Self-organizing maps for robot control (original) (raw)

Lecture Notes in Computer Science, 1997

Abstract

Many tasks in robotics are difficult to tackle with explicit models, based on “first principles”. Consequently, neural networks with their inherent learning ability can offer feasible alternatives to more traditional approaches. Focusing on neural network approaches that have been derived from the Self-Organizing Map, we review several examples, how such networks can contribute to the solution of typical tasks in robotics, such as map building, object recognition and, in particular, the coordination of multi-joint movements. We argue, that a frequent common structure is a “continuous associative memory” for the flexible representation of continuous relations between degrees of freedom, and show how such representation can be obtained with a parametrized Self-organizing Map.

Klaus P. hasn't uploaded this paper.

Let Klaus know you want this paper to be uploaded.

Ask for this paper to be uploaded.