Selection of fuzzy control rules using automatic tuning of membership functions

Tuning of membership functions using learning procedure in a neuro-like approach has been developed to select fuzzy control rules. The tuning method is applied to simulation of driving control of a model car to run on a straight road. Simulation results bring out similar optimal trajectories of the car in both cases of 3/spl times/3 (=9 rules) and 7/spl times/7 (=49 rules) control rule types after tuning. Estimation function of errors used in the tuning of 3/spl times/3 rule type rapidly decreases to the convergent value of that used in 7/spl times/7 with increasing learning iteration.<
>