An image-based visual servoing method is developed for the position control of a robot with a vision sensor. For its successful applications to many robotic tasks, it is necessary to calibrate the robotic system parameters relevant to camera and hand-eye by itself. In this thesis, we present a new self-calibration method using vanishing points to compute all system parameters, including the extrinsic parameters as well as the intrinsic parameters through four deliberate robot motions.
This thesis also develops a feature Jacobian for controlling a six D.O.F. robot with stereo cameras and a feature extraction method based on line segments and their junctions for robust tracking. For an accurate grasping of the robot, a model-based set-point reconstruction method is used.
In a series of simulations and real experiments using a six D.O.F robot and a stereo camera system, the proposed algorithm which uses self-calibration and image-based visual servoing shows accurate and robust robot control. The fully integrated system shows a reliable grasping of an 3D object in an arbitrary position in space.