Authors:
Achille Morenville, Eric Piette

Venue:
Rencontres des Jeunes Chercheurs en Intelligence Artificielle (RJCIA, PFIA), 2024

Topics:
general game playing, imperfect-information games, reinforcement learning, game description languages

Links: PDF · Conference page

Imperfect-information games remain a major challenge for General Game Playing agents due to hidden information and the need for reasoning under uncertainty.

This work proposes to extend the Ludii system by improving game descriptions to include hidden information, adopting a modelling framework closer to reinforcement learning, and adapting search algorithms to better exploit game descriptions.

These improvements aim to enable the simulation and analysis of a wide range of imperfect-information games without relying on domain-specific knowledge.

This paper outlines early research directions toward extending general game systems to handle imperfect information in a principled and domain-independent way.

It highlights limitations of existing approaches, including reliance on handcrafted knowledge and difficulty in generalisation across games.

The proposed ideas contribute to the broader objective of developing general and human-like agents, and connect directly to later work such as the Belief Stochastic Game model.

Morenville, A., Piette, E. (2024). Vers une Approche Polyvalente pour les Jeux à Information Imparfaite sans Connaissance de Domaine. In RJCIA (PFIA).

@inproceedings{morenville2024rjcia,
  author    = {Morenville, Achille and Piette, Eric},
  title     = {Vers une Approche Polyvalente pour les Jeux à Information Imparfaite sans Connaissance de Domaine},
  booktitle = {RJCIA (PFIA)},
  year      = {2024},
  url       = {https://pfia2024.univ-lr.fr/Pr%C3%A9sentations/RJCIA---Pr%C3%A9sentation-2.4/}
}