(Dr. Harvey Lipkin, advisor)
"Model Independent Offset Tracking with Virtual Feature Points"
This thesis builds on a dynamic quasi-Newton controller by Piepmeier that guides an uncalibrated robot with uncalibrated cameras. The controller minimizes error vectors sensed on the image place between target and end-effector features. The end-effector cannot be positioned to Cartesian offsets from the target since there are no physical features to sense on the image plane. Instead, “virtual feature points” are used to generate the image error vectors.
Virtual feature points are specified on an end-effector moving coordinate frame. Their image positions are calculated using sensed end-effector features as a reference. Virtual feature points make it possible by the dynamic quasi-Newton controller to track a target at desired offsets. Using a series of offsets, the approach path of the end-effector with respect to the target can be controlled for a given task.
A sufficient number of features on the end-effector are identified to determine the moving frame to image plane transformation in every image cycle. As a result, if the cameras are disturbed during the operation, errors are not propagated to successive cycles. In addition, as the target gets closer to the end-effector, the accuracy of the camera model becomes less important for calculating the precise positions of virtual feature points on the image plane.