Peter Stone's Selected Publications

Classified by TopicClassified by Publication TypeSorted by DateSorted by First Author Last NameClassified by Funding Source


Learning Complementary Multiagent Behaviors: A Case Study

Shivaram Kalyanakrishnan and Peter Stone. Learning Complementary Multiagent Behaviors: A Case Study. In Jacky Baltes, Michail G. Lagoudakis, Tadashi Naruse, and Saeed Shiry Ghidary, editors, RoboCup 2009: Robot Soccer World Cup XIII, pp. 153–165, Springer Verlag, 2010.
BEST STUDENT PAPER AWARD WINNER at RoboCup International Symposium.
Some simulations referenced in the paper.

Download

[PDF]200.5kB  [postscript]1.4MB  

Abstract

As machine learning is applied to increasingly complex tasks, it is likely that the diverse challenges encountered can only be addressed by combining the strengths of different learning algorithms. We examine this aspect of learning through a case study grounded in the robot soccer context. The task we consider is Keepaway, a popular benchmark for multiagent reinforcement learning from the simulation soccer domain. Whereas previous successful results in Keepaway have limited learning to an isolated, infrequent decision that amounts to a turn-taking behavior (passing), we expand the agents' learning capability to include a much more ubiquitous action (moving without the ball, or getting open), such that at any given time, multiple agents are executing learned behaviors simultaneously. We introduce a policy search method for learning ``GetOpen'' to complement the temporal difference learning approach employed for learning ``Pass''. Empirical results indicate that the learned GetOpen policy matches the best hand-coded policy for this task, and outperforms the best policy found when Pass is learned. We demonstrate that Pass and GetOpen can be learned simultaneously to realize tightly-coupled soccer team behavior.

BibTeX Entry

@incollection{LNAI09-kalyanakrishnan-1,
  author = "Shivaram Kalyanakrishnan and Peter Stone",
  title = "Learning Complementary Multiagent Behaviors: A Case Study",
  booktitle = "{R}obo{C}up 2009: Robot Soccer World Cup {XIII}",
  editor = "Jacky Baltes and  Michail G. Lagoudakis and Tadashi Naruse
 and Saeed Shiry Ghidary",
  Publisher="Springer Verlag",
  year = "2010",
  pages="153--165",
  abstract = {
    As machine learning is applied to increasingly complex tasks, it is
    likely that the diverse challenges encountered can only be addressed by
    combining the strengths of different learning algorithms.  We examine
    this aspect of learning through a case study grounded in the robot
    soccer context.  The task we consider is Keepaway, a popular benchmark
    for multiagent reinforcement learning from the simulation soccer
    domain.  Whereas previous successful results in Keepaway have limited
    learning to an isolated, infrequent decision that amounts to a
    turn-taking behavior (passing), we expand the agents' learning
    capability to include a much more ubiquitous action (moving without
    the ball, or getting open), such that at any given time, multiple
    agents are executing learned behaviors simultaneously.  We introduce a
    policy search method for learning ``GetOpen'' to complement the
    temporal difference learning approach employed for learning ``Pass''.
    Empirical results indicate that the learned GetOpen policy matches the
    best hand-coded policy for this task, and outperforms the best policy
    found when Pass is learned.  We demonstrate that Pass and GetOpen can
    be learned simultaneously to realize tightly-coupled soccer team
    behavior.
  },
  wwwnote={<b>BEST STUDENT PAPER AWARD WINNER</b> at RoboCup International Symposium.<br> Some <a href="http://www.cs.utexas.edu/~AustinVilla/sim/keepaway-getopen/">simulations</a> referenced in the paper.},
}

Generated by bib2html.pl (written by Patrick Riley ) on Fri Sep 05, 2014 12:17:34