EfEcient Motor Learning by Self-organizing Implicit Linear Transformations
Jiirgen Monnerjahn · 1994
The subject of this paper are training algorithms for robot control with selforganizing feature maps (SOFM). This work is based on the research of Ritter et al. [RMS91] who used an extended SOFM concept to make a three-joint robot arm system learn its inverse kinematics by visually supervised trial movements. A good familiarity with their algorithms and simulation environment in [RMS91] is necessary to understand this texf. The disadvantages of their approach are the great need for computational power and the necessity of vey many (several thousand) trial movements. This paper presents algorithms developed to reduce the computational cost and the number of trial movements to a minimal amount. The most efficient algorithm only needs about one trial movement per neuron to reach an optimal training result.