Enhancing Procedural Game Level Generation using Transformer-based Neural Architectures
Tianze Zhao, Zhijun Fan · 2024
In the realm of computer game design, Procedural Content Generation (PCG) stands as a pivotal technique, enabling the automated creation of diverse and intricate game levels. Traditional PCG methodologies, while effective, often present limitations in terms of creativity and variability. Addressing these constraints, this paper introduces a novel approach leveraging the power of Transformer-based neural architectures, specifically tailored for game level generation. By training on extensive datasets of existing game levels, our model is adept at internalizing intricate level structures and patterns, thereby facilitating the generation of levels that not only exude creativity but also resonate with the expectations of players. Through rigorous experimentation on various popular computer games, our findings attest to the superiority of the proposed model in comparison to conventional PCG techniques, especially in terms of ingenuity, diversity, and overall player experience.