Generating Rooms using Generative Grammars and Genetic Algorithms

Bi-dimensional game environments are usually not designed as a single piece, but built on top of smaller pieces called tiles which, are typically grouped into tilesets. These work as a palette for creating backgrounds, so the resulting map also demands careful positioning of game objects. However, many modern games demand a high volume of scenarios composed of a myriad of tilesets, i.e., creating this kind of game content may be difficult and challenging for humans, so Procedural Content Generation (PCG) techniques can be used to address this problem. In this paper, we propose a novel PCG technique to speed up this process. We model generative grammars whose association rules yield strings that represent the objects arranged in the scene. We show that it is possible to define simple generation directives leaving it to a genetic algorithm process to control the best distributions of weights on the rules. We evaluate our technique under the scenario of generating game rooms for the popular JRPG (Japanese Role-Playing Game), resulting in varied and good-looking rooms.