改进代数算法神经网络在机器人逆运动学中的应用

Neural network based on modified algorithm applying to inverse kinematics of robot manipulator

用多层前向INI神经网络建立机器人逆运动学模型。采用一种改进代数算法来学习神经网络待求权和自由权,该算法选择很广一类的隐层神经元函数训练网络,将复杂的非线性优化问题转化为简单的代数方程组求解问题,求解速度快;在网络训练之前就可以根据给定的问题确定隐层神经元个数,可以方便地求得全局最优点,实现样本空间的精确映射,不存在局部极小、收敛速度慢等问题。提出的求解机器人逆运动学新算法可以得到高精度的解,有仿真结果为证。

The multi-layer forward INI neural networks are used to establish the inverse kinematics models for robot manipulator.An improved algebra algorithm is presented to update the free weights and target weights.The algorithm transforms the complicated nonlinear optimization problem into linear algebraic equations with a wide variety of functions of hidden neurons and results can be got fast.Before training the neural networks,the number of hidden neurons can be got according to the given problem.The algorithm can get the global minimum easily and the precision mapping in the sample space,and there are no problems such as local minima and slow rate of convergence.The simulations show that the new algorithm to inverse kinematics problem for robot manipulator can get high precision results.