Tuning scaling factors of fuzzy logic controllers via reinforcement learning policy gradient algorithms

In this study a gain scheduling method for the scaling factors of the input variables to the fuzzy logic controller by means of policy gradient reinforcement learning algorithms has been proposed. The motivation for using PG algorithms is that they can scale RL problems into continuous high dimensional state-action spaces without the need for function approximation methods. Without incorporating any a-priori knowledge of the plant, the proposed method optimizes the cost function of the learning algorithm and tries to find optimal solutions for the scaling factors of the fuzzy logic controller. To show the effectiveness of the proposed method it has been applied to a PD type fuzzy controller along with a nonlinear model of an inverted pendulum. By performing different simulations, it is observed that the proposed method can find optimal solutions within a small number of learning iterations.