12:00, Wed Apr 12, WeH 1327

		   "Teaching Robots to Play Soccer"

			     Peter Stone

Abstract:  Robotic soccer is emerging as a new and challenging problem
for AI researchers and for roboticists.  Several researchers are
studying the problems that this complex, hostile, real-time
environment presents.  In fact, if all goes as planned, there may be a
robotic soccer World Cup tournament at IJCAI '97 in Japan!  In this
talk, I will give a quick overview of some of the current research in Robot
Soccer, particularly from the angle of Reinforcement Learning.  Then I
will describe two separate Machine Learning experiments that I have
worked on in a simulated environment.  The first is an adversarial
task in which an agent learns to score against a moving defender.  The
second is a cooperative task in which an agent learns to shoot a
moving ball into a goal after it has been passed from a teammate.

If all goes as planned, there will be a live computer demo for your
edification and entertainment.