Tucker Balch, RI/CS Joint Faculty Candidate
Monday, April 26
3:30pm-4:30pm
Wean 4623


TITLE: Controlling Large-scale Multirobot Teams:
       From collective behavior to diversity


ABSTRACT:


Multiple robots offer significantly more reliable and efficient
solutions to real-world tasks than a single agent.  Reliability
is provided through redundancy.  Increased performance is achieved
by exploiting unique opportunities available in the multirobot
domain: cooperation, communication, selection of heterogeneous
solutions when appropriate and the integration of adaptive strategies
like reinforcement learning to a multiagent domain.  My research has
addressed all of these issues.  In this talk I will focus on the
design of cooperative behavior and the use of behavioral diversity
in multirobot teams.

In some tasks (e.g. soccer) diversity seems to be preferred, while
in others (such as foraging) homogeneous systems perform better.
But how do we quantify diversity?  How can we know which tasks are
appropriate for heterogeneous solutions, and how and when does
diversity arise in learning teams?  These are some of the challenges
that are addressed by the contributions of this work which include:
quantitative metrics of individual robot difference and multirobot
system diversity, a new multirobot configuration system integrating
behavior-based control and reinforcement learning and a research
platform supporting experiments in simulation and on mobile robots.
I will show how these tools have been used experimentally to evaluate
learning multirobot soccer and foraging teams.


Another contribution (jointly with Ron Arkin) is a new paradigm
for cooperative robot behavior called  "social potential fields."
The potential field approach is a well-known strategy for robot
navigation.  In this work we extend it to multiple robots so that the
potential field impacting a robot's path is shaped by the presence
of team or opponent robots.  This provides an elegant means for
specifying team strategies in tasks like foraging, soccer and
cooperative navigation.


Finally, I will close with a review of my current work at CMU,
involving the integration of planning and execution in multirobot
systems, and directions for future research.


BIO:

Tucker Balch earned the B.S. and Ph.D. degrees in Computer
Science at Georgia Tech in 1984 and 1998, respectively.
Prior to his work as a Research Assistant in Georgia Tech's
Mobile Robot Lab, he was a Computer Scientist at the Lawrence
Livermore National Laboratory (1984-1988) and an officer in the
U.S. Air Force where he accumulated over 1000 hours as
pilot-in-command of fighter aircraft (1988-1996).  Dr. Balch
was also a Member of the Technical Staff at NASA/JPL in the Robotic
Vehicles Group (1996).  He is currently a Postdoctoral Fellow
in the Computer Science Department at CMU.


Contact sriso@cs.cmu.edu to schedule a meeting appointment.