Map Generation for Roguelike Games from Textual Descriptions Using Large Language Models

• Gustavo Gurgel Medeiros UFC
• Cristiano Bacelar de Oliveira UFC

DOI: https://doi.org/10.5753/sbgames.2026.25721

Resumo

Introduction: Procedural Content Generation (PCG) enhances replayability in roguelike games but remains difficult to control via algorithmic parameters. Large Language Models (LLMs) introduce a “text-to-map” paradigm but still struggle with spatial reasoning. Objective: We propose an automated system that generates playable and structurally coherent roguelike maps from textual descriptions. Methodology: The architecture combines an LLM-driven Assets Generator for semantic content with a deterministic PCG algorithm for spatial layout generation and uses Retrieval-Augmented Generation (RAG) to map textual intent into visual assets, based on a dataset of 489 pixel-art images. Results: Evaluation indicates high semantic fidelity, with coherence scores of up to 93.6%, yielding visually cohesive and playable maps.

Palavras-chave: Procedural Content Generation, Large Language Models, Roguelike Games, Map Generation

Referências

Auxtero, A. L. S. (2023). Game environment design creator using artificial intelligence procedural generation. Master thesis (computer science and engineering), NOVA University, Lisbon, Portugal. Acesso em: 27 jan. 2026.

Ehrig, H., Pfender, M., e Schneider, H. J. (1973). Graph-grammars: An algebraic approach. In Proceedings of 14th Annual Symposium on Switching and Automata Theory (swat 1973), pages 167–180, Iowa City, Iowa. Institute of Electrical and Electronics Engineers. Acesso em: 27 jan. 2026.

Epyx Inc. (1984). Rogue. Digital game.

Gallotta, R., Liapis, A., e Yannakakis, G. (2024). Consistent game content creation via function calling for large language models. In Proceedings of 2024 IEEE Conference on Games (CoG), pages 1–4, Milan, Italy. Acesso em: 27 jan. 2026.

Huang, S. (2025). Word2minecraft: Generating 3d game levels through large language models. arXiv preprint arXiv:2503.16536. Acesso em: 27 jan. 2026.

Kenney (2026). 1-Bit Pack. [link]. Accessed: 2026-01-23.

Koesnaedi, A. e Istiono, W. (2022). Implementation drunkard’s walk algorithm to generate random level in roguelike games. International Journal of Multidisciplinary Research and Publications, 5(2):97–103. Acesso em: 27 jan. 2026.

Laurindo, E. K., Oti, K. H., Brocatto, M. d. A., e Belchior, V. (2024). Relatório técnico de desenvolvimento do jogo digital thorn ascent. Trabalho de conclusão de curso (curso superior de tecnologia em jogos digitais), Faculdade de Tecnologia de Americana “Ministro Ralph Biasi”, São Paulo. Acesso em: 27 jan. 2026.

Lazaridis, L. e Fragulis, G. F. (2024). Creating a newer and improved procedural content generation (pcg) algorithm with minimal human intervention for computer gaming development. Computers, 13(11):304. Acesso em: 27 jan. 2026.

Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W.-t., Rocktäschel, T., Riedel, S., e Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive nlp tasks. NIPS ’20, Red Hook, NY, USA. Curran Associates Inc.

Lindenmayer, A. (1968). Mathematical models for cellular interactions in development i. filaments with one-sided inputs. Journal of Theoretical Biology, 18(3):280–299. Acesso em: 27 jan. 2026.

Motion Twin (2018). Dead cells. Digital game.

Nasir, M. U., James, S., e Togelius, J. (2024). Word2world: Generating stories and worlds through large language models. arXiv preprint arXiv:2405.06686. Acesso em: 27 jan. 2026.

Pearson, K. (1905). The problem of the random walk. Nature, 72(1865):294. Acesso em: 28 jan. 2026.

Perlin, K. (1985). An image synthesizer. In Proceedings of the 12th Annual Conference on Computer Graphics and Interactive Techniques, pages 287–296, New York, NY, USA. Association for Computing Machinery. Acesso em: 27 jan. 2026.

Perlin, K. (2001). Noise hardware. In Olano, M., editor, Real-Time Shading SIGGRAPH Course Notes. ACM SIGGRAPH. Course Notes. Acesso em: 27 jan. 2026.

Salmela, T. (2022). Game development using the open-source godot game engine. Bachelor’s thesis (media and arts), Tampere University of Applied Sciences, Tampere, Finlândia. Acesso em: 27 jan. 2026.

Sepänmaa, T. (2024). Procedural level generation in 2d roguelite games. Bachelor’s thesis (business information systems), Tampere University of Applied Sciences, Tampere, Finlândia. Acesso em: 27 jan. 2026.

Shorten, C., Pierse, C., Smith, T. B., Cardenas, E., Sharma, A., Trengrove, J., e van Luijt, B. (2024). Structuredrag: Json response formatting with large language models. arXiv preprint arXiv:2408.11061. Acesso em: 27 jan. 2026.

Stiny, G. e Gips, J. (1972). Shape grammars and the generative specification of painting and sculpture. In Freiman, C. V., editor, Proceedings of Information Processing 71, pages 1460–1465, Amsterdam. North-Holland. Acesso em: 27 jan. 2026.

Sudhakaran, S., González-Duque, M., Freiberger, M., Glanois, C., Najarro, E., e Risi, S. (2023). Mariogpt: open-ended text2level generation through large language models. In Proceedings of the 37th International Conference on Neural Information Processing Systems, pages 54213–54227, Red Hook, NY, USA. Curran Associates Inc. Acesso em: 27 jan. 2026.

Team Cherry (2017). Hollow knight. Digital game.

Thorium Entertainment (2020). Undermine. Digital game.

Togelius, J., Yannakakis, G. N., Stanley, K. O., e Browne, C. (2011). Search-based procedural content generation: a taxonomy and survey. IEEE Transactions on Computational Intelligence and AI in Games, 3(3):172–186. Acesso em: 28 jan. 2026.

Yan, H., Hu, X., Wan, X., Huang, C., Zou, K., e Xu, S. (2023). Inherent limitations of llms regarding spatial information. arXiv preprint arXiv:2312.03042. Acesso em: 27 jan. 2026.