David Brogan
University of Virginia
Yannick Loitiere
University of Virginia
International Joint Conference on Autonomous Agents and Multiagent Systems, 2002
Abstract
The dynamics of how groups move through space to accomplish
common goals must be understood to create realistic
synthetic environments. One potential method for creating
such multiagent behaviors is to replay prerecorded examples
of group movements. While these data-driven methods effectively
capture the original performance for a particular
instance, the success of these methods for interactive, multiagent
applications is limited by the large number of potential
agent movements that must be prerecorded. To mitigate
the scaling effects of data-driven multiagent behavior algorithms,
we propose a behavior model that reduces the dimensionality
of prerecorded data and decreases the amount
of data required by effectively using available data. We have
chosen to investigate the sport of simulated soccer and have
developed behaviors for simulated soccer players from the
data acquired from recent RoboCup games.
Paper
|
|
|