Empirical Comparison of Heuristic Optimisation Methods for Automated Car Setup

Abstract

Tuning a race car to improve its performance by adopting an effective setup is crucial and an extremely challenging task. The Open Racing Car Simulator, referred to as TORCS, is a well-known simulator in which a race car requires a configuration of twenty two real-valued parameters for an optimal setup. In this study, various modern (meta)heuristic techniques, such as, evolutionary algorithms, swarm intelligence algorithm and selection hyper-heuristics, are evaluated using TORCS to solve the car setup optimisation problem across a range of tracks. An in-depth performance comparison and analysis of those techniques on the car setup optimisation problem are provided with a discussion on their strengths and weaknesses. The empirical results indicate the success of Covariance Matrix Adaptation Evolutionary Strategy for the car setup optimisation problem.