Collaborative Robotics
Research
Our group is a joint initiative of Bielefeld University and the Fraunhofer IOSB-INA in Lemgo. We are part of the research centre "Cognitive Interaction Technology" (CITEC) and the Faculty of Engineering at Bielefeld University. Our focus is on the intuitive use of collaborative robot systems in manipulation and assembly with the help of imitation learning. In particular, we concentrate on reconfigurable and modular robots, intelligent product transport and their neural control and regulation in safe interaction with human production participants. We pay particular attention to magnetic levitation systems, which represent a new trend in industrial automation and are being further developed as part of our current research work in the field of "planar robotics".
Magnetic Levitation
Our planar motor systems consist of mobile permanent magnets (“movers”) and static motor modules (“tiles”). During operation, the movers hover above the tiles and can be controlled in six dimensions by adjusting the currents in the coils embedded within the tiles. We are working on implementing electromagnetic control using neural networks. GPU acceleration is required to perform these calculations in real time. Motion planning and object manipulation remain active areas of research.
XTS
The XTS is a linear motor system frequently used in industry, for example in food packaging. The movers can be moved individually and centrally controlled along the rail system. Currently, the movements must be implemented through a time-consuming process. We are working on improving the coordination of the movers using machine learning. Other areas of research include, for example, wear detection.
T-IRMA
Transformer-based Imitation learning for Robot-driven Assembly Applications
Small and medium-sized enterprises (SMEs) are facing major challenges, primarily due to the shortage of skilled labour caused by demographic change. Flexible automation and robotics are promising solutions, but SMEs often lack the expertise and financial resources to implement them. Therefore, Simple, robust, cost-effective automation systems developed specifically for SMEs are needed. The project aims to develop imitation learning algorithms based on transformer architectures to transfer manipulative skills from experts to robotic assembly processes. These technologies should enable SMEs to utilise low-cost robotic cells that learn directly from demonstrations, significantly reducing development efforts.
Funded as part of the internal programmes of the Fraunhofer-Gesellschaft.
Grant number: SME 40-09551
enableATO
Rural areas urgently need new mobility concepts to fulfil the needs of the population. The "enableATO" project will make a significant contribution to this. This project, which is based at the RailCampus OWL in Minden and is part of the German Centre for Future Mobility (DZM), aims to enable sustainable and connected mobility through automated, rail-based mobility concepts.
Over the coming years, a broad consortium of universities, Fraunhofer institutes and companies will drive forward technologies that enable the efficient utilisation of existing infrastructure while improving the convenience and quality of services. Driverless transport systems such as the MONOCAB will be demonstrated in Minden and Extertal before being deployed in regular test operations.
Under the leadership of Thomas Günther and József Lurvig, we are actively involved in two important work packages (WP) of the project: WP 1.1.2 Environment and obstacle detection and WP 1.1.3 Terrain simulation.
Environment and obstacle detection is crucial for the safety and efficiency of automated rail transport. Our objective is to develop innovative technologies that enable the vehicle to record and interpret relevant information in real-time. This includes recognising obstacles along the tracks, identifying signals and adjusting speed according to the given environmental conditions. The successful realisation of these goals requires the seamless integration of various sensors, the performance of which must be validated on a test vehicle. The simulated data from the terrain simulation supplement the rich real-world data generated by the test drives. This data serves as the basis for developing an AI-based environment and obstacle detection.
