数据驱动的冗余机器人双层运动规划  

Data-driven double-level motion planning for redundant robots

作　　者：刘梅 尚明生[1] Mei LIU;Mingsheng SHANG(Chongqing Institute of Green and Intelligent Technology,Chinese Academy of Sciences,Chongqing 400714,China;University of Chinese Academy of Sciences,Beijing 101408,China)

机构地区：[1]中国科学院重庆绿色智能技术研究院,重庆400714 [2]中国科学院大学,北京101408

出　　处：《中国科学：信息科学》2024年第4期960-975,共16页Scientia Sinica(Informationis)

基　　金：国家自然科学基金(批准号:62372427,62306130)资助项目。

摘　　要：冗余机器人在智能制造、救灾抢险与太空探索等领域扮演着重要角色.然而,对结构信息不准确或未知的冗余机器人进行运动规划是一个棘手的问题.此外,若不考虑冗余机器人末端执行器的姿态约束,可能会因姿态的不可控性导致任务失败.为解决该问题,在考虑末端姿态保持和多层关节物理限制的基础上,本文针对模型未知的冗余机器人提出了一种基于数据驱动技术的双层运动规划方案.该方案在关节角加速度层求解的同时能实现最小化关节角速度范数的要求.进而,构建一个离散神经动力学求解器以在线求解该方案.理论分析、仿真和物理实验验证了所提数据驱动的冗余机器人双层运动规划方案的可行性和有效性.Redundant robots play an important role in intelligent manufacturing,rescue and relief work,and space exploration.However,motion planning of redundant robots with unknown structural information is a tricky problem.In addition,if the orientation of the end-effector of a redundant robot is not considered,a task may fail due to the change of orientation.To solve this problem,a data-driven double-level motion planning scheme is proposed,which is formulated at the joint acceleration level to minimize the norm of the joint velocity,maintaining the orientation of the end-effector and considering the multiple physical constraints of joints.Furthermore,a discrete neural dynamics solver is constructed to solve the proposed scheme online.Theoretical analysis,computer simulations,and experiments verify the feasibility and effectiveness of the proposed data-driven doublelevel motion planning scheme for redundant robots.

关 键 词：冗余机器人 姿态保持 数据驱动 神经动力学 非线性优化 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]