Procedural level generation with answer set programming for general Video Game playing

This paper proposes an automatic way of evolving level generators for arbitrary 2D games, which are described in the Video Game Description Language (VGDL). The process works as follows: A game described in VGDL is interpreted and transformed in a set of rules defined in Answer Set Programming (ASP), along with other general and customizable rules. Although a set of rules described in ASP can generate multiple levels, not all of them will be playable or well designed. Therefore, an evolutionary process is run to determine the values of the parameters of those customizable rules. The different level generators are evaluated with general video game playing agents, which are able to play any game and level in the framework. The aim is to maximize the difference between their performance in the levels generated, under the assumption that levels are better designed if good skilled players play better than poor agents. This work presents some initial experiments that suggest that it is possible to evolve interesting level generators using this technique, and outlines some lines of future work.