Identifying Catastrophic Failures in Offline Level Generation for Mario

Abstract

Video games are pushing the boundaries of the creative medium to be more realistic. This realism demands the game content to be tailored to improve the gaming experience. Generating content is a challenging task and automated approaches based on Artificial Intelligence techniques can help the gaming industry with this problem. The focus of our research is to produce adaptive levels for action-adventure games. We present a technique to identify catastrophic failures in offline level generation of the popular game "Mario". Our approach produces levels that have high replay value and have limited catastrophic failures, thereby improving the quality of the levels generated. This paper also presents taxonomy of Procedural content generation and Search-based PCG techniques. That is to our best knowledge the first wide-ranging survey of both the approaches.