Short Videos




Pointing with a Robot: Leaning the Visuo-motor Map


1991 © J. Walter (03:43)

Learning the robot coordination map from visual to joint coordinates using a SOM neural network architecture; Beckman Institute, UIUC, Illinois, USA
Details in:
"Implementation of self-organizing neural networks for visuo-motor control of an industrial robot". IEEE TNN, 4(1):86-95, 1993 by Jörg Walter and Klaus Schulten, more...


Predicting Timeseries with Self-Organizing-Maps:

Mackey-Glass Time Series


1992 © J. Walter (03:00)

The first uses of SOM neural networks for the prediction of a time series, here representing deterministic chaos. In the video the dots represent the reference vectors of the SOM neurons in the first 3 dimensions of the time delay embedding. After random initialization, the dots represent the attractor of the Mackey-Glass time series, with the embedding time delay parameter Delta=6. The data generation process used the chaos determination parameter tau=16 (quasi periodic curve resembling the double looped double filament) and tau=17 (the attractor gets chaotic with fractal dimension 2.1). More details in Dimplomarbeit (in german).


Robot Vision Checker Player


1995 © J. Walter, Neuroinformatics Group, University of Bielefeld (06:27)

Playing Checkers with a Robot. See detail in "Service Object Request Management Architecture: SORMA concepts and examples" by Jörg Walter and Helge Ritter. Technical Report SFB360-TR-96-3, 1996. more...


3D Visual and Force Tracking Robotarm


1996 © J. Walter (01:27)

Unimate PUMA classical industrial robot follows a ball. An end-effector mounted camera and a "wrist" 6D-force-torque sensor are used for the tracking algorithm. Pulling and pushing is sensed at the wrist, therefore the reaction is soft below and stiff at the arm above the sensor. more...