Skip to Main content Skip to Navigation
Theses

Une contribution à la résolution des processus décisionnels de Markov décentralisés avec contraintes temporelles

Abstract : This thesis deals with distributed multiagent decision-making under
uncertainty. We formalize this problem with Decentralized Markov
Decision Processes (DEC-MDP) which extends Markov Decision Processes
(MDP) to multi-agent settings. Even if DEC-MDPs describe an expressive
framework for cooperative multiagent decision, they suffer from a high
complexity and fail to formalize constraints on task execution.
Despite the wide variety of approaches to solve DEC-MDPs, computing
a solution for large problems remains a serious challenge even for
approximation approaches.

We develop an approach that can solve large problems, and that can
deal with more complex time and action representations. We therefore
define a class of DEC-MDP, OC-DEC-MDP, that allows us to consider
several possible durations for each task taking into account
constraints on task execution. Having considered the representation
of the problems we deal with, we turn to OC-DEC-MDP resolution. Our
purpose is to develop an efficient planning approach that computes
each agent's policy even for large missions. Given the high
complexity of finding an optimal solution, we aim at computing an
approximate solution. We also split the multiagent decision problem
into a set of MDPs. For purposes of coordinating the agents, we then
introduce the notion of Opportunity Cost.
Document type :
Theses
Complete list of metadata

Cited literature [94 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00112014
Contributor : Hal System <>
Submitted on : Tuesday, November 7, 2006 - 8:55:27 AM
Last modification on : Tuesday, February 5, 2019 - 12:12:10 PM
Long-term archiving on: : Tuesday, April 6, 2010 - 7:09:40 PM

Identifiers

  • HAL Id : tel-00112014, version 1

Citation

Aurélie Beynier. Une contribution à la résolution des processus décisionnels de Markov décentralisés avec contraintes temporelles. Informatique. Université de Caen, 2006. Français. ⟨tel-00112014⟩

Share

Metrics

Record views

216

Files downloads

480