Loading...

Deep Learning-Based Procedural Content Generation for Video Games

Morakabi Esfahani, Mohammad Hadi | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58175 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Hashemi, Matin
  7. Abstract:
  8. Procedural Content Generation (PCG) is a critical challenge in video game design, aimed at automating the creation of complex maps and levels. This thesis presents a novel deep learning-based approach for PCG that integrates transformer architectures with reinforcement learning techniques. Building on previous works, this research seeks to enhance the quality, diversity, and success rate of generated playable levels. The proposed method utilizes transformer architectures to model sequences of actions, states, and rewards, predicting optimal game content based on past trajectories. Offline datasets, generated by semi-expert agents trained in the PCGRL framework, serve as the foundation for training our model. The integration of transformers enables the model to generate coherent and meaningful maps while maintaining flexibility across various game environments. Moreover, a comprehensive analysis of stopping criteria and success rates indicates that the proposed model effectively utilizes map tiles to construct playable levels. This capability makes it a powerful tool for content generation in video games, offering a promising solution for the challenges of generating complex game content
  9. Keywords:
  10. Reinforcement Learning ; Procedural Content Generation (PCG) ; Sequence Modeling ; Two Dimentional Video Game Level Generation ; Video Games

 Digital Object List

 Bookmark