Reinforcement tuning of fuzzy rules

Fuzzy rules for control can be effectively tuned via reinforcement learning, which only requires information on the success or failure of the control application. The tuning process allows one to generate fuzzy rules which are unable to accurately perform control and have them tuned to be rules which provide smooth control. The paper explores a new simplified method of using reinforcement learning for the tuning of fuzzy control rules. Results from the domain of pole balancing are given and compared to another approach. It is shown that the learned fuzzy rules are able to provide smoother control in the pole balancing domain than another tuning approach.