Evolutionary Learning of Fuzzy Control Rule Base for an Autonomous Vehicle

This paper presents a hybrid learning method in which fuzzy logic controllers (FLC) are automatically designed by means of a genetic algorithm (GA). A messy coding scheme is proposed which allows a compact and exible representation of the fuzzy rules in a genetic string. The complexity and size of the rule base are reduced which enables the GA to solve the design task even for FLCs with a large number of input variables. A dynamically weighted objective function is suggested for control problems with multiple tasks, which prevents the GA from premature convergence on FLCs that are specialized exclusively to the easier subtasks. In order to achieve a robust control behaviour for a broad spectrum of situations a second GA coevolves a set of training situations to evaluate the performance of the FLCs. The method is applied to learn a FLC, which implements a behaviour of a mobile robot. The robot is given the task to reach a goal point and to avoid collisions with obstacles on its way, which it perceives by means of ultrasonic sensors. The performance of the FLCs, which are learned in simulated environments, is tested afterwards in real world experiments with the mobile robot.