Self-tuning fuzzy logic controller for reactive navigation

Low-level tasks for the navigation of autonomous vehicles can be accomplished efficiently by behavioral-based approaches mapping sensory data onto control commands in a reactive way, without using internal representations. Such a direct mapping can be usefully realized using fuzzy logic. All the choices involved in fuzzy controller design, impacting the final result in terms of complexity and performance, are generally made intuitively and/or by trial-and-error processes. Automatic learning and self-tuning of rules are required for adaptive controllers operating in real dynamic environments instead. This paper presents a method for minimizing the number of rules in a fuzzy controller without affecting its overall performance. The experiments have been made reducing the number of rules of a fuzzy controller developed for driving a TRC Labmate based vehicle along the wall on its right-hand side mapping ultrasonic sensor readings onto steering velocity values. Fuzzy rules have been learned automatically from training data collected during operator-driven runs of the vehicle. Experimental results, comparing the original and the optimized versions of the controller successfully driving the vehicle along arbitrarily shaped walls in real unknown environments, are provided.