Abstrakt
Procedural content generation (PCG) is now used in many games to generate a wide variety of content. One way of evaluating this content is by artificial intelligence (AI) controlled players.
Inversely, PCG content can also be used when training AI players to ensure generalization. Evolutionary algorithms are employed in both AI and PCG fields, but rarely simultaneously.
In this work, we use evolutionary algorithms for both AI players and level generation in the platformer game Super Mario. We further combine them into a coevolution, where the AI players are evaluated by adapting level generators, and vice versa, level generators are evaluated by adapting AI players.
This yields an AI player trained on gradually more difficult levels and a sequence of level generators with gradually increasing difficulty. Such sequence of generators might be useful for human game playing in commercial games.