Generation of multi-floor levels for first-person shooters

Abstract

This work introduces a Search-Based Procedural Content Generation tool for First-Person Shooters similar to Quake and Unreal Tournament. Procedural Cement Generation is used in many of today's games and has been growing in the academia scene, from the generation of platformer levels, racing tracks and even game rules. Not much research related to the generation of levels for First-Person Shooters exist so this thesis aims on contributing to that work by ex:;ending on few of the works already done. We use a Genetic Algorithm to manipulate several elements of the level that form the shape of our map, together with the in-game entities that one finds in a typical first-person shooter such as health recovery and weapons. The physical form of our maps are generated as either one-floor or two-floor maps, adding an extra layer to the map. The evaluation of our maps is a combination of simulation and direct-evaluation functions that encourage balance for the players, resulting in emergent maps that may exhibit interesting design patterns. The above is possible by proposing heuristics from work related to level design patterns and thus produce the above methodology.