摘要： 求解机械臂逆运动学问题可以采用神经网络来建立逆运动学模型，通过遗传算法或BP算法训练神经网络的权值从而得到问题的解，在求解精度和收敛速度上有待进一步改进。采用人工免疫原理对RBF网络训练数据集的泛化能力在线调整隐层结构，生成RBF网络隐层。当网络结构确定时，采用递推最小二乘法确定网络连接权值。由此对神经网络的网络结构和连接权进行自适应调整和学习。通过仿真可以看出，用免疫原理训练的神经网络收敛速度快，泛化能力强，可大幅提高机械臂逆运动学求解精度。

关键词: 机械臂, 免疫原理, RBF神经网络, 逆运动学求解

Abstract: Neural network is applied to solving inverse kinematics of manipulator, whose weight is got through genetic algorithm or back propagated algorithm. These methods need to be further improved in solving accuracy and convergence rate. Through the generalization ability of artificial immune principle to RBF network data set training, the structure of hidden layer is adjusted and generated. When the structure is determined, recursive least squares method is used to determine weight value of network output, so as to adjust and learn the structure and weight value of the neural network self-adaptively. The simulation results verify that the trained neural network by the immune principle has better generalization ability and fast convergence. It can solve accuracy of manipulator inverse kinematics.

Key words: manipulator, immune principle, RBF neural network, solution of inverse kinematics

中图分类号: 

• TP241