Mania Archetype: Chart Generation for Rhythm Action Games with Human Factors

Abstract: In recent years, it has been proven feasible to generate rhythm game charts through machine learning models. This has reduced the cost of chart generation and served as an auxiliary tool for novice chart creators. Machine learning generation allows players to experience new songs and charts earlier. However, the article does not provide sufficient discussion on how to generate challenging and high-quality human-like charts for the versatile 4k Mania mode, which uses 4 scrolling 'note highways' to display the notes to be played. The focus of this article is on improving the generation of 4k charts in OSU! Mania through research on machine learning chart generation, with the aim of creating more interesting and accurate charts. Additionally, we propose a more comprehensive and humane standard for evaluating the quality of generated charts. This was informed by interviews with experienced players and chart creators.

Keywords: Rhythm game, Deep learning, Level generation, Music feature extraction, Game experience.

DOI: 10.54941/ahfe1005000

Cite this paper:

​