• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2026, Vol. 62 ›› Issue (17): 293-308.doi: 10.3901/JME.260821

• 特邀专栏:人形机器人与具身智能 • 上一篇    

扫码分享

基于全状态跟踪的机械臂空间—时间尺度缩放控制方法

杨鹏辉1,2, 张哲硕1, 戴家文1,2, 李刘杰3, 张邦基1, 凌亮4   

  1. 1. 浙大城市学院信息与电气工程学院 杭州 310015;
    2. 湖南大学机械与运载工程学院 长沙 410082;
    3. 杭州智元研究院有限公司 杭州 310000;
    4. 西南交通大学轨道交通运载系统全国重点实验室 成都 610031
  • 收稿日期:2026-01-30 修回日期:2026-06-30 发布日期:2026-11-20
  • 作者简介:杨鹏辉,男,2002年出生。主要研究方向为机器人示教学习,自适应鲁棒约束跟随控制。E-mail:phyang@hnu.edu.cn;张哲硕(通信作者),男,1989年出生,博士,副教授,硕士生导师。主要研究方向为矢量旋翼无人机、自适应鲁棒控制。E-mail:zhangzs@hzcu.edu.cn;戴家文,男,1997年出生,博士研究生。主要研究方向为无人机控制。E-mail:jiawendai@hnu.edu.cn;李刘杰,男,1984年出生,高级工程师。主要研究方向为智能弹药和分队体系总体设计。E-mail:303493297@qq.com;张邦基,男,1967年出生,博士,教授,博士生导师。主要研究方向为智能汽车和非线性系统振动控制。E-mail:zhangbj@hzcu.edu.cn;凌亮,男,1986年出生,博士,研究员,博士生导师。主要研究方向为车辆-轨道相互作用与行车安全控制。E-mail:liangling@swjtu.edu.cn
  • 基金资助:
    江西省2026年高新领域重点研发计划(20261BCE310030)和2025年度高层次留学回国人员在杭创新创业资助项目(抗强风全向检测矢量旋翼无人机项目)。

A Spatio-temporal Scaling Control Method for Robot Manipulators Based on Full-state Tracking

YANG Penghui1,2, ZHANG Zheshuo1, DAI Jiawen1,2, LI Liujie3, ZHANG Bangji1, LING Liang4   

  1. 1. School of Information and Electrical Engineering, Hangzhou City University, Hangzhou 310015;
    2. College of Mechanical and Vehicle Engineering, Hunan University, Changsha 410082;
    3. Hangzhou Zhiyuan Research Institute Co., Ltd., Hangzhou 310000;
    4. State Key Laboratory of Rail Transit Vehicle System, Southwest Jiaotong University, Chengdu 610031
  • Received:2026-01-30 Revised:2026-06-30 Published:2026-11-20

摘要: 在机械臂轨迹学习与泛化应用中,传统仅基于位置约束的轨迹跟踪方法难以适应起始条件变化及运动节奏调整的需求,且在位置存在偏差时难以有效的间接跟踪速度与加速度。其根本原因在于,现有方法在跟踪过程中仅对位置进行约束而忽略了速度与加速度等高阶状态,导致在空间与时间尺度缩放或起止条件变化时,难以保持示教轨迹所蕴含的运动特征。针对上述问题,提出一种基于全状态(位置、速度和加速度)跟踪的控制方法。首先,基于人类操作员示教轨迹数据,采用改进的动态运动基元方法,构建统一描述位置、速度与加速度演化关系的轨迹生成模型,并在空间与时间尺度灵活缩放的同时保留示教轨迹的运动特征;随后,以缩放轨迹的全状态为控制目标,结合作业机械臂动力学模型,设计一种全状态约束跟随控制器,实现机械臂在空间、时间尺度缩放及起止条件变化情况下对缩放轨迹的稳定跟踪控制,并基于李雅普诺夫方法严格证明闭环系统的稳定性;仿真与实验结果表明,该方法能够在保持轨迹运动特征与示教轨迹形状一致性的前提下,实现不同空间尺度、时间尺度及起止条件下的有效轨迹泛化,并对轨迹全状态实现高精度跟踪。结果进一步表明,相比传统轨迹跟踪控制方法,所设计的全状态跟踪策略更适用于轨迹学习与泛化应用。

关键词: 示教学习, 动态运动基元, 空间—时间尺度缩放轨迹, 约束跟随控制, 全状态跟踪

Abstract: In robot manipulator trajectory learning and generalization, conventional trajectory tracking methods that rely solely on position constraints struggle to accommodate variations in initial conditions and motion timing, and often fail to ensure accurate tracking of velocity and acceleration when position deviations occur. The fundamental reason lies in neglecting higher-order states during tracking, which prevents the preservation of intrinsic motion features of demonstrated trajectories under spatial and temporal scaling or changes in start–end conditions. To address this issue, a full-state tracking control method incorporating position, velocity, and acceleration is proposed. First, based on demonstration data collected from a human operator, a modified dynamic movement primitive framework is employed to construct a unified trajectory generation model that describes the coupled evolution of position, velocity, and acceleration, while preserving motion features under flexible spatial and temporal scaling. Subsequently, taking the full state of the scaled trajectory as the control objective, a full-state constraint-following controller is designed in conjunction with the manipulator dynamics to achieve stable trajectory tracking under spatial and temporal scaling as well as varying initial and terminal conditions. The stability of the closed-loop system is rigorously proven using Lyapunov theory. Simulation and experimental results demonstrate that the proposed method enables effective trajectory generalization across different spatial scales, temporal scales, and boundary conditions, while achieving high-precision tracking of the full trajectory state. The results further indicate that, compared with conventional trajectory tracking approaches, the proposed full-state tracking strategy is more suitable for trajectory learning and generalization tasks.

Key words: learn from demonstration, dynamic movement primitives, spatio-temporal scaling trajectory, constraint-following control, full-state tracking

中图分类号: