A Hierarchical Approach to Generating Maps Using Markov Chains

In this paper we describe a hierarchical method for procedurally generating maps using Markov chains. Our method takes as input a collection of human-authored two-dimensional maps, and splits them into high-level tiles which capture large structures. Markov chains are then learned from those maps to capture the structure of both the high-level tiles, as well as the low-level tiles. Then, the learned Markov chains are used to generate new maps by first generating the high-level structure of the map using high-level tiles, and then generating the low-level layout of the map. We validate our approach using the game Super Mario Bros., by evaluating the quality of maps produced using different configurations for training and generation.