基于改进型BP神经网络多关节机器人逆向运动学求解

Solving Converse Kinematics Problems for Multi-Joint Robot Based on Modified BP Neural Network

提出采用一种3层改进型快速BP神经网络(Modified fast BP neural network,MFBPNN)求解一个5自由度多关节机器人逆向运动学问题。使用正向运动学计算获得的样本向量进行离线学习,然后充分利用人工神经网络的泛化特性,实现了机器人末端作用器位姿到各个关节转角变量之间的非线性映射。仿真结果表明,采用MFBPNN算法以后,绝对误差不超过0．005°,计算精度和处理速度能够满足机器人实时控制的要求,并且可以应用于机器人路径规划控制场合。

The three-layer modified fast BP neural network （MFBPNN） is used to solve the converse kinematics problems of five-freedom multi-joint robot. The sample vectors of forward kinematics are used to train the weights of MFBPNN off-line. Then the generalization characteristic of artificial neural network （ANN） is fully utilized to accomplish the nonlinear map of the end-effector position and gesture to all joint rotating angles. Simulation results demonstrate that the calculation precision and the velocity satisfy the demands of the real-time control. The absolute angle error is less than 0. 005°, and the method can be applied to the situation of the robot route plan.