Mutual modeling of teammate behavior
Jelle R. Kok and Nikos Vlassis. Mutual modeling of teammate behavior. Technical Report IAS-UVA-02-04, Informatics Institute, University of Amsterdam, 2002.
Download
Abstract
In multiagent systems the action that an agent chooses depends on the state of other agents. It is therefore important to have an accurate representation of the current world state. In case the world is only partially observable, an agent does not continually perceive the state of the other agents. It is then essential to model the actions of the other agents to predict their future state. In this paper, we introduce ``Mutual Modeling of Teammate Behavior'' (MMTB). This model simulates the executed action of a teammate. In combination with the known world dynamics this action can be used to determine the teammate's future state. In this model, we assume that all teammates are homogeneous and follow a policy that is common knowledge among them. Furthermore, we will describe how MMTB is implemented in our RoboCup soccer simulation team UvA Trilearn and give empirical results about its effectiveness.
BibTeX Entry
@TechReport{Kok02ias04,
author = {Jelle R. Kok and Nikos Vlassis},
title = {Mutual modeling of teammate behavior},
institution = {Informatics Institute, University of Amsterdam},
address = {The Netherlands},
year = {2002},
month = aug,
number = {IAS-UVA-02-04},
postscript = {2002/Kok02ias04.ps.gz},
pdf = {2002/Kok02ias04.pdf},
abstract = {In multiagent systems the action that an agent
chooses depends on the state of other agents. It is
therefore important to have an accurate
representation of the current world state. In case
the world is only partially observable, an agent
does not continually perceive the state of the other
agents. It is then essential to model the actions of
the other agents to predict their future state. In
this paper, we introduce ``Mutual Modeling of
Teammate Behavior'' (MMTB). This model simulates the
executed action of a teammate. In combination with
the known world dynamics this action can be used to
determine the teammate's future state. In this
model, we assume that all teammates are homogeneous
and follow a policy that is common knowledge among
them. Furthermore, we will describe how MMTB is
implemented in our RoboCup soccer simulation team
UvA Trilearn and give empirical results about its
effectiveness.}
}
Generated by bib2html.pl (written by Patrick Riley) on Tue Oct 31, 2006 19:33:42 UTC