EUVIMEDEUVIMED entdecken
Crossref

Reducing Reward Interference in PCGRL Through Multilayer Level Generation

Ken Kidokoro, Youichiro Miyake, Daiki Satoi, Sou Asei, Shinpei Sakata, Yuta Mizuno

Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment · 08.11.2026 · 22 (1) · 118-127

DOI: 10.1609/aiide.v22i1.43570 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, online erschienen am 08.11.2026. Verfasst von Ken Kidokoro und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Procedural Content Generation via Reinforcement Learning (PCGRL) is a reinforcement learning-based approach to level generation that treats level editing as sequential decision-making. Manually designed reward functions allow PCGRL to train generators without relying on datasets of existing content; however, conventional methods often assume a single-layer representation, which differs from the multilayer representations commonly used in game development. Consequently, when multiple objectives are aggregated into a single reward, an edit that improves one objective may degrade another, causing learning signals to interfere. This study proposes Multilayer PCGRL that decomposes levels into semantic layers and trains them progressively with layer-specific rewards. Experiments in two game environments with different characteristics show that the proposed method achieves a higher learning efficiency than a single-layer PCGRL.

Zitieren

APA

Kidokoro, K., Miyake, Y., Satoi, D., Asei, S., Sakata, S., & Mizuno, Y. (2026). Reducing Reward Interference in PCGRL Through Multilayer Level Generation. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment, 22(1), 118-127. https://doi.org/10.1609/aiide.v22i1.43570

Vancouver

Kidokoro K, Miyake Y, Satoi D, Asei S, Sakata S, Mizuno Y. Reducing Reward Interference in PCGRL Through Multilayer Level Generation. Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment. 2026;22(1):118-127. doi:10.1609/aiide.v22i1.43570

BibTeX
@article{Kidokoro2026,
  author = {Kidokoro, Ken and Miyake, Youichiro and Satoi, Daiki and Asei, Sou and Sakata, Shinpei and Mizuno, Yuta},
  title = {Reducing Reward Interference in PCGRL Through Multilayer Level Generation},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment},
  year = {2026},
  volume = {22},
  number = {1},
  pages = {118--127},
  doi = {10.1609/aiide.v22i1.43570}
}
Kostenlos starten →