Real time control of robot manipulator using a neural network based learning controller

A neurocontroller is presented for the tracking control of a SCARA robot. The structure of the controller consists of an inverse dynamics model of which the parameters are to be learnt in real time and a feedback servo to guarantee stability. By exploiting the a priori knowledge about the dynamics of the robot, a single layer linear network is obtained to model the inverse dynamics thereby reducing the training time. Real time learning of the synaptic weights which represent the parameters of the inverse dynamics of the robot can be completed in a few minutes. Experimental results demonstrated that the performance of the neurocontroller improved rapidly during learning. Accurate trajectory tracking is achieved within the first ten presentations of the training trajectory pattern