引用本文:孔颖,吴佳佳,雷景生,胡汤珑.基于有限时间对偶神经网络的冗余机械臂重复运动规划[J].控制理论与应用,2023,40(1):139~148.[点击复制]
KONG Ying,WU Jia-jia,LEI Jing-sheng,HU Tang-long.Finite-time dual neural network for solving repetitive motion planning of redundant manipulator[J].Control Theory and Technology,2023,40(1):139~148.[点击复制]
基于有限时间对偶神经网络的冗余机械臂重复运动规划
Finite-time dual neural network for solving repetitive motion planning of redundant manipulator
摘要点击 1505  全文点击 509  投稿时间：2021-12-13  修订日期：2022-01-22
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2022.11214
  2023,40(1):139-148
中文关键词  对偶神经网络  有限时间  冗余机械臂  关节物理限制  重复运动规划
英文关键词  dual neural network  finite-time  redundant manipulator  joints limitation  repetitive motion planning
基金项目  国家自然科学基金项目(61803338, 61972357), 浙江省重点研发计划(2019C03135), 浙江省自然科学基金项目(LZY22E050002)资助.
作者单位E-mail
孔颖* 浙江科技学院信息与电子工程学院 kongying-888@163.com 
吴佳佳 浙江科技学院信息与电子工程学院  
雷景生 浙江科技学院信息与电子工程学院  
胡汤珑 浙江科技学院信息与电子工程学院  
中文摘要
      在考虑关节物理极限的情况下, 将冗余机械臂的逆运动学解析问题抽象为带约束的重复运动规划(RMP) 方案. 针对速度层的带约束RMP方案, 本文提出了一种新型的递归神经网络, 即有限时间对偶神经网络(FTDNN), 用以求解该类带约束RMP方案. 相比于传统的递归神经网络, 该FTDNN模型具有有限时间收敛特性, 不仅能够改进 收敛的速度, 并且能够获得较高的收敛精度. 通过李雅普诺夫稳定性定理验证了FTDNN模型的渐近稳定性, 并进一 步计算出FTDNN模型求解带约束RMP方案最优解的时间上界. 基于冗余机械臂PA10的计算仿真结果验证了FTDNN模型求解带约束RMP方案的有效性和可行性. 最后在Dobot Magician实物机械臂上的实验结果表明本文提出的 有限时间对偶神经网络方法可以有效实现机械臂各关节角的重复运动.
英文摘要
      Resolution of the inverse-kinematics redundancy problem for redundant manipulators can be transformed into a repetitive motion planning (RMP) problem by incorporating joint-angle limits and joint-velocity limits. In this paper, a new recurrent neural network, finite-time dual neural network (FTDNN), is proposed for solving this type of constrained RMP problem in velocity layer. Compared with the traditional recurrent neural networks, the FTDNN model has finite-time convergence characteristics, which not only improves the convergent speed but also achieves higher convergent accuracy. Furthermore, asymptotic stability of the FTDNN model is verified by the Lyapunov’s stability theorem, and the time upper bound for the optimal solution of the FTDNN model with constrained RMP problem is presented. Computational simulation results based on the redundant manipulator PA10 demonstrate the validity and feasibility of the FTDNN model for solving the constrained RMP problem. Finally, experimental results on the Dobot Magician manipulator show that the proposed FTDNN model can effectively realize the repetitive kinematics of each joint angle.