EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Interagent Beliefs for Learning to Communicate in Large-Scale Multirobot Visual Object Search

Jernej Puc, Gašper Škulj, Jan Pleterski, Primož Podržaj, Rok Vrabič

IEEE Transactions on Robotics · 2026 · 42 · 1729-1748

DOI: 10.1109/tro.2026.3677944 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Robotics, 2026. Verfasst von Jernej Puc und 4 weiteren Autor:innen. Die Arbeit nennt 54 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: The Slovenian Research and Innovation Agency.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

We present Century Maze, a simulated environment for cooperative navigation of groups numbering over 100 robots with primarily visual observations. The underlying task of large-scale object search is designed to push the scalability and efficiency of decentralized multirobot systems relying on communication to overcome the limitations of separate units. To enable the training of robotic agents in large-scale visual embodied applications such as Century Maze, we propose two reinforcement learning (RL) components that facilitate the emergence of effective communication: Differentiable Interagent Belief Learning (DIABL), which provides a clear learning signal by optimizing the agents' beliefs about their goals and propagating gradients back over a differentiable communication channel to include the appropriate observers, and informative event replay (infer), which compensates for the scarcity of samples relevant to DIABL by maintaining a dedicated buffer for auxiliary learning. We demonstrate our combined method through an actor-critic neural network architecture encompassing a decentralized multiagent system of variable size with individual goal-based belief modules. Results on Century Maze show that our method is able to significantly improve performance through cooperation over 100 agents against comparable baselines and maintains its effectiveness in realistic conditions.

Förderung

  • The Slovenian Research and Innovation Agency

Zitieren

APA

Puc, J., Škulj, G., Pleterski, J., Podržaj, P., & Vrabič, R. (2026). Interagent Beliefs for Learning to Communicate in Large-Scale Multirobot Visual Object Search. IEEE Transactions on Robotics, 42, 1729-1748. https://doi.org/10.1109/tro.2026.3677944

Vancouver

Puc J, Škulj G, Pleterski J, Podržaj P, Vrabič R. Interagent Beliefs for Learning to Communicate in Large-Scale Multirobot Visual Object Search. IEEE Transactions on Robotics. 2026;42:1729-1748. doi:10.1109/tro.2026.3677944

BibTeX
@article{Puc2026,
  author = {Puc, Jernej and Škulj, Gašper and Pleterski, Jan and Podržaj, Primož and Vrabič, Rok},
  title = {Interagent Beliefs for Learning to Communicate in Large-Scale Multirobot Visual Object Search},
  journal = {IEEE Transactions on Robotics},
  year = {2026},
  volume = {42},
  pages = {1729--1748},
  doi = {10.1109/tro.2026.3677944}
}
Kostenlos starten →