Probabilistic Calibration and Catheter Tracking with Robotic Systems
Johns Hopkins University
A significant boost in robotics technology has been observed in recent years and more and more tasks are being automated by robots such as robotic surgery, autonomous driving, package delivery, etc. Not only has the precision of robots been improved, but the number of robots involved in a specific task has also grown in many scenarios. An important part in a robotic automated task involves the relative pose estimation among objects, and this often boils down to calibration and tracking. The dissertation begins with a robotic catheter tracking system and then focuses on calibration of robotic systems. The presentation first introduces a novel robotic catheter tracking system which uses an embedded active piezoelectric element at the tip of the catheter. Catheter intervention procedure is performed exclusively with X-ray, while ultrasound comes as an alternative modality which is radiation free. However, the catheter tip is usually very small and hard to be differentiated from human tissue in an ultrasound image. Moreover, an ultrasound photographer needs to hold the ultrasound probe during the procedure which can easily last for over an hour. The proposed system can tackle these issues using a robot arm and the active echo signal, and is, to the best knowledge of the author, the first robotic catheter tracking system using ultrasound. It is demonstrated in both the simulation and experiment that a robotic arm holding the ultrasound probe can track the catheter tip without image input. To better assist the tracking process, other procedures can be automated such as catheter insertion and phantom localization, etc. All these require introducing an extra robot and a precise calibration between robots and targets of interest. Out of many calibration approaches, the most classical one is called the hand-eye calibration problem formulated as AX = XB which takes in data from sensors in different locations to solve for an unknown rigid-body transformation. A generalization of this problem is the AX = YB robot-world and hand-eye calibration, where two unknowns need to be recovered simultaneously. The above two approaches mainly deal with the calibration of a single robot system. For multi-robot systems, a problem cast as the AXB = YCZ formulation arises where three unknowns need to be solved given three sensor data streams. The second portion of the presentation investigates in the probabilistic approaches toward all three problems above. Different methods based on the probabilistic theory on Lie group are developed to show their superior performance over non-probabilistic equivalents when there is partial knowledge of the correspondence among sensor data.
Robotics, Calibration, Catheter Tracking, Lie Group