基于改进QPSO-NN的冗余机械臂逆运动学算法

针对某种冗余机械臂逆运动学求解的问题,提出了一种基于改进量子粒子群神经网络的求解算法。以冗余机械臂末端位姿为输入,经神经网络求得其逆解;针对神经网络输出结果误差较大的问题,把神经网络求初值加入初始化的粒子群中,通过基于Metropolis准则改进量子粒子群算法,避免了量子粒子群算法的早熟现象;以关节坐标经正向运动学求得的末端位姿和期望位姿的误差为适应度函数,对机械臂关节坐标迭代寻优。仿真结果表明该方法结合了神经网络算法的快速性和改进量子粒子群算法的精确性,满足求冗余机械臂逆运动学问题的速度和精度要求。

Inverse Kinematics of Redundant Manipulators Based on Improved QPSO-NN

A technique based on improved QPSO-NN is proposed to solve the inverse kinematics of redundant manipulators. The posture of manipulator’s end-effector is used as input of neural networks and the inverse kinematics results are got from neural networks. To minimize the error of neural networks’ output,a type of QPSO is selected. The results of neural network are added to initialized particle swarm and optimized through improved QPSO. QPSO is improved by Metropolis process,which could avoid premature convergence phenomenon. The orientation and position error of end effector calculated from joints’angle by forward kinematics is performed as fitness function. The simulation results show that the algorithm combines velocity of neural network and precision of improved QPSO,can satisfy the application in inverse kinematics of redundant manipulators.