Talk: Self-organizing robots


SIIT Image and Vision Computing Laboratory Seminar

Abstract:

Self-organization and the phenomenon of emergence which play an
essential role in living systems form a major challenge to artificial
life systems. This is not only because systems become more life like
but also since self-organization may help in reducing the design efforts in
creating complex behavior systems. The talk exemplifies our general
approach to the self-organization of behavior which has been developed
and tested in various examples in recent years. The robots dispose of a
neural network controller and a world model which is also realized by a
neural network. Our general paradigm of minimizing the so called time
loop error is used in order to learn both the model and the controller
concomitantly from scratch. We apply this approach to different robots
with complicated physical properties which are completely unknown to
the system. Nevertheless after some time the robots develop behaviors which
are both body and environment related in a completely self-organized
way. The applications are demonstrated by several videos of wheeled
robots, a “rocking stamper”, a spherical robot and various snake like
artifacts. More information on our video page
http://robot.informatik.uni-leipzig.de/research/videos/.


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.