基于非线性偏最小二乘的操作臂逆解算法

操作臂逆运动学问题是机器人控制中的一项重要内容。目前使用较多的神经网络法大多为多输入多输出或者 多输入单输出方式，需要大量运算。非线性偏最小二乘法( NLPLS)建立的模型分为内部和外部模型，样本数据经外部模 型处理后才用于训练若干个单输入单输出的神经网络。对PUMA560操作臂的仿真试验表明，在相同隐层神经元数的情 况下，该算法比普通神经网络法具有更好的预测精度。这也表明，NLPLS只需较少的隐层神经元数就可以达到普通方法 的精度，从而减少运算量。

Inverse Kinematics Solution of Manipulator Based on Nonlinear PLS

The inverse kinematics of manipulator is a vital problem in robot control. Generally, most of the neural networks are applied by many researchers as numerical solutions with multi-input-multi-output or multi-input-sigle-output ones. and that means a large amount of computations. Models built by NLPLS were divided into outer models and inner models. The samples were used for the training process of several neural networks after they were processed by outer models. Simulations based on the PUMA 560 manipulator show that NLPLS has a better prediction accuracy than ordinary solutions when they have the same number of hidden units. The results also indicate that, by applying NLPLS, less computation is needed to meet the same accuracy requirements.