AI-Evaluated Procedural Content Generation for Video Game Level Design

  • Daniel Millward University of Derby


Procedural Content Generation (PCG) has long been used in video games to produce content that would otherwise be restricted by time, cost, or memory limitations. A relatively new area of research within the field is that of Search-Based PCG, which uses a process inspired by biological evolution to evolve content over iterative generations. Expanding on this work, we present a prototype of AI-Evaluated PCG, which aims to generate content with as few human-authored rules as possible, by using an AI-driven Bot for content evaluation. Our working prototype focuses on level generation for a 2D platforming game similar to Nintendo's Super Mario Bros. We develop a bot which, through the use of an artificial neural network, learns from player behaviour across a number of training sessions, and replicates that behaviour to test-run and evaluate generated levels. Level evaluation is used to score potential level solutions, and a genetic algorithm employed to evolve our level content. Ultimately our prototype is able to successfully generate complete and playable levels, with room for improvement. We detail a list of recommendations for further research.
How to Cite
MILLWARD, Daniel. AI-Evaluated Procedural Content Generation for Video Game Level Design. Discovery, Invention & Application, [S.l.], june 2017. Available at: <>. Date accessed: 06 dec. 2019.