Team reinforcement learning: Robocup theorems
Speaker: Dr Mark Pendrith, SRI International
When: 2003-01-30 16:00:00
Venue: 78-420
Host: Professor Peter Lindsay
Abstract:The study of reinforcement learning for the special case of a single
learning agent represents a relatively mature body of research
within the field of machine learning. On the other hand, our
understanding of the dynamics of agents learning in competitive and
co-operative group situations is much less well-developed. In this
talk, I present some novel theoretical results that have practical
import for cooperative team reinforcement learning. These theorems
can be naturally applied and understood in the context of a team of
soccer-playing learning agents, hence I refer to these results as
"Robocup theorems".
Biography:Mark Pendrith received his Ph.D. in Computer Science and
Engineering from the University of New South Wales in 1998. His
thesis topic studied the application of reinforcement learning to
situations where only partial or otherwise imperfect state
information was available to the learning agent, motivated by the
problem of learning gait-control for a physical six-legged robot.
From late 1997 through to late 1999 he was employed as a research
scientist by the Daimler-Benz Research and Technology Center, Palo
Alto, California, where he researched the application of multi-agent
reinforcement learning to traffic flow optimisation problems. From
late 1999 to early 2002 he was a research engineer at SRI
International, Menlo Park, California, where he was co-principal
investigator on the DARPA/ITO MARS (Mobile Autonomous Robotics
Systems) program, and also joint technical lead and champion for the
internally funded Janus project, conducting research in the area of
pervasive/ubiquitous computing. Since returning with his family to
live in Australia, he directs research part-time at the
Indooroopilly Research Institute, and is also a visiting fellow at
UNSW.
Type: ITEE Seminar
Contact:Professor Peter Lindsay, seminar host (pal@itee.uq.edu.au)
or Guido Governatori (ITEE seminar co-ordinator)
(guido@itee.uq.edu.au)