A scheme for visual tracking of robot manipulator using neural network

The authors describe a control scheme and a strategy for a robotic manipulator using visual information to track a moving object. The proposed system directly integrates the visual data into the control process without calculating a transformation from world coordinate to workpiece coordinate, and without solving the inverse kinematics of the manipulator. Neural networks are used for learning the reproduction of the nonlinear relationship between image data and control signal in the joint angle space to achieve the desired pose. After they have finished learning such a task, the neural networks are used as a controller to track a moving object