Topology Acquisition in Unknown Environment and Learn the Route to Destination Point by Autonomous Mobile Robots

Abstract

It is an important task for mobile robots that search the target and learn the route while recognizing the unknown environment topology. Usually, reinforcement learning is used as a learning method to know the route to the target while exploring the environment. However, in an unknown environment, it is difficult to predict the number of state division. Particularly, when the state division is too fine, the amount of calculation increases exponentially. In this paper, we propose a method to dynamically construct the state space of the environment using Growing Neural Gas and simultaneously search and learn the route to the target using Q-Learning. We applied multiple autonomous mobile robots to increase searching efficiency. The experimental result shows the effectiveness of the proposed method that can respond to dynamic environmental change.