Abstract
We describe a new algorithm for vector quantization and control. The algorithm, in addition to generating a discrete representation of input data by means of Voronoï polyhedra, also generates a tesselation associated with these polyhedra. The tesselation corresponds to a graph which connects neighbouring Voronoï polyhedra and, hence, reflects neighborhood relationships of the embedding space of the data. The algorithm can be extended to approximate through ‘training’ arbitrary functions defined on the data points. The tesselation allows one to speed up the ‘training’ through cooperative learning involving nearest, next-nearest, etc. Voronoï polyhedra, reducing the range of cooperation progressively during training. The algorithm produces a table look-up program, assigning optimally tables to inputs and generating rapidly optimal table entries. The entries can be complex data structures, e.g., combinations of scalars, vectors, and tensors. The use of the algorithm has been demonstrated for time series prediction, surpassing existing algorithms, and for visuo-motor control of an industrial robot, e.g., for precise end effector position control. We will attempt to demonstrate by the time of the lecture also an application of the algorithm for visuo-motor control of a pneumatically driven robot arm, a Bridgestone ‘RUBBERTUATOR’. This light-weight robot, capable of compliant motion, can be operated in contact with humans. The presented algorithm can acquire the complex response characteristics of this arm through training and, thereby, allows accurate and swift control of pneumatic robot motion.
Preview
Unable to display preview. Download preview PDF.
Similar content being viewed by others
References
Neuronale Netze: Eine Einführung in die Neuroinformatik selbstorganisierender Abbildungen, H. Ritter, Th. Martinetz, and K. Schulten (2nd enlarged edition, Addison-Wesley, Bonn, 1990 ) Neural Computation and Self-Organizing Maps: An Introduction, H. Ritter, Th. Martinetz, and K. Schulten (revised, English edition, Addison-Wesley, New York, 1991 )
Industrial Robot Learns Visuo-Motor Coordination by Means of ‘Neural Gas’ Network, J.A.Walter, Th.M.Martinetz, and K.Schulten, Proceedings of the International Conference on Artificial Neural Networks, Helsinki, 1991)
Neural Network with Hebbian-like Adaptation Rules Learning Control of a PUMA Robot, Th.Martinetz and K.Schulten, (submitted to NIPS-91)
Hierarchical Neural Net for Learning Control of a Robot’s Arm and Gripper, Th.Martinetz and K.Schulten, IJCNN International Joint Conference on Neural Networks, San Diego, California, July 1990, pp. 11–747 to 11–752 ( The Institute of Electrical and Electronics Engineers, New York, 1990 )
A ‘Neural Gas’ Network Learns Topologies, Th.Martinetz and K.Schulten, Proceedings of the International Conference on Artificial Neural Networks, Helsinki, 1991)
S. Berkovitch, Th, Martinetz and K. Schulten, application to time series prediction, manuscript in preparation
Ph. Dalger, B. Noël, Th. Martinetz and K. Schulten, mathematical analysis of Delaunay tesselation on the basis of 1st and 2nd order Voronoï polygons in case of 2-dimensional task spaces, manuscript in preparation
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 1991 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Berkovitch, S. et al. (1991). Vector Quantization Algorithm for Time Series Prediction and Visuo-Motor Control of Robots. In: Brauer, W., Hernández, D. (eds) Verteilte Künstliche Intelligenz und kooperatives Arbeiten. Informatik-Fachberichte, vol 291. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-76980-1_41
Download citation
DOI: https://doi.org/10.1007/978-3-642-76980-1_41
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-54617-7
Online ISBN: 978-3-642-76980-1
eBook Packages: Springer Book Archive