Bi-criteria optimal control of redundant robot manipulators using LVI-based primal-dual neural network
Bi-criteria optimal control of redundant robot manipulators using LVI-based primal-dual neural network
复制标题
使用基于 LVI 的原对偶神经网络冗余机器人机械臂的双准则最优控制
DOI：
10.1002/oca.897
复制
发表时间：
2010-05-01
影响因子：
1.8
通讯作者：
Zhang, Yunong
中科院分区：
文献类型：
--
作者：
Cai, Binghuang;Zhang, Yunong
本文针对冗余机器人操作臂的最优运动控制提出了一种双准则加权方案。为了减少纯无穷范数速度最小化（INVM）方案的不连续现象，所提出的双准则冗余分解方案通过一个加权因子将最小动能方案和INVM方案相结合。关节物理极限，如关节极限和关节速度极限，也可同时纳入方案公式中。最优运动学控制方案最终可重新表述为一个二次规划（QP）问题。作为实时QP求解器，还开发了一种基于线性变分不等式（LVI）的原始 - 对偶神经网络（PDNN），它具有简单的分段线性结构，并能全局指数收敛到最优解。由于基于LVI的PDNN无需矩阵求逆，与对偶神经网络相比，它具有更高的计算效率。基于PUMA560操作臂进行的计算机仿真说明了这种用于冗余机器人的双准则神经最优运动控制方案的有效性和优势。版权所有（C）2009约翰威立父子有限公司
In this paper, a bi-criteria weighting scheme is proposed for the optimal motion control of redundant robot manipulators. To diminish the discontinuity phenomenon of pure infinity-norm velocity minimization (INVM) scheme, the proposed bi-criteria redundancy-resolution scheme combines the minimum kinetic energy scheme and the INVM scheme via a weighting factor. Joint physical limits such as joint limits and joint-velocity limits could also be incorporated simultaneously into the scheme formulation. The optimal kinematic control scheme can be reformulated finally as a quadratic programming (QP) problem. As the real-time QP solver, a primal-dual neural network (PDNN) based on linear variational inequalities (LVI) is developed as well with a simple piecewise-linear structure and global exponential convergence to optimal solutions. Since the LVI-based PDNN is matrix-inversion free, it has higher computational efficiency in comparison with dual neural networks. Computer simulations performed based on the PUMA560 manipulator illustrate the validity and advantages of such a bi-criteria neural optimal motion-control scheme for redundant robots. Copyright (C) 2009 John Wiley & Sons, Ltd.