Reinforcement learning relies on the association between a goal and a scalar signal, interpreted as reward or punishment. The objective is not to reproduce some reference signal, but to progessively find, by trial and error, the policy maximizing the rewards. This paper presents the basis of reinforcement learning, and two model-free algorithms, Q-Learning and Fuzzy Q-Learning.

This paper presents a fuzzy navigation system for an autonomous robot. A behavior-based control system provides the robot with the adaptibility necessary for coping with a dynamically changing environment. Moreover, a reinforcement learning method is used for on-line rule optimization. keywords: Fuzzy Logic, obstacle avoidance, wall-following, reinforcement learning, on-line optimization. 1 Introduction Many manufacturing systems require autonomous robots capable of coping with dynamically changing unknown environments: transport of parts from one place to another, removing any undesired objects from floors, surveillance, and so on. Traditional methods have met great difficulties in application to realtime problems. They often require large computational power and memory, incompatible with embedded systems and real-time requirements. On the other hand, reactive or behavior-based systems offer an interesting alternative and have two main advantages: ffl prior knowledge can be easily...

The paper explores the possibilities of learning steering control behaviors for navigation from a human operator. Behaviors allow the development of structured navigation control in the face of uncertain environment models. However, their design is rendered difficult by the nature of real sensor readings, which are nonlinear, uncertain and even contradictory. Hence, learning strategies are interesting in that they incorporate actual sensor information from human-driven real-world experiments. In this sense, a modelling method is presented that efficiently generates fuzzy behavior rules from a set of input-output data. Yet, the application of these techniques poses particular problems due to user-vehicle interaction issues, which are discussed in the paper. Experimental results illustrate the whole procedure applied to a real mobile robot outfitted with ultrasonic sensors.