Kinematic Calibration of Serial Manipulators Using Relative Measurements

Abstract

Kinematic calibration is necessary to enhance the accuracy of robotic manipulators. It is typically desired to perform this task in both a cost-effective and time-efficient manner. Many techniques exist in the literature that achieve both goals. In this thesis, a novel model-based kinematic calibration method using relative measurements is developed and implemented, which has proven to be useful for optical measurements, and is capable of achieving the same level of accuracy as the typical absolute measurement methods. Furthermore, the effects of measurement noise, the number of measured poses, and the best pose configurations for kinematic calibration are investigated. Both a Thermo CRS A465 and a 7 DOF WAM Arm were used in this thesis. The results for the WAM Arm lead to a significant improvement in the end-effector pose accuracy. The implication is that the relative measurement concept is a valid tool for model-based kinematic calibration of serial manipulators. iii iv Table of Contents v List of Tables ix List of Figures xi Appendix B Links Velocities Representations 122

Kinematic Calibration of a Serial Robotic Arm Using a Linear Movement Constraint

“…Following this, the relative measurement technique described in [34] can be used to determine the other DH parameter offsets. This is done by solving:…”

“…Calibration is performed by observing this board using a CCD camera mounted on the robot's end effector. A similar technique is used in [34], which is discussed further in Subsection 2. 3…”

“…Additionally, a stereo laser system is used to measure the distance from the straightedge to the tool. In [34], both relative position and orientation are measured. This is done using a camera mounted on the robot's end effector and a reference object that can be recognized by analyzing the images from the camera.…”

“…The values of these synthetic errors were assigned such that the error in joint angle was between -1 and 1 degree. This approximate value was chosed based on the value of several similar experiments such as [34]. The error in link twist was assigned to be less than 10% of this value.…”