Procedural Content Generation in Mirror’s Edge using Markov Chains

Abstract

This research presents a novel map generation system for a parkour-based game, called Mirror’s Edge, that operates outside the constraints of a grid, allowing for more organic level design. While the system enables flexible map generation, it offers limited control over the process, relying on transition probabilities and a user-defined seed. A key challenge in the system is resolving intersections, which can lead to unsolvable maps. Performance evaluation showed significant variability, with map generation taking an average of 106 seconds without intersections and up to 600 seconds in more complex cases. A user study comparing procedurally generated maps to a manually designed Pure Time Trial Map revealed no significant differences in player experience, indicating the potential of the generation system to produce maps of comparable quality. However, the small sample size limits the conclusiveness of these findings.