Generating non-monotone 2D platform levels and predicting difficulty

Abstract

This thesis presents a method to procedurally generate levels for 2D side scrolling platformer games. The method yields non-linear gameplay and guarantees bi-directional paths. This method is based on a depth-first search technique to find an arrangement of building pieces, which represent dis- cretized jumps, and that satisfies a set of constraints. We also present a method to estimate the difficulty of the generated levels. A statistical model is constructed for individual challenges, based on three general parameters of the challenges. This model is used to find the path that yields the highest probability of successfully traversing the level. We verify our results by investigating the relationship between the computed difficulty values and the number of deaths measured.