Development of Computational Intelligence-based Control System using Backpropagation Neural Network for Wheeled Robot

Abstract

This study aims to develop an optimal autonomous control system for a three-wheeled robot with two motors. The focus of this research is a neural network direct inverse controller system that is trained using backpropagation algorithm. Autonomous control system training is carried out by using the real data of manually controlled wheeled robot. This study analyzed the use of two Backpropagation learning algorithms namely Levenberg Marquardt Backpropagation and Bayesian Regularization Backpropagation, and compares 3 different controller network configurations, i.e., 13–10-2, 13– 20-2 and 13–26-2. The simulation results revealed that the best control system network architecture is 13–10-2 which was trained using Bayesian Regularization Backpropagation algorithm. This result serves as early evidence that a neural network-based control system can be used for autonomous wheeled robots.