Assessing the Quality of Car Racing Controllers in a Virtual Setting under Changed Conditions

Abstract

This paper discusses several controllers based on fuzzy logic and evolutionary concepts applied to a car racing simulation and their robustness to changing physics of the cars. The challenge is to design a car controller that passes the next three arising waypoints faster than an opponent car controller. Two fuzzy controllers are compared to two evolutionary optimized controllers in solo races as well as in head-to-head competitions, where all controllers compete head-to-head against the other controllers. The influence of some parameter settings is investigated as well. The results emphasize the robustness of the fuzzy controllers, not differing much from each other. Overall, the fuzzy controllers perform better with different parameter settings of the driving physics except when reverse speed is equal to forward speed.