In flexible SMEworkshops, workpieces are often prepared manually without CAD data due to “lot size 1 production”. Using a robot like the welding robot assistant, the workspace is scanned in 3D with a time-of-flight camera to recognise individual workpieces. Welding seams are detected between the workpieces and are illustrated directly onto the workspace using projectors. The worker can use touch screens or gestures to reorder, parameterise and refine the sequence of welding seams.
Using learning algorithms such as Hidden Markov Models, the sequence and parameters for various types of welding seams are learned and continuously improved. Online sensor-based control enables the welding system to handle the large uncertainties in the setup. Large uncertainties and exceptions are handled by interacting with the worker and strategies are learned for improving the capabilities of the robot system.