Evolving a TORCS Modular Fuzzy Driver Using Genetic Algorithms

Abstract

This work presents an evolutionary approach to optimize the parameters of a Fuzzy-based autonomous driver for the open simulated car racing game (TORCS). Using evolutionary algorithms, we intend to optimize a modular fuzzy agent designed to determine the optimal target speed as well as the steering angle during the race. The challenge in this kind of fuzzy systems is the design of the membership functions, which is usually done through a trial and error process, but in this paper an adapted real-coded Genetic Algorithm with two different fitness functions - has been applied to find the best values for these parameters, obtaining a robust design for the TORCS controller. The evolved drivers were tested and evaluated competing against other TORCS controllers in practice mode, without rivals, and real races. The optimized fuzzy-controllers yield a very good performance, mainly in tracks that have many turning points, which are, in turn, the most difficult for any autonomous agent. Thus, this is a real enhancement of the baseline fuzzy controllers which had several difficulties to drive in this kind of circuits.