机器人逆运动学的微分进化与粒子群优化BP神经网络求解  

Solving robot inverse kinematics based on differential evolution and particle swarm optimization BP neural network

作　　者：陈桂[1] 陈耀忠 林健[1] 温秀兰[1] 

机构地区：[1]南京工程学院自动化学院,江苏南京211167 [2]北方信息控制集团有限公司,江苏南京211153

出　　处：《南京理工大学学报》2014年第6期763-768,共6页Journal of Nanjing University of Science and Technology

基　　金：江苏省自然科学基金(BK2010479);南京工程学院校级科研基金(YKJ201319)

摘　　要：针对采用传统反向传播(BP)神经网络算法进行逆运动学求解收敛速度慢的问题,提出将微分进化(DE)与粒子群优化(PSO)算法相结合,对用于机器人逆运动学求解的BP神经网络进行优化。基于机器人正解映射建立优化算法的目标函数,在PSO过程中,引入DE操作优化粒子进化方向,并将此混合算法用于BP神经网络权值与阈值的优化。对KUKA机器人进行仿真实验,结果表明:采用该文方法对机器人逆运动学问题的求解精度高,求得的关节角度误差小于0.1°;逆运动学求解结果所对应位姿矩阵的位置误差在0.1 mm数量级,具有较好的泛化能力。该文方法满足机器人位置和姿态方面的精度要求。Aiming at the problem of slow convergence speed of traditional back propagation （BP） neural network algorithms, differential evolution （DE）and particle swarm optimization （PSO）are combined to optimize BP neural network for robot inverse kinematics. An objective function of the optimization algorithm is formulated based on the mapping of robot forward kinematics. DE operation is employed to optimize particle evolution direction in PSO, and the weights and thresholds of the BP neural network are optimized. A simulation experiment is proposed for a KUKA robot, and the result shows that：the solution accuracy of robot inverse kinematics of the algorithm proposed here is high, and the joint angle error is below 0.1 ° ; the position error between the initial pose matrix of the robot and that solved by the algorithm proposed here is of the order of magnitude of 0.1 mm, and has good generalization ability. The algorithm proposed here satisfies the accuracy requirements of robot locations and postures.

关 键 词：微分进化 粒子群优化 反向传播神经网络 机器人 逆运动学 收敛速度 权值 阈值 关节角度误差 位置误差 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]