Designer-driven Procedural Game Content Generation using Multi-agent Evolutionary Computation
Creating computer game levels that offer good playability and interesting layouts is a laborious and costly task. In particular, multiplayer game levels necessitate careful balancing of gameplay between teams and clear objectives for players. Expert knowledge that draws on a good understanding of player experience and strong level design skills results in maps that players enjoy and appreciate. The success of new level designs hinges on this expertise. This study introduces a Multi-Agent System based on heuristics developed from expert level designers, that is able to augment game designers of all levels of expertise and help them create First Person Shooter levels with good playability. An interactive evolutionary algorithm is employed to provide designers with several level design options. The human user stays in full control of the decisions made during the design process, while the agent system provides suggestions that promise good playability and a positive player experience. The system has been evaluated using game designers with varying levels of experience, and the results show great promise. A Multi-Agent System in addition to the Interactive Genetic Algorithm outperforms a purely human-centric solution: Game designers of all skill levels draw heavily on the suggestions made by the Multi-Agent System. Recommendations for future developments are given.