Evolving controllers for simulated car racing

That car racing is a challenging problem, generating considerable public excitement, is evident from the huge amount of time and money invested both in practising and watching physical car racing, and in developing and playing racing games. For the same reasons, the problem(s) cannot sensibly be considered "trivial" or "solved" - no one would want to watch a race where the drivers were perfect. Though experiments with neural and evolutionary methods have undoubtedly taken place in commercial game studios, these have not been published for reasons of commercial confidentiality. The academic evolutionary computation community has apparently not devoted much energy to the car racing domain. One exception is Wloch and Bentley [11], who used evolutionary algorithms to optimize the parameters of simulated Formula 1 racing car with good results. However, they did not try to evolve the car controller, but rather used the simulator's built-in controller.