Intelligent Control of Mobile Robots with ANN Assisted Improved Q-learning: IQ-CRL Algorithm

Abstract

This paper presents a novel algorithm, IQ-CRL (Improved Q-learning using Classification and Regression with ANN), which is a control architecture that uses the existing Q-learning algorithm and integrates it with Artificial Neural Networks (ANNs). The proposed algorithm addresses the limitations of traditional Q-learning by incorporating an adaptive mechanism that leverages ANN to find an optimal control policy that is more accurate and at lower computational expense. IQ-CRL is designed to be used efficiently with dynamical systems such as mobile robots. The objective of this study is to evaluate the performance of IQ-CRL in comparison to classical Q-learning and PID controller. The results show that the IQ-CRL algorithm outperforms both the classical Q-learning and PID controller in terms of learning efficiency, control accuracy and computing complexity.