现代制造工程 ›› 2026, Vol. 553 ›› Issue (10): 73-81.doi: 10.16731/j.cnki.1671-3133.2026.10.009

• 机器人技术 • 上一篇    下一篇

融合LPI-DK-MPC的多自由度机械臂轨迹控制*

吴文杰1,2, 何勋3   

  1. 1 商丘职业技术学院机电工程学院,商丘 476000;
    2 菲律宾凯迪雷拉大学,碧瑶 2600;
    3 河南农业大学机电工程学院,郑州 450000
  • 收稿日期:2026-01-22 出版日期:2026-10-18 发布日期:2026-11-05
  • 作者简介:吴文杰,博士,副教授,主要研究方向为机器人智能控制、人工智能。E-mail:www2el@126.com
  • 基金资助:
    *河南省普通高校就业创业课题项目(JYB2025085);商丘市级科技攻关计划项目(2024039)

Trajectory tracking control of robotic manipulators by integrating physical-informed Koopman operators with model predictive control

Wu Wenjie1,2, He Xun3   

  1. 1 School of Mechanical and Electrical Engineering,Shangqiu Vocational and Technical College,Shangqiu 476000,China;
    2 University of the Cordilleras,Baguio 2600,Philippines;
    3 College of Mechanical and Electrical Engineering,Henan Agricultural University,Zhengzhou 450000,China
  • Received:2026-01-22 Online:2026-10-18 Published:2026-11-05

摘要: 针对多自由度机械臂在强非线性动力学、模型不确定性及严格物理约束下难以实现高精度轨迹控制的难题,提出了一种融合拉格朗日物理信息深度Koopman与模型预测控制(Lagrangian Physics Informed-Deep Koopman-Model Predictive Control,LPI-DK-MPC)的控制策略。首先,通过将欧拉-拉格朗日力学方程的残差嵌入到深度神经网络的训练损失函数中,构建了具备物理一致性的全局线性化Koopman预测模型,在保证预测精度的同时严格满足了系统的能量守恒约束;其次,引入Koopman障碍函数以处理非线性约束,并以此设计了线性模型预测控制器来实现高效的闭环控制;最后,以UR5机械臂为研究对象开展了仿真与物理实验平台验证。仿真结果表明,在LPI-DK-MPC控制策略作用下,机械臂末端在X、Y和Z方向上的最大跟踪误差分别为8.04、7.41和6.55 mm,控制精度较传统MPC与反馈MPC分别平均提升了30.24 %和38.95 %;实物实验结果进一步表明,X和Y方向的最大跟踪误差分别为7.62和8.81 mm。研究结果充分验证了所提LPI-DK-MPC控制策略能够有效实现多自由度机械臂的高精度、安全轨迹控制。

关键词: 机械臂控制, Koopman算子, 模型预测控制, 物理信息约束

Abstract: To address the challenge of achieving high-precision trajectory tracking for multi-degree-of-freedom robotic arms under strong nonlinear dynamics,model uncertainties, and strict physical constraints, it proposes a novel control framework that integrates Lagrangian Physics Informed-Deep Koopman-Model Predictive Control (LPI-DK-MPC). Firstly, by embedding the Euler-Lagrange equations of motion into the training loss function of a deep neural network, a globally linearized Koopman prediction model with physical consistency is constructed, which ensures the conservation of energy while maintaining prediction accuracy. Secondly, a Koopman control barrier function is introduced to handle nonlinear state constraints,and a linear model predictive controller is designed based on this to achieve efficient closed-loop control. Finally, the UR5 robotic arm is used as the research object for simulation and physical experiment platform verification. Simulation results show that under the LPI-DK-MPC strategy,the maximum tracking errors of the robotic arm′s end effector in the X,Y and Z directions are 8.04,7.41 and 6.55 mm, respectively, with control accuracy improved by an average of 30.24 % and 38.95 % compared to traditional MPC and feedback MPC. Physical experiment results further demonstrate that the maximum tracking errors in the X and Y directions are 7.62 and 8.81 mm, respectively. The research results fully validate that the proposed LPI-DK-MPC control strategy can effectively achieve high-precision and safe trajectory control for multi-degree-of-freedom robotic arms.

Key words: manipulator control, Koopman operator, Model Predictive Control (MPC), physical information constraints

中图分类号: 

版权所有 © 《现代制造工程》编辑部 
地址:北京市东城区东四块玉南街28号 邮编:100061 电话:010-67126028 电子信箱:2645173083@qq.com
本系统由北京玛格泰克科技发展有限公司设计开发 技术支持:support@magtech.com.cn