Probabilistic plan recognition in multiagent systems

Suchi Saria, Sridhar Mahadevan

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

We present a theoretical framework for online probabilistic plan recognition in cooperative multiagent systems. Our model extends the Abstract Hidden Markov Model (AHMM) (Bui, Venkatesh, & West 2002), and consists of a hierarchical dynamic Bayes network that allows reasoning about the interaction among multiple cooperating agents. We provide an in-depth analysis of two different policy termination schemes, T all and T any for concurrent action introduced in (Rohanimanesh & Mahadevan 2003). In the T all scheme, a joint policy terminates only when all agents have terminated executing their individual policies. In the T any scheme, a joint policy terminates as soon as any of the agents terminates executing its individual policy. Since exact inference is intractable, we describe an approximate algorithm using Rao-Blackwellized particle filtering. Our approximate inference procedure reduces the complexity from exponential time in N, the number of agents and K, the number of levels, to time linear in both N and K̂ ≤ K (the lowest-level of plan coordination) for the T all termination scheme and O(N log N) and linear in K̂ for the T any termination scheme.

Original languageEnglish (US)
Title of host publicationProceedings of the 14th International Conference on Automated Planning and Scheduling, ICAPS 2004
EditorsS. Zilberstein, J. Koehler, S. Koenig
Pages287-296
Number of pages10
StatePublished - Dec 1 2004
Externally publishedYes
EventProceedings of the 14th International Conference on Automated Planning and Scheduling, ICAPS 2004 - Whistler, BC, Canada
Duration: Jun 3 2004Jun 7 2004

Publication series

NameProceedings of the 14th International Conference on Automated Planning and Scheduling, ICAPS 2004

Other

OtherProceedings of the 14th International Conference on Automated Planning and Scheduling, ICAPS 2004
CountryCanada
CityWhistler, BC
Period6/3/046/7/04

ASJC Scopus subject areas

  • Engineering(all)

Fingerprint Dive into the research topics of 'Probabilistic plan recognition in multiagent systems'. Together they form a unique fingerprint.

Cite this