Using patterns as objectives for general video game level generation

摘要

Automatically generating multiple levels and creating infinite game-play for video games is a complicated task. In this paper, we introduce a pattern-based level generator for general video games. Our generator uses a genetic algorithm to automatically generate levels. The fitness function promotes contiguous patterns and penalizes levels with single or isolated elements. The generator generates levels using two different types of fitness functions; alpha and beta. alpha levels are generated using the original fitness function and beta levels through the inverse of the fitness function. In our experiments, Relative performance profiles indicate that controllers using more advanced techniques achieve better overall performance on alpha generated levels. The generator finished at the third spot in the 2018 general video game level generation competition.