基于DMPs的双机器人协同搬运轨迹的学习与泛化  

Learning and Generalization of Dual-Robot Cooperative Handling Trajectory Based on Dynamic Movement Primitives

作　　者：曹学鹏[1] 王德硕 冯艳丽 樊豪 张弓 张新荣[1] 赵睿英[1] CAO Xuepeng;WANG Deshuo;FENG Yanli;FAN Hao;ZHANG Gong;ZHANG Xinrong;ZHAO Ruiying(Institute of Mechatronics and Hydraulics,Chang’an University,Xi’an 710064,Shaanxi,China;Xi’an Aerospace Times Precision Electromechanical Company Limited,Xi’an 710100,Shaanxi,China;Guangzhou Institute of Advanced Technology,Chinese Academy of Science,Guangzhou 511458,Guangdong,China)

机构地区：[1]长安大学机电液一体化研究所,陕西西安710064 [2]西安航天时代精密机电有限公司,陕西西安710100 [3]广州中国科学院先进技术研究所,广东广州511458

出　　处：《华南理工大学学报（自然科学版）》2023年第12期9-20,共12页Journal of South China University of Technology(Natural Science Edition)

基　　金：国家自然科学基金面上项目(62073092);陕西省重点研发计划项目(2021ZDLGY09-02)。

摘　　要：针对双机器人协作系统执行具有强协调约束关系的仿人任务时,存在轨迹学习复杂、协调约束分析欠缺的问题,提出了基于动态运动基元(DMPs)的双机器人协同搬运轨迹学习及泛化方法。首先,从双机器人协同搬运任务出发,分析了双机器人协调约束关系,建立了双机器人运动约束模型。然后,将机器人运动轨迹解耦为位置轨迹和姿态轨迹,采用四元数实现姿态轨迹的无奇异描述,分别建立位置轨迹和姿态轨迹的动态运动基元模型,结合双机器人运动约束模型与动态运动基元模型,兼顾各自的任务要求和相对位姿约束,进而获得双机器人的运动轨迹。接着,开展了双机器人协同搬运轨迹的仿真与实验,结果表明:采用双机器人协同搬运轨迹的学习与泛化方法,当改变起、终点状态时,双机器人定姿态协同搬运的起、终点位置误差分别为0.0292mm和0.1127mm,变姿态协同搬运的起、终点位置误差分别为0.0323mm和0.1131mm,终点的四元数姿态偏差为0.0014、0.0027、0.0018、0.0030,表明该协同搬运轨迹的学习与泛化方法具有较高的运动控制精度,即使起、终点任务参数改变,泛化轨迹仍可保证目标的可达性,验证了提出的双机器人协调运动轨迹控制方法的合理性和有效性。该方法可学习人体搬运过程并能够准确泛化出新的运动轨迹,实现了双机器人的协调运动,具有重要的工程应用价值。Aiming at the problems of complex trajectory learning and lack of coordination constraint analysis when a dual-robot collaborative system performs humanoid tasks with strong coordination constraints,this paper proposed a dual-robot cooperative handling trajectory learning and generalization method based on dynamic movement primitives(DMPs).Firstly,starting from the dual-robot cooperative handling task,the coordination constraints of the dual-robot were analyzed,and the motion constraint model of the dual-robot was established.Then,the robot motion trajectory was decoupled into position trajectory and orientation trajectory,and the quaternion was used to realize the non-singular description of the orientation trajectory.And the dynamic movement primitives model of position trajectory and orientation trajectory were established respectively.They were combined with the dual robot motion constraint model and DMPs model,and the dual-robot movement trajectory was obtained,taking into account their respective task requirements and relative pose constraints.Finally,the simulation and experiments of the cooperative handling trajectory of the two robots were carried out.The results show that:using the learning and generalization method of the dual-robot cooperative handling trajectory,when the starting and ending states are changed,the position errors of start point and end point of the dual-robot cooperative handling with the fixed orientation are 0.0292 mm and 0.1127 mm respectively;the position errors of start point and end point of variable orientation coordinated handling are 0.0323 mm and 0.1131 mm respectively;and the quaternion orientation errors of the end point are 0.0014,0.0027,0.0018,0.0030,indicating that the cooperative handling trajectory learning and generalization method has high motion control accuracy;even if the task parameters of the starting and ending are changed,the generalized trajectory can still ensure the accessibility of the target,which verified the scientificity and effectiveness of the propo

关 键 词：动态运动基元 双机器人协同 轨迹学习 协调约束 四元数 

分 类 号：TP242.2[自动化与计算机技术—检测技术与自动化装置]