结合深度强化学习与变形五次多项式的搬运机器人关节轨迹规划  

Joint Trajectory Planning of Transfer Robots Combining Deep Reinforcement Learning and Deformed Quintic Polynomials

作　　者：李英杰 陈乃建[1] 尹逊瑞 郑家坤 张山林 李映君[1] LI Yingjie;CHEN Naijian;YIN Xunrui;ZHENG Jiakun;ZHANG Shanin;LI Yingjun(School of Mechanical Engineering,University of Jinan,Jinan 250022,Shandong,China)

机构地区：[1]济南大学机械工程学院,山东济南250022

出　　处：《济南大学学报（自然科学版）》2024年第2期234-241,共8页Journal of University of Jinan(Science and Technology)

基　　金：国家自然科学基金项目(51875250);山东省自然科学基金项目(ZR2022ME177);济南市“新高校20条”扶持项目(202228116);德州市现代产业领军人才工程项目。

摘　　要：针对在生产节拍固定的情况下搬运机器人各关节的运动轨迹及时间分配问题,构建基于深度强化学习与变形五次多项式相结合的搬运机器人关节轨迹时间分配模型;设计以节拍要求、速度约束和加速度约束为目标的奖励函数,搭建神经网络,利用MATLAB/Simulink软件,得到满足生产节拍和运动学约束的时间序列,并利用单臂四自由度搬运机器人仿真实验,验证所构建模型的可行性和有效性。结果表明:单臂四自由度搬运机器人各关节的运行时间均为5.89 s,其中平移关节1、 3的最大速度分别为2 597.84、 1 697.97 mm/s,最大加速度分别为19 532.11、 31 302.61 mm/s^(2);旋转关节2、 4的速度大小相等且方向相反,最大角速度均为137.53 (°)/s,最大角加速度均为1 180.51 (°)/s^(2),均未超过运动学约束;所构建的模型可以解决在指定生产节拍下搬运机器人关节轨迹时间分配问题,实现搬运机器人搬运过程中各关节的运动均衡,改善搬运机器人运行的稳定性和有效性。Aiming at motion trajectories and time allocation of all joints of transfer robots under the condition of fixed production cycle time,a joint trajectory time allocation model of transfer robots based on deep reinforcement learning combining with deformed quintic polynomials was constructed.A reward function with the target of cycle time require-ments,speed constraints,and acceleration constraints was designed,a neural network was built,and MATLAB/Simulink software was used to obtain time series meeting with the production cycle time and kinematic constraints.A single-arm four-degree-of-freedom transfer robot simulation experiment was used to verify feasibility and effectiveness of the constructed model.The results show that the running time of each joint of the single-arm four-degree-of-freedom transfer robot is 5.89 s,in which the maximum velocity of translation joint 1 and translation joint 3 are respectively 2597.84 mm/s and 1697.97 mm/s,and the maximum acceleration are respectively 19532.11 mm/s^(2) and 31302.61 mm/s^(2) respectively.The velocities of rotary joint 2 and rotary joint 4 are equal in magnitude and opposite in direction.Both of the maximum angular velocities are 137.53(°)/s,and both of the maximum angular accelerations are 1180.51(°)/s^(2).All of the above do not exceed kinematic constraints.The constructed model can solve the time allocation problem of transfer robot joint trajectory under specified production cycle time,achieve motion balance of all joints during transfer process of transfer robots,and improve stability and effectiveness of transfer robot operation.

关 键 词：机器人技术 关节轨迹规划 深度强化学习 搬运机器人 生产节拍 

分 类 号：TP241[自动化与计算机技术—检测技术与自动化装置] TP38[自动化与计算机技术—控制科学与工程]