23. Juni 2020, 11:06 Uhr | Andreas Knoll
Applications such as the assembly of printed circuit boards require high positioning accuracy. Here you can see a detailed view.
The aim of a new research and development project on AI-based robot calibration (KI-basierte Roboter-Kalibrierung, KIRK) is to develop new software-driven calibration methods for industrial robots through machine learning to increase their accuracy.
Initiators of the joint project are the University of Stuttgart, the DHBW Karlsruhe and the robotics expert and software manufacturer ArtiMinds Robotics.
Industrial robots normally carry out their tasks reliably and precisely. To ensure this accuracy, the systems have to be individually recalibrated at regular intervals. This is cost- and time-intensive and means considerable additional work, especially for SMEs. In addition, more and more low-cost robot arms are coming onto the market, which can cause considerable inaccuracies in positioning for mechanical reasons.
With currently available calibration methods, only geometric errors can be corrected. Temperature or load-dependent inaccuracies, for example, can only be compensated insufficiently. Recalibration during operation, which would be important for a sustainable optimization process, is also not feasible.
In order to close these gaps and to develop new software-driven calibration methods for practical use through machine learning, the robotics expert ArtiMinds Robotics, the University of Stuttgart and the DHBW Karlsruhe have now launched the KIRK AI project. "The possibility of automated data acquisition and analysis reduces the effort for users and makes it easier, especially for SMEs, to build up the necessary competence to make optimal use of a robot system," explains Darko Katic, technical contact person for the KIRK project and ArtiMinds team leader for Artificial Intelligence.
The aim is to increase the accuracy with the help of software in order to be able to use robots flexibly for a wide range of applications, to simplify workflows with a solution that is independent of the robot type and manufacturer, and to save time for skilled personnel. "The basis for making the complex interrelationships of external factors as well as the time-varying characteristics of the individual robot controllable and thus increasing the positioning accuracy is formed by deep neural networks, i.e. deep learning," explains AI researcher Prof. Marco Huber from the Institute for Industrial Manufacturing and Factory Operation (IFF) at the University of Stuttgart.
The IFF and the Robot-and-Human-Motion-Lab (RaHM-Lab) of the Baden-Wuerttemberg Cooperative State University in Karlsruhe will carry out basic research in the project. Together with ArtiMinds Robotics as industrial partner, the results will be transferred to real industrial applications. Finally, the newly developed methods will also be incorporated into the Robot Programming Suite (RPS) programming software. The project is scheduled to end in spring 2022.