Robot-self-learning visual servoing algorithm using neural networks

A self-learning controller of a robot manipulator visual servoing system with a camera in hand to track a moving object is presented, where neural networks are involved in making a direct transition from visual to joint domain without requiring calibration. A technique, which uses monocular vision without explicitly estimating the visual depth, is also given in this paper. In this case, the visual sensory input is directly translated into joint accelerations. Simulation results show that this method can drive the static tracking error to zero quickly and keep good robustness and adaptability at the same time.