Q-Learning Algorithm for a Mobile Robot Obstacle Avoidance in an Unknown Environment Based on Artificial Neural Networks

Abstract

In this paper neural network representation for the Q-learning algorithm of a mobile robot is presented. The property of adaptability in modern intelligent manufacturing environments is the significant advantage comparing with the traditional designed facilities. Presented approach characterizes its efficiency and simplicity considering the learning process of the intelligent agent - a mobile robot. Experience gathered from the external sensors in an obstacle avoidance task presents the input of the neural network, enabling the mobile robot to learn the value of the selected actions as the output of the neural network, gradually improving its behaviour. If the more learning epochs are conducted mobile robot could became autonomous, which can be crucial advantage for the 21st century manufacturing systems.