Self-Organizing Interval Type-2 Fuzzy Q-learning for reinforcement fuzzy control

This paper proposes a reinforcement fuzzy control method using Self-Organizing Interval Type-2 Fuzzy Q-learning (SOIT2FQ). The fuzzy rules are of zero-order Takagi-Sugeno-Kang (TSK) type and the antecedent part uses interval type-2 fuzzy sets in order to improve fuzzy controller robustness. There are no fuzzy rules initially. The SOIT2FQ generates all fuzzy rules during control process using an online rule generation algorithm. The consequent part of each generated fuzzy rule is selected from a predefined discrete set containing all candidate values. The SOIT2FQ selects the consequent candidate values according to their Q-values in order to obtain a successful interval type-2 fuzzy controller. The SOIT2FQ is applied to reinforcement truck-backing control problem in clean and noisy environments, where only two reinforcement signals “success” and “failure” are used for training. Simulations show effectiveness and efficiency of the SOIT2FQ. Comparisons with type-1 fuzzy controller verify the noise robustness ability of the SOIT2FQ-designed interval type-2 fuzzy controller.