On-line learning of a fuzzy controller for a precise vehicle cruise control system

Usually, vehicle applications need to use artificial intelligence techniques to implement control strategies able to deal with the noise in the signals provided by sensors, or with the impossibility of having full knowledge of the dynamics of a vehicle (engine state, wheel pressure, or occupants' weight). This work presents a cruise control system which is able to manage the pedals of a vehicle at low speeds. In this context, small changes in the vehicle or road conditions can occur unpredictably. To solve this problem, a method is proposed to allow the on-line evolution of a zero-order TSK fuzzy controller to adapt its behaviour to uncertain road or vehicle dynamics. Starting from a very simple or even empty configuration, the consequents of the rules are adapted in real time, while the membership functions used to codify the input variables are modified after a certain period of time. Extensive experimentation in both simulated and real vehicles showed the method to be both fast and precise, even when compared with a human driver.