This talk describes the use of a socially motivated evolutionary algorithm, Cultural Algorithms, to design a controller for a 3D racing game. The 3D Racing Game was one of the competitive events held at the IEEE World Congress in Hong Kong in 2008. The goal of the project was to use the Cultural Algorithm to generate a set of control rules for an autonomous driver that would compete in the events. The driver was modeled as a state machine and the rules were associated with actions that can be taken in each state. The Cultural Algorithm consists of a Population Space, a Belief Space, and a communication protocol that connects them together. Here the population was viewed as a network of chromosomes whose aggregate behavior was generalized in the Belief Space. The emergent patterns in the belief space serve to influence the modifications of individuals in the network using the genetic operators. The impact that different network configurations had on the resultant rule performance was studied. What network configuration that works best for rule learning here and why will be discussed. The in-race behavior of the controller as part of the “pack” will be highlighted.

See link below for background article.