Axis Motion Based System Calibration of an Automated Disassembly Work Cell
Abstract This paper presents an approach to calibrate a robotic cell consisting of a robot, a positioning table and a stereo vision system in an autonomous manner. The approach is designed to simplify the error relationships and parameter updates and thus eliminating the need for a large nonlinear search. The accumulation of error in the kinematic model is avoided by calibrating one joint at a time from the manipulator hand to the manipulator base. The error in the manipulator and sensor models are identified by using least squares estimates. The manipulator kinematic model is parameterized by the joint axes position and orientation instead of the Denavit-Hartenberg parameters. This approach leads to a more “user-friendly” interface to the calibration results. The model is derived using screw geometry, resulting in a simple relationship between the joint axis parameters and the path produced by moving a particular joint. The robot model provides an example of a chain of revolute joints while the positioning table provides an example of prismatic joints. Model simplifications result from each of these simplified motions. As with other methods, this formulation produces a four by four homogeneous transformation matrix which defines the motion of any point on the hand of the manipulator in terms of the sensed joint angles. It is shown that each camera can independently estimate the manipulators’ paths using the image data and distances along the path from the manipulator model. Error in position and orientation between the resulting two path estimates identify the relative error between the camera models. It is shown that a solution exists for any set of three or more points generated from one axis.