Learning multiple arm postures of a real arm robot

Implementation of a self-organizing adaptive algorithm for visuo-motor control of a real arm robot (Rubbertuator) is reported. The algorithm supports the learning of multiple postures of the redundant arm, reproducing all possible postures for any arbitrary end-effector position of the arm in the three-dimensional workspace. The computational algorithm accomplishes three different artificial neural network components, responsible for visual and motor maps. These components are interconnected, forming a synergetic neural structure where the arm postures are memorized. Learning multiple arm postures is essential to guide the arm end-effector in its workspace. Positioning error of the robot's end-effector is discussed.