GREY SYSTEM BASED REACTIVE NAVIGATION OF MOBILE ROBOTS USING REINFORCEMENT LEARNING

A reactive navigation approach for autonomous mobile robots in unknown dynamic environments is presented using grey theory and reinforcement learning techniques. Reactive navigation is one of the basic tasks for mobile robots to achieve memoryless intelligence. In this paper, a grey system is designed for the sensor-based reactive control of mobile robots with several primitive behaviors, i.e., obstacle-avoidance, goal-seeking, local-trap escaping and emergency behavior. The uncertainties in the sensory information and decision making are represented and operated using frequency grey numbers. A grey reinforcement learning method is proposed for the learning of grey rules and behavior decisions to coordinate the primitive behaviors. The experimental results show that this approach is effective for reactive navigation and learning control with incomplete environmental information, in unknown dynamic environments.