Three population co-evolution for generating mechanics of endless runner games

Abstract

Procedural content generation (PCG) is increasingly used to generate many aspects in a variety of games. However, using PCG to generate mechanics of games is rarely attempted.

Past approaches include using cooperative coevolution to generate the rules and the environment of a game, which had interesting results. Our approach extends this idea by using coevolution with three populations, also generating the evaluating player.

We used this approach to generate endless runner games, for which it was able to generate novel mechanics.

Keywords

• coevolution
• evolutionary algorithms
• game design
• procedural content generation