Qualitative control for mobile robot navigation based on reinforcement learning and grey system

A qualitative control method using reinforcement learning (RL) and grey system is developed for mobile robot navigation in an unknown environment. New representation and computation mechanisms are key approaches for learning and control problems with incomplete information or in large probabilistic environments. In this paper, the uncertainties in sensor information and qualitative localization are represented using grey systems. Traditional RL methods are also combined with grey theory for mobile robot navigation control and a new grey reinforcement learning (GRL) method is proposed to solve complex problems where environment information is incomplete and grey models are constructed through representing incomplete information with grey algebra. The experimental results verify the effectiveness and superiority of the qualitative control system and the presented approaches also show alternative ways for the measurement and learning control problems with incomplete information of the environment. Copyright © 2008 SoftMotor Ltd.