Self-organizing maps for the kinematics problem of a simulated two-arm robot

A self-organizing map is studied in terms of its relationship with the kinematics of a simulated two-arm robot operating in 2D workspace. A two-dimensional array of neural units will be used for the map. Each unit in the map is responsible for a particular region in the workspace. An input from the workspace activates a unit in the map. The activation of such unit causes it to transmit its output values to the joint motors and these lead the end effectors to the target. This method is motivated by the reflex mode present in the control of human arms. It will be shown that after the learning process, the input space is topographically represented on the competitive layer and finally on the output space. Both supervised and unsupervised SOM\u27s are studied, and the results for 2D workspace are extended to 3D. Simulations results are presented and analyzed