Incorporating fuzzy logic to reinforcement learning [mobile robot navigation]

摘要

Proposes a sensor-based navigation method that utilizes fuzzy logic in reinforcement learning algorithms for navigation of a mobile robot in uncertain environments. The sonar readings are codified in distance notions by fuzzy sets and a modification in the R-learning algorithm by incorporating fuzzy logic is proposed. Fuzzy logic is used for weighting the immediate reward value, that is a variable present in most reinforcement learning algorithms. The effectiveness of the modified algorithm, R'-learning, is verified in several tests and compared to the performance of the R-learning algorithm.