Implementations of Robot Visual Servo by Learning

This paper uses learning methods to acquire image features from eigen subspace transform and acquire the nonlinear relation between image features and robot control commands through a wavelet function neural network. The stages include data-sampling, model-learning, and robot visual servo. The servo stage consists of acquiring an image, transforming it into image features, computing robot joints by the trained neural network, communicating and controlling the robot to move, until the desired image features are achieved. Implementations of related algorithms are given. Experiments are carried out successfully on a real robot-vision system.