Procedural Generation of Metroidvania Style Levels (thesis)

Abstract

Video game maps can become dull with repeated play-throughs and handcrafting a variety of maps can be a tedious and time consuming process. This is especially true for games of the Metroidvania genre, games which focus on exploration. If there was a way to adequately automate the creation of levels, then in theory, the games would have enhanced replay value. Previous researchers have used artificial intelligence and genetic programming techniques to engineer new mappings. But, is it possible to procedurally generate levels using graph theory and without using training examples or simply placing pre-built assets? In this paper we propose a system to model Metroidvania maps as directional graph structures. The system uses an algorithm that crafts graphs meeting all of the constraints necessary for level generation. These generated graphs are verified as winnable with the keys assigned to appropriate nodes. Once the graph has been created and validated it is rendered into a 2-D level using pygame. During the rendering process, the game demo constructs the walls and platforms essential to the game. We were able to procedurally generate Metroidvania levels of varying sizes and gating techniques using this sequence of steps.