Computer Vision Approach for the Automated Tool Alignment of an Orbital Sanding Robot

Abstract

In automated production lines for manufacturing complex geometries like car bodies, automated sanding processes based on compliant robots are used to achieve time and cost efficient production while fulfilling high quality constraints. However, the replacement of sanding paper fixed to an orbital tool using velcro is difficult to automate. In this contribution, a measurement setup and evaluation procedure using Computer Vision is presented. It detects position and orientation of the orbital tool from a single photo (captured by a calibrated monocular setup). To achieve the required accuracy, distortion introduced by the camera system is removed, and significant and robust features are detected. Considering imperfect scene illumination found on a real production site, all available contours are combined to a single solution space, which allows the robust detection of position and orientation, even if the target is partially occluded. In this contribution, the required Computer Vision algorithms are outlined, and promising measurement results are presented.