Work

Advancing Level Generation in Super Mario Bros.: A Proposal for Combining Diffusion Models with Real-Time Playability Assessment

Year: 2024
Type: article
Abstract: This thesis presents an innovative approach in the realm of Procedural Content Generation (PCG) for video games, with a specific focus on the iconic Super Mario Bros.Leveraging a diffusion model built on a U-Net architecture, the study explores new frontiers in level generation, training the model on a diverse array of existing levels.The primary objective is to capture the intricate mechanics and design ethos of the game, ensuring the generated levels are not only diverse but also align with the game's original spirit.Initial results demonstrate the model's success in producing both high-quality and playable levels.As a novel extension, the research proposes integrating a real-time playability assessment mechanism, employing differentiable A* and a classifier system, into the training phase.This proposed integration, though yet to be experimentally validated with the diffusion model, is anticipated to enhance the model's efficiency in generating inherently playable levels.This advancement signifies a potential leap in PCG, suggesting a method that could refine game development processes and inspire future explorations in game design and artificial intelligence.
Institution: Waseda University
Language: English
Open Access status: green
OpenAlex ID: W7017207843
FWCI: 0.16
Cites: 40
Cited by: 2
Related to: 10