Human computation for procedural content generation in platform games

Abstract

One of the major challenges in procedural content generation in computer games is to automatically evaluate whether the generated content has good quality. In this paper we describe a system which uses human computation to evaluate small portions of levels generated by an existing system for the game of Infinite Mario Bros. Several such evaluated portions are then combined into a full level of the game. The composition of the small portions into a full level is done by accounting for the human-annotated information and the mathematical model of tension arcs used in interactive drama and storytelling. We tested our system with human subjects and the results show that our approach is able to generate levels with better visual aesthetics and that are more enjoyable to play than other existing approaches.