Below you will find video demonstrations of some of my research. The
videos show the CMUnited and CMDragons RoboCup soccer teams, in
competition, interacting with humans, and learning using the
techniques I developed for my dissertation. All the videos are
recorded at real time, playback speed may vary depending on the
machine and software.
NOTICE: The videos successfully play under linux using
mplayer. The MPEG-1 (lower resolution) videos play under Windows
Media Player, but MPEG-2 videos do not. Enjoy!
Adversarial Robot Learning
This is a video of my most recent work applying WoLF and policy
gradient techniques to an adversarial, simultaneous, robot learning
problem. The robot at the top of the video is trying to get inside
the circle on the field, while the closer robot is trying to stop it.
Learning was carried out in simulation for 2000 trials, and on the
robots for an additional 2000 trials. The video shows the resulting
policies after learning.
MPEG-2, 75.1MB, 720x480, ~2m
MPEG-1, 30.4MB, 360x240, ~2m
CMDragons Interacting with People...
CMDragons'02 is CMU's most recent RoboCup small-size team. I was
highly involved in the development of this team, responsible for
motion control, tactical behavior, and strategic team behavior through
a play-based, adapting team architecture. We advanced to the
quarterfinals at RoboCup 2002. The video below shows this team in
action against a human playing with a hockey stick. Although our
robots can't see the woman or the hockey stick, they can still put up
a strong defense.
Windows Media, 25.6MB, 320x240, ~5m
This is is a 4 minute video giving an overview of CMUnited '98, the
RoboCup small-size robot champion team. I was the team leader and
also narrate the video.
MPEG-2, 142MB, 640x480, ~4m
MPEG-1, 103MB, 320x240, ~4m
"Multiagent Learning in the Presence of Limited Agents",
Job Talk, 2003.
"Learning, Equilibria, Limitations, and Robots",
NIPS 2002 Workshop on Mulitagent Learning.
"Rational and Convergent Learning of Mixed Equilibria in Stochastic Games",
UAI 2000 Workshop: Beyond MDPs.
"Convergence Problems of General-Sum Multiagent Reinforcement Learning",