Reinforcement learning for fuzzy agents: application to a pighouse environment control

摘要: Fuzzy Actor-Critic Learning (FACL) and Q-learning (FQL) are reinforcement learning methods based on Dynamic Programming (DP) principles. In this chapter, they used to tune line the conclusion part of Inference Systems (FIS). The only information available for is system feedback, which describes in terms reward punishment task fuzzy agent has realize. At each time step, receives a reinstate. problem involves optimizing not direct reinforcement, but also total amount reinforcements can receive future. To illustrate use these two methods, we first applied them have find controller drive boat from one bank another, across river with strong non-linear current. Then, well-known Cart-Pole Balancing Mountain-Car problems be able compare our other focus important characteristic aspects FACL FQL. experimental studies had shown superiority respect related literature. We found that generic allow us learn every kind (continuous states, discrete/continuous actions, various types functions). Thanks flexibility, been successfully an industrial problem, discover policy pighouse environment control.