Using genetic algorithms to find cellular automata rule sets capable of generating maze-like game level layouts

The video game industry has grown substantially over the last decade and the quality of video games has also been advancing rapidly. In recent years, video games have been advancing to a point that the increased time required to manually create their content is making this process too costly. This has made procedural content generation a desirable option for game developers due to its speed of generating content, and the variety of content that a single PCG method can produced. The main purpose of this dissertation is to detail a new approach to procedurally generate video game level layouts, and to aid in further research in the area of procedural video game content generation. The new PCG approach investigated and developed in this study combined a genetic algorithm with cellular automata and a maze generation technique into a method for generating game level layouts with desired maze-like properties. The GA in this approach was utilized to evolve CA rules that, when applied to a maze configuration, would produce layouts with desired properties. This research discovered that CA rules could be evolved to generate level layouts with desired properties, and that there were a number of parameters which could affect the layouts these rules produced. These parameters include the number of cell states used in the CA, as well as the CA’s neighbourhood size and the number of times the CA rules were applied to their maze configurations. This research also discovered that the one factor that had the largest impact on the visual aspect of the generated layouts was the chosen chromosome representation