Precise industrial robot positioning and path-tracking over large surfaces using non-calibrated vision

This paper presents a methodology for precise robot positioning and path tracking, performed by industrial robots over large surfaces of arbitrary size, shape and orientation. The methodology is based on a vision-based, calibration-free robot control method known as camera-space manipulation. The path, defined and stored in a CAD file, is later adapted to the curved, work-surface by using a mapping procedure. When applied to large surfaces, the precision of the positioning and path-tracking maneuvers depends on several factors like the resolution of the cameras per unit physical space and the mapping procedure, which may introduce distortion specially in the case of non-developable surfaces. In order to reduce the influence of camera-resolution, this paper presents two alternatives: the use of multiple cameras and the application of cameras mounted over non-expensive, sensorless pan/tilt units. In the case of the distortion produced by the mapping procedure, the paper discusses several options like a modified geodesic mapping and virtual projection. The proposed techniques were tested multiple times over flat and deformed surfaces, by using a large work-envelope, industrial robot.