Reinforcement Learning can be seen as a way of programming agents by reward and punishment for solving specific tasks through repeated interactions with the environment. In this work, the performance of the most important reinforcement learning algorithms: Qlearning, Rlearning, H-learning is investigated in the context of a navigation task avoiding obstacles. Furthermore, this work proposes a sensor-based navigation method, named R'-learning, which incorporates fuzzy logic into the Rlearning algorithm for mobile robot navigation in uncertain environment. An application is realized consisting of teaching the robots to find small objects in a corridor. For this, a state set mapping has been proposed through force field concepts. R'learning algorithm has been used for this navigation task. The robot showed to have satisfactory behaviors in performing this task.
展开▼
