Gaussian-based adaptive fuzzy control

Fuzzy logic controllers are well known for robustness and performance, and it has been proven that fuzzy logic controllers can approximate an optimal controller with arbitrary accuracy. However, adapting fuzzy rule bases towards an optimal solution has proven challenging. This paper describes a new adaptive algorithm for a fuzzy logic controller. Previously studied adaptive algorithms such as Lyapunov methods and neural-network methods effectively reduce error, but they are complex and potentially distort the rule base so that it loses robustness. The proposed method “shapes” the fuzzy rule base as a normal distribution to maintain robustness. The algorithm is tested on a robotic clamping operation; in three experiments, it reduced the error variance by 35%, 44%, and 53%.