In this video I show how it is possible to use Self-Organizing Maps (SOM) for developing a representation of a robot body. I use the Aldebaran NAO to perform random movements of the head and then training the SOM to group the different poses. The SOM has 16x16 neurons and each neuron represent a possible head orientation. In the video the position associated with each neuron is represented with an arrow. At the beginning the arrows are in random position. After the first iterations the arrows are organized in clusters.
You can find the code on my repository:
github.com/mpatacchiola/pyERA
Personal webpage:
mpatacchiola.github.io/