Gal A. Kaminka's Publications

Sorted by DateClassified by Publication TypeClassified by TopicGrouped by Student (current)Grouped by Former Students

On Integrated Multi-Agent Intention Recognition Systems

Nirom Cohen-Nov-Slapak. On Integrated Multi-Agent Intention Recognition Systems. Master's Thesis, Bar Ilan University,2008.

Download

[PDF]863.3kB  

Abstract

Intention recognition is the ability to reason and infer informationabout others, based on observations of their behavior.Unfortunately, to date there have been only a handful ofinvestigations into the integration of intention recognition intoagent architectures. In particular, there are open questions as tothe effect of intention recognition on the computational resourcesavailable to the agent. The issue is of particular importance inmodern virtual environments, where the agent may be interacting withmultiple other agents. This work tackles this question analyticallyand empirically. First, we examine run-time considerations, andoffer a novel view of plan-recognition as a sampling process. Underthis view, an existing work can be viewed as if trying to reduce thecomputational load on the agent by reducing the number of hypothesesit considers. In contrast, we reduce the Frequencyby which the recognition process is sampled. We provide ananalytical model allowing selection of a fixed-frequencyrecognition, and examine a number of heuristics fordynamically-changing plans. In the second part of the thesis, weconsider a method of integrating plan recognition processes, calledMirroring, in which the executable knowledge of theagent is re-used, as is, for recognition.

Additional Information

BibTeX

 @MastersThesis{nirom-msc,
author = {Nirom Cohen-Nov-Slapak},
title = {On Integrated Multi-Agent Intention Recognition Systems},
school = {{B}ar {I}lan {U}niversity},
year = {2008},
OPTkey = {},
OPTtype = {},
OPTaddress = {},
OPTmonth = {},
OPTnote = {},
OPTannote = {},
  wwwnote = {}, 
  abstract = { Intention recognition is the ability to reason and infer information
about others, based on observations of their behavior.
Unfortunately, to date there have been only a handful of
investigations into the integration of intention recognition into
agent architectures. In particular, there are open questions as to
the effect of intention recognition on the computational resources
available to the agent.  The issue is of particular importance in
modern virtual environments, where the agent may be interacting with
multiple other agents. This work tackles this question analytically
and empirically. First, we examine run-time considerations, and
offer a novel view of plan-recognition as a sampling process. Under
this view, an existing work can be viewed as if trying to reduce the
computational load on the agent by reducing the number of hypotheses
it considers. In contrast, we reduce the \emph{Frequency}
by which the recognition process is sampled. We provide an
analytical model allowing selection of a fixed-frequency
recognition, and examine a number of heuristics for
dynamically-changing plans. In the second part of the thesis, we
consider a method of integrating plan recognition processes, called
\emph{Mirroring}, in which the executable knowledge of the
agent is re-used, as is, for recognition.
 }
}

Generated by bib2html.pl (written by Patrick Riley ) on Sun Jul 23, 2017 22:08:49