Improvement of a car racing controller by means of Ant Colony Optimization algorithms

The performance of a car racing controller depends on many factors. Although there are strong dependencies among these, some subproblems intrinsic to the design process can be addressed independently. Thus, if the aim is to minimize the lap time on a track, it becomes necessary to find the right trace around it and to determine the maximum speed on each stretch. This study proposes a hybrid solution to achieve this. Traces, which are described as a configuration of points the car must move towards, are found by means of an ant colony optimization algorithm, the ant system. As the maximum speed strongly depends on such traces, it must be adjusted for each one of them. In order to do this, a local search procedure is used in two ways: either when the best trace has been found, or during the search for such a trace. Results show a significant improvement with this technique in comparison with the original heuristic controller in terms of lap time. Moreover, the number of laps required for the algorithms to reach the solution makes them viable as a learning mechanism in real-time simulation environments.