A Hybrid Approach to Procedural Generation of Roguelike Video Game Levels

Algorithmic generation of data, known as procedural content generation, is an attractive prospect within the game development industry as a means of creating infinitely fresh and varied content. In this paper, we present an approach to level generation for roguelike dungeon style levels, based on our examination of the suite of existing approaches used in formal research. Our generator aims to create simple dungeon style level layouts that are always playable. We utilise a hybrid technique combining context free grammars to generate a description of levels and a cellular automata inspired process to generate the physical space. The generator proves successful at consistently generating dungeon layouts that maintain completability at all times with sufficient variation, when accounting for the occasional need for corrective actions. We conclude that there is substantial value in hybrid approaches to automated level design and propose a new heuristic by which to assess dungeon style level content.