Continuous Path Control of Manipulator Using an Improved CMAC

Abstract

The Cerebellar Model Arithmetic Controller (CMAC) is a means to achieve neural networks, and this has the advantages that the convergence time is short and learning speed per step is fast, compared with those of standard back-propagation neural networks. In the conventional design method of CMAC, a learning gain and quantizing intervals are determined by trial and error, and these values are constant. Therefore, it is very difficult to improve both the learning speed and the accuracy of CMAC at once through the conventional design method.
In this paper, a conventional design method of CMAC has been improved to overcome this problem. Furthermore, a continuous path control system of manipulator using this CMAC is presented. The effectiveness is verified by experimental results.