Automatic level generation for platform videogames using genetic algorithms

In this document we present an investigation on automatically generating levels for platform videogames. Common approaches for this problem are rhythm based, where input patterns are transformed in a valid geometry, and chunk based, where samples are humanly created and automatically assembled like a puzzle. The proposal hereby presented is to explore this challenge with the usage of Genetic Algorithms, facing it as a search problem, in order to achieve higher expressivity and less linearity than in rhythm based approach and without requiring human creation as it happens with the chunk based approach. With simple heuristics the system is able to generate playable levels in a small amount of time (one level is created in less than a minute) and with considerable diversity, as our results show.