用分级神经网络学习机器人的动力学特点

Learning Robot Dynamic Characteristic with Classification Neural Network

给出了解决机器人控制问题一种神经网络方法。使用一个分级神经网络(NN)结构学习刚体机器人动力学特点。对于一般类别的机械手,使用前训练一系列的三层前馈网络模块,然后把这些基函数实时地用于第四层。使用线性控制原理,辅以非线性补偿控制机械手,使学得的机械手动力学知识创建一个在整个工程中高速控制机械手的控制器。模拟结果表明控制器的性能得到了大大提高。

This paper presents a neural network method to solve robot control problem. A classification neural network is used to learn rigid robot dynamic characteristic. To common manipulator training a serial of three-layer forward –feedback module before using, then these radical function is used in four-layer in real time. Based on principle of linear control and compensated with nonlinearity the gained manipulator dynamic knowledge establishes a controller that control manipulator in the whole process. Simulatio...