System integration

Considering that the role of the vision system should be to determine the deburring profile and to measure the length and depth of burrs, as burrs in this kind of machining process are characterized by depth/length of 1-2 millimeters, the accuracy of the vision system should be in the order of tens of millimeters. If the profiles detected during subsequent experiments are compared on the basis of the image plane coordinates only the intrinsic camera parameters affect the precision; instead if the 3D coordinates of the profile are compared, the accuracy will be affected by the extrinsic camera parameters as well.
At this point, the accuracy of the self-calibration procedure used to determine the extrinsic camera parameters is not well enough to guarantee the required accuracy in the reconstruction of the 3D burr profile. A better self-calibration algorithm has thus been developed and is currently under validation.
Further, it must be considered that a reconstruction of the profile in the 3D space will be very useful, as it will give more information to the robot control system: the robot will be able to execute the path previously determined measuring the contact force (that gives an indirect measure of the burr length/depth) and thus deriving more information for the learning algorithm. On the other hand it will not be strictly necessary, as the profile reconstructed in the image plane should be sufficient to measure the burr characteristics during subsequent deburring experiments (and this represents the minimal amount of information required by the learning system).
Here can be found two video of a scan from the image processing side and from the robot side.

Learning the deburring path

A paper concerning the study developed to learn a deburring path has been submitted to ICRA 2013.