Maximising the Performance of Quality-Diversity Algorithms within PCG Systems

点击翻译按钮获取中文摘要

Quality-Diversity (QD) algorithms are a new branch of high performing search algorithm, the benefits of which make them especially relevant to video-game Procedural Content Generation (PCG). However, at time of writing there have been no studies contrasting and evaluating the performances of alternative QD algorithms within a PCG domain, a gap this project aims to address. Through comparative experimentation I aim to discover which QD algorithm is best able to generate diverse and high quality game content, and which factors are most important for optimising its performance. The project would first focus on the domain of generating Super Mario Bros levels and then extend to multiple different game PCG spaces. Successful completion of this work will substantially deepen understanding of this area, allowing the creation of more powerful and sophisticated content generation tools using QD algorithms.