Mixed-integer linear programming for transition-independent decentralized MDPs

  • Authors:

  • Jianhui Wu;Edmund H. Durfee

  • Affiliations:

  • University of Michigan, Ann Arbor, MI;University of Michigan, Ann Arbor, MI

  • Venue:
  • AAMAS '06 Proceedings of the fifth international joint conference on Autonomous agents and multiagent systems
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

Markov decision process (MDP) provides a framework for computing optimal policies for individual agents operating in uncertain environments. However, extending single-agent MDP techniques to multiagent problems is not straightforward. Previous complexity analyses have shown that the general decentralized Markov decision process (Dec-MDP) is NEXP-complete, which means that optimally solving a Dec-MDP is extremely difficult. The class of problems studied in this paper is a subclass of Dec-MDP in which two or more cooperative agents are tied together through the rewards of completing joint tasks but the actions taken by one agent do not impact other agents' transitions. Although this reduces the complexity class to NP-complete [4], efficiently solving such transition independent Dec-MDPs is still nontrivial.