12:00, Wed 3 December 1997, WeH 7220

Title: Behavioral Diversity in Learning Robot Teams

Tucker Balch
Mobile Robot Laboratory
Georgia Institute of Technology

Abstract:

Research  indicates that just as in the case of individual agents,
multi-robot  teams  benefit  from the adaptability and reliability
of  reinforcement  learning  techniques.   This  presentation will
focus  on new  issues arising  from learning in multi-agent teams.
Of particular interest is how agents arrive at team  solutions for
tasks,  whether  the  individuals  specialize  or  settle on homo-
geneous  policies,  and how  the  choice of reinforcement function
impacts  diversity  of  the  team.   As  tools  for this research,
metrics  of robot  behavioral difference  and behavioral diversity
are proposed.   Behavioral difference  refers to disparity between
two  specific agents,  while diversity  is a measure  of an entire
society.   The  presentation  will  conclude  with a discussion of
results using these metrics in simulated robot soccer and foraging
tasks.


Biography:

Tucker Balch is a  graduating Ph.D. student in Intelligent Systems
at Georgia Tech.   He has investigated a number of topics in auto-
nomous  robotics at the  Mobile  Robot Laboratory, including: nav-
igation, communication  and cooperation in behavior-based systems,
formation for multi-robot teams and learning in multiagent systems.
Tucker  was a Computer  Scientist  at Lawrence  Livermore National
Laboratory  from  1984-1988.   In  1996  he  was  a  Member of the
Technical  Staff  in  the  Robotic  Vehicles  Group at  NASA's Jet
Propulsion Laboratory.  He is currently a pilot in the Georgia Air
National Guard.