基于RBF神经网络的MOTOMAN-UPJ型机器人运动学逆解

利用D—H参数对MOTOMAN—UPJ型机器人建立坐标系，推导出正运动学公式。将由此得到的运动学正解作为训练样本，利用RBF神经网络的局部逼近的优点，将求解机器人运动学逆解转化为对神经网络的权值进行训练。实现了机器人从工作空间到关节空间的非线性映射。使用12输入，单输出的RBF网络，对6自由度的MOTOMAN—UPJ机器人进行了计算仿真，验证了该方法的可行性。与传统解析法相比，大大减少了求解运动方程的计算量。与BP神经网络相比，加快了收敛速度，解决了实时性差的问题。

A Method for Solving Inverse Kinematics of MOTOMAN-UPJ Manipulator Based on RBF Network

Appropriate coordinate of MOTOMAN - UPJ robot manipulator was established by using Denavit -Hartenberg method, and a forward kinematics equation was deduced. Because of its local approaching ability, the inverse kinematics problem of the manipulator can be transformed into the weight training problem by using some forward kinematics results as training data set. The mapping from joint -variable space to operation - variable space was realized. The RBF network of 12 inputs and 1 output was designed, and the simulation of a 6 redundant MOTOMAN - UPJ robot manipulator was done to validate the feasibility of this method. The computation was reduced greatly compared with the conventional analytical method, and the problem of real time was resolved by accelerating the convergence to the BP network.