Integrating Expert Knowledge into Fuzzy Reinforcement Learning

The field of the reinforcement learning (RL) is the set of the trial-and-error types machine learning methods. The learning of the conventional RL algorithms, such as Q-learning, SARSA, and various Fuzzy Q-learning variants, begin without an initial knowledge. Before the learning, the knowledgebase of the system is empty. It is gradually collected during the learning process based on the reinforcement signals of the environment. Integrating a priori expert knowledge into reinforcement learning methods can lead to more efficient learning process. The goal of the paper is to provide a brief overview of the possibility of incorporating expert knowledge into fuzzy reinforcement learning. Additionally, it aims to introduce a Fuzzy Rule Interpolation-based RL method suitable for integrating external expert knowledge into the system. This method is capable of fine-tuning the initially imprecisely defined expert knowledge during the learning process, thereby correcting and optimizing it. Furthermore, the paper will also discuss the efficiency of the expert knowledge injection by using a simple RL simulation example.