Dynamics modeling of robotic manipulators using an artificial neural network

Abstract

Abstract Dynamics modeling is important for the design, analysis, simulation, and control of robotic and other computer‐controlled mechanical systems. The complete dynamic modeling of such systems involves the computationally intensive solution of a set of non‐linear, coupled differential equations. Artificial neural networks are well suited for this application due to their ability to represent complex functions and, potentially, to operate in real time. The application of an artificial neural network to dynamics modeling of robotic systems is investigated. The Cerebellar Model Arithmetic Computer (CMAC) is employed. A hybrid implementation of CMAC is proposed to allow use of the model for either simulation or control of robotic manipulators. The success of the simulated results and the accuracy of the generated outputs after a few training cycles demonstrate great promise for further development of the method and its implementation in control systems. © 1994 John Wiley & Sons, Inc.