Automatic Generation of Diverse Cavern Maps with Morphing Cellular Automata
Matthew Kreitzer, Daniel Ashlock, Rajesh Pereira · 2019 IEEE Conference on Games (CoG) · 2019
Cellular automata can be used to rapidly generate complex images, but controlling the character of those images can be difficult. This study continues experimentation with fashion-based cellular automata that generate cavern-like level maps and provides the beginning of a mathematical theory. Fashion-based automata are defined by a competition matrix with different cell states competing to capture territory. This study co-evolves pairs of competition matrices to permit the evolution of automata rules that can be spatially morphed to provide substantially more diverse types of maps than earlier systems using fashion-based cellular automata. As in earlier studies, the cellular automata rules function in local neighborhoods, meaning that the level generation system scales smoothly to any desired level map size. This reusability also permits variation of the type of morph used: a variety of spatial morphing styles are tested with the evolved rules. The theoretical treatment includes the derivation of a normal form for the cellular automata rules that informs the design of the fitness function and has application to understanding the fitness landscape of fashion based automata.