Hind, D., Harvey, C., 2022.

A NEAT Approach to Wave Generation in Tower Defense Games

Neural networks have shown promise when applied to video games and have proven effective at performing tasks such as dynamic difficulty adjustment (DDA). This paper explores how an evolving neural network can be applied to a tower defense game in order to generate dynamic content with the intent of increasing player engagement through the principals of flow. A NeuroEvolution of Augmenting Topologies (NEAT) neural network (NN) was trained as a wave manager to observe the current game state and generate an opposing enemy wave which best challenges the players' current tower defenses. The resulting network was compared against manually designed human waves in a blind A/B test using the Games Experience Questionnaire (GEQ) to evaluate the waves across a range of criteria. The results show that an approach like this could be viable if used for content generation purposes as no discernible difference existed in reported player experience between AI and human designed waves. However, the findings regarding subsequent increases to player engagement were inconclusive. More research is required in this field to conclusively determine if machine learning generated content can exceed the quality of content created by human designers, but the findings of this paper indicate that this approach may prove valuable to game developers in the near future by allowing them to save time and money by having AI generate content instead of requiring costly human game designer time.