Society For Neuroscience, San Diego, November 2001
Program Number:
295.9 Day /
Time: Monday, Nov. 12, 8:00 AM - 9:00 AM
DYNAMIC FEATURES OF NEURAL ACTIVITIES IN THE LAMPREY'S BRAINSTEM: A STUDY
WITH A NEURO-ROBOTIC SYSTEM
A.Karniel*1; K.M.Fleming1; V.Sanguineti2;
S.T.Alford3; F.A.Mussa-Ivaldi1
1. Department of Physiology, Northwestern University Medical School,
Chicago, IL, USA; 2. Dipartimento di Informatica Sistemistica e Telematica,
Universita' di Genova, Genova, Italy; 3. Department of Biological Sciences,
University of Illinois at Chicago, Chicago, IL, USA
In order to study learning mechanisms, we developed a research tool that
includes the brainstem of a lamprey and a two-wheeled robot that are
interconnected in a closed loop (Artificial Life 6:307,2000). Two electrodes
applied stimulations to the axons of the octavomotor nuclei. The stimulations
rates were proportional to the light intensity measured by sensors on the right
and left sides of the mobile robot. The velocity command to the right and left
wheels was proportional to the population spike rates recorded by two
electrodes in the spinal cord. In each trial, one of five lights was turned on
and in most cases, the robot moved roughly toward the light.
We fitted various two-input/two-output neural network models and observed the
generalization error. The outputs were modeled with two units that received
polynomial functions of the two inputs. Two families of networks were
considered: (i) a static neural network model, where the output at each time
step was a function of the excitations in the previous time step; and (ii) a
dynamic network model, where the output was a function of the previous
excitation and of the previous output. We found that a dynamic model was
significantly superior to a static model even when the number of parameters was
smaller. We discuss the results in the context of possible cellular mechanisms
that might explain the behavior of this neuro-robotic preparation.
Supported by: ONR grant #N00014-99-1-0881 and AASERT grant #N00014-97-1-0714