Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in

BCCN AG-Seminar

Tuesday, 13.02.2007 16 c.t.

About self-referential dynamical systems and playful robots

by Prof. Dr. Ralf Der
from Universität Leipzig

Contact person: J. Michael Herrmann


Seminarraum Haus 2, 4. Stock (Bunsenstr.)


Dynamical systems are a powerful tool for both understanding and controlling the behavior of agents in the sensorimotor loop. We consider robots in closed loop control realized by a neural network. Our aim is the emergence of a self-motivated and playful behavior of complex robots in unstructured and highly dynamical environments. Self-motivated means that the adaptation of the controller (parameter dynamics) is formulated in terms of the system dynamics alone without any recourse to external information. This is what we call a self-referential dynamical system. Technically the parameter dynamics is given as the gradient flow on the so-called time loop error. The talk presents both new theoretical results on self-referential dynamical systems and the a number of videos demonstrating the emergence of behavior with high sensorimotor coordination with robots with up to 20 active degrees of freedom. The results may be useful for developmental robotics in the early sensorimotor stage.

back to overview