Computational Intelligence in Racing Games

This chapter surveys the research of us and others into applying evolutionary algorithms and other forms of computational intelligence to various aspects of racing games. We first discuss the various roles of computational intelligence in games, and then go on to describe the evolution of different types of car controllers, modelling of players’ driving styles, evolution of racing tracks, comparisons of evolution with other forms of reinforcement learning, and modelling and controlling physical cars. It is suggested that computational intelligence can be used in different but complementary ways in racing games, and that there is unrealised potential for cross-fertilisation between research in evolutionary robotics and CI for games. 1 On the Roles of Computational Intelligence in Computer Games Computational Intelligence (henceforth “CI”) is the study of a diverse collection of algorithms and techniques for learning and optimisation that are somehow inspired by nature. Here we find evolutionary computation, which uses Darwinian survival of the fittest for solving problems, neural networks, where various principles from neurobiology are used for machine learning, and reinforcement learning, which borrows heavily from behaviourist psychology. Such techniques have successfully tackled many complex real-world problems in engineering, finance, the natural sciences and other fields. Recently, CI researchers have started giving more attention to computer games. Computer games as tools and subjects for research, that is. There are two main (scientific) reasons for this: the idea that CI techniques can add value and functionality to computer games, and the idea that computer games can act as very good testbeds for CI research. Some people (like us) believe that computer games can provide ideal environments in which to evolve complex general intelligence, while other researchers focus on the opportunities competitive games provide for comparing different CI techniques in a meaningful manner. J. Togelius et al.: Computational Intelligence in Racing Games, Studies in Computational Intelligence (SCI) 71, 39–69 (2007) www.springerlink.com c © Springer-Verlag Berlin Heidelberg 2007 U nc or re ct ed P ro of 40 J. Togelius et al. We would like to begin this paragraph by saying that conversely, many game developers have recently started giving more attention to CI research and incorporated it in their games. However, this would be a lie. At least in commercial games, CI techniques are conspicuously absent, something which CI researchers interested in seeing their favourite algorithms doing some real work often attribute to the game industry’s alleged conservatism and risk aversion. Some game developers, on the other hand, claim that the output of the academic Computational Intelligence and Games (“CIG”) community is not mature enough (the algorithms are too slow or unreliable), and more importantly, that the research effort is spent on the wrong problems. Game developers and academics appear to see quite different possibilities and challenges in CI for games. With that in mind, we will provide a brief initial taxonomy of CIG research in this section. Three approaches to CI in games will be described: optimisation, innovation and imitation. The next section will narrow the focus down to racing games, and survey how each of these three approaches have been taken by us and others in the context of such games. In the rest of the chapter, we will describe our own research on CI for racing games in some detail.