Programmation et apprentissage bayésien pour les jeux vidéo multi-joueurs, application à l'intelligence artificielle de jeux de stratégies temps-réel

Abstract : This thesis explores the use of Bayesian models in multi-player video games AI, particularly real-time strategy (RTS) games AI. Video games are an in-between of real world robotics and total simulations, as other players are not simulated, nor do we have control over the simulation. RTS games require having strategic (technological, economical), tactical (spatial, temporal) and reactive (units control) actions and decisions on the go. We used Bayesian modeling as an alternative to (boolean valued) logic, able to cope with incompleteness of information and (thus) uncertainty. Indeed, incomplete specification of the possible behaviors in scripting, or incomplete specification of the possible states in planning/search raise the need to deal with uncertainty. Machine learning helps reducing the complexity of fully specifying such models. We show that Bayesian programming can integrate all kinds of sources of uncertainty (hidden state, intention, stochasticity), through the realization of a fully robotic StarCraft player. Probability distributions are a mean to convey the full extent of the information we have and can represent by turns: constraints, partial knowledge, state space estimation and incompleteness in the model itself. In the first part of this thesis, we review the current solutions to problems raised by multi-player game AI, by outlining the types of computational and cognitive complexities in the main gameplay types. From here, we sum up the transversal categories of prob- lems, introducing how Bayesian modeling can deal with all of them. We then explain how to build a Bayesian program from domain knowledge and observations through a toy role-playing game example. In the second part of the thesis, we detail our application of this approach to RTS AI, and the models that we built up. For reactive behavior (micro-management), we present a real-time multi-agent decentralized controller inspired from sensory motor fusion. We then show how to perform strategic and tactical adaptation to a dynamic opponent through opponent modeling and machine learning (both supervised and unsupervised) from highly skilled players' traces. These probabilistic player-based models can be applied both to the opponent for prediction, or to ourselves for decision-making, through different inputs. Finally, we explain our StarCraft robotic player architecture and precise some technical implementation details. Beyond models and their implementations, our contributions are threefolds: machine learning based plan recognition/opponent modeling by using the structure of the domain knowledge, multi-scale decision-making under uncertainty, and integration of Bayesian models with a real-time control program.
Liste complète des métadonnées

Cited literature [157 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00780635
Contributor : Abes Star <>
Submitted on : Thursday, January 24, 2013 - 3:22:13 PM
Last modification on : Thursday, October 11, 2018 - 8:48:02 AM
Document(s) archivé(s) le : Thursday, April 25, 2013 - 3:53:26 AM

File

29588_SYNNAEVE_2012_archivage....
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-00780635, version 1

Collections

Citation

Gabriel Synnaeve. Programmation et apprentissage bayésien pour les jeux vidéo multi-joueurs, application à l'intelligence artificielle de jeux de stratégies temps-réel. Informatique et théorie des jeux [cs.GT]. Université de Grenoble, 2012. Français. ⟨NNT : 2012GRENM075⟩. ⟨tel-00780635⟩

Share

Metrics

Record views

5991

Files downloads

4454