Searching for the optimal racing line using genetic algorithms

Finding the racing line to follow on the track is at the root of the development of any controller in racing games. In commercial games this issue is usually addressed by using human-designed racing lines provided by domain experts and represents a rather time consuming process. In this paper we introduce a novel approach to compute the racing line without any human intervention. In the proposed approach, the track is decomposed into several segments where a genetic algorithm is applied to search for the best trade-off between the minimization of two conflicting objectives: the length and the curvature of the racing line. The fitness of the candidate solutions is computed through a simulation performed with The Open Racing Car Simulator (TORCS), an open source simulator used as testbed in this work. Finally, to test our approach we carried out an experimental analysis that involved 11 tracks provided with the TORCS distribution. In addition, we compared the performance of our approach to the one achieved by a related approach, previously introduced in the literature, and to the performance of the fastest controller available for TORCS. Our results are very promising and show that the presented approach is able to reach the best performance in almost all the tracks considered.