Multiple self-organizing maps to facilitate the learning of visuo-motor correlations

This paper presents an application of bi-directional neural modularity: a chaining of several self-organizing maps (SOM) is used to represent the motor and sensorial position correlations of a robotic platform. Two active cameras follow the movements of a robot manipulator in 3-D space. The mapping of image positions and camera orientations into arm angular joint positions can be learned by a neural network. However, decomposing the problem and using several neural networks turns out to be a better way. In our approach, the neural modules do not need to be adapted independently. Based on the principle of bi-directionality, the modular architecture can be adapted globally, using the sensor-motor data directly.