This research try to use single camera on robot to estimate three dimensional positions. The main idea of this research is moving machine arm terminal symmetrically, according to the central point of camera. It causes parallax on vision system. We use the parallax from previous step to estimate the three dimensional positions of the machine arm. Furthermore, using Kalman filter to reduce the estimated error. Finally, we combined any angle of machine arm and the coordinate of vision system, using Back-propagation Neural Network lessons three dimensional position of machine arm. We can inference the machine arm terminal position instantly.