AN ADAPTIVE UNIDIRECTIONAL LINEAR-RESPONSE FUZZY CONTROLLER BASED ON REINFORCEMENT LEARNING

This paper presents an adaptive fuzzy controller using unidirectional linear response (ULR) elements. The basic functions for a fuzzy contro ller including membership, minimum and defuzzification functions are r ealized by the ULR elements. Because the ULR element has diode-like ch aracteristics, it can be implemented by a diode-connected mean-opinion score (MOS) transistor in current-mode implementations. The hardware implementation of the fuzzy controller using ULR elements should also be very simple and straightforward. The ULR fuzzy controller is applie d to an inverted pendulum problem, and the effectiveness of the propos ed ULR controller architecture and its learning capability based on re inforcement learning are demonstrated.