Authors:
Achille Morenville, Eric Piette

Venue:
Computer and Games (CG), 2024

Topics:
imperfect-information games, general game playing, knowledge representation, belief states

Links: PDF · ACM · Springer

Imperfect-information games are a major challenge for General Game Playing agents because they require reasoning under hidden information.

This paper introduces the Belief Stochastic Game model, a framework that externalises state estimation from the agent to the game model itself, allowing agents to focus directly on strategy development.

By exploiting common structures found in many imperfect-information games, the model provides a more general and standardised approach for reasoning about hidden information and supports the design of more adaptable agents.

This paper extends your research beyond perfect-information games and addresses one of the major open challenges in game AI: how to build more general agents for games with hidden information.

The proposed Belief-SG model shifts the burden of belief-state construction away from the agent and into the formal game model, which is particularly relevant for general-purpose reasoning and search.

The work is also an important step toward future integration of imperfect-information games into broader general game systems such as Ludii, and toward applications in areas such as historical card games and interdisciplinary game studies.

Morenville, A., Piette, E. (2024). Belief Stochastic Game: A Model for Imperfect-Information Games with Known Positions. In Computer and Games (CG).

@inproceedings{morenville2024beliefsg,
  author    = {Morenville, Achille and Piette, Eric},
  title     = {Belief Stochastic Game: A Model for Imperfect-Information Games with Known Positions},
  booktitle = {Computer and Games (CG)},
  year      = {2024},
  url       = {https://dl.acm.org/doi/10.1007/978-3-031-86585-5_14}
}