimal control parameters.Finally,robot constant force tracking experiments were expanded on Coopeliasim and MATLAB software platforms.The simulation results show that the adaptive PID algorithm based on the dueling deep Q-network can achieve good force-tracking effects,verifying the algorithm′s feasibility;compared with a deep Q network algorithm,the average absolute value of the force error is reduced by 51.6%,and the convergence speed is improved,allowing the robot to track the external environment better.

关 键 词：机器人 恒力控制 自适应PID控制 对决深度Q网络 

分 类 号：TP242.2[自动化与计算机技术—检测技术与自动化装置]