Constructing a fuzzy logic controller using evolutionary Q-learning

This paper proposes an evolutionary Q-learning algorithm for the design of a fuzzy logic controller. By defining Q-values as a functional value of state and each fuzzy logic controller, Q-learning is easily applied to the group of fuzzy logic controllers. An evolutionary algorithm which uses Q-values for the evaluation of the fitness value is proposed to extract the best fuzzy logic controller from the group of fuzzy logic controllers. This algorithm can generate a fuzzy logic controllers when only a binary reinforcement signal is available. The feasibility of the proposed algorithm is shown through the simulations on cart-pole balancing problem