Panel discussion on autonomous driving of the future in the context of “nxtAIM”
The German automotive industry is in the midst of a transformation process towards electromobility and software-based driving. Generative AI has the potential to take autonomous driving to a new level: Because it is only through its use that a safe, highly automated driving function is possible in any traffic context.
Representatives from politics and industry discussed this topic on Monday in a panel hosted by the Federal Ministry for Economic Affairs and Climate Action (BMWK). The occasion for the event was the project nxtAIM - generative methods for perception, prediction and planning - initiated by the German Association of the Automotive Industry (VDA) Lead Initiative for Autonomous and Connected Driving funded by the BMWK.
Prof. Dr. Astrid Lambrecht was present on behalf of Forschungszentrum Jülich - as the Jülich Supercomputing Centre (JSC) is directly involved in nxtAIM, which develops generative models for precisely this purpose. For one thing, the required computing capacity can be applied for and provided at JSC. Secondly, it contributes its own expertise with several scientists around Dr. Stefan Kesselheim (JSC) and Dr. Andreas Lintermann (JSC) to develop novel methods in the field of highly scalable AI on supercomputers. In addition to Ms. Lambrecht's participation in the panel discussion, Marcel Aach (JSC) was also at the event and took the opportunity to exchange ideas with other experts on the use of supercomputers for autonomous driving and AI.
Generative AI and the courage for innovation
In what way and how quickly will autonomous and connected driving become a reality in all traffic situations through the use of “generative AI”? Where do Germany and Europe stand in terms of AI applications in a global comparison and what contribution can industry, politics and research make together to keep Germany competitive and fit for the future? These and other questions were addressed in the 60-minute debate by Astrid Lambrecht, the Federal Minister of Economics Robert Habeck, board members from the industry - Mercedes-Benz CTO Markus Schäfer, Continental Automotive CTO Gilles Mabire, Bosch Mobility CTO Dr. Mathias Pillin, Valeo Germany CEO Holger Schwab, AVL EVP Jens Poggenburg and BIT TS Managing Director Cornelia Denk - as well as VDA Managing Director Dr. Marcus Bollig. They agreed that the ambitious goals can only be achieved together. This is because it requires innovative technology, fast computing, large data sets, financial support and a consensus on opportunities and risks: safety, it became very clear, is everyone's top priority. This includes both the safety of future AI-controlled cars on the road and the safety of the data.
Large AI models, such as those required for autonomous driving, require supercomputers of the highest performance level on which they can be trained. JUPITER, one of the most powerful AI computers in the world, is currently being built at Forschungszentrum Jülich. It will be used jointly by science and industry in this project. Close collaboration on this unique machine is the key to innovations that will keep Europe sovereign and competitive.”
Foundation models for the autonomous driving of the future
The core approach of nxtAIM is to create foundation models for autonomous driving. These are AI models that are trained on very broad data sets for a very long time with a general goal. This general training means that they can later be adapted to a variety of more specific tasks. Once trained, they can generate precisely fitting scenes that can be used to test the safety of existing systems in autonomous vehicles. Other applications in the field of autonomous driving are also possible.
The performance of generative AI models, especially foundation models, largely depends on two factors: The volume of data on which training is performed and the computing resources used to train the model. In nxtAIM, these two basic requirements for successful AI models are brought together: The participating industrial partners provide large data sets for training (while complying with data protection constraints). The required computing time can be made available at JSC upon application (peer review).
The JSC is contributing experience from numerous projects in the field of generative AI (e.g. “OpenGPT-X” and projects of the “Helmholtz Foundation Model Initiative”) to the project. In the nxtAIM project, it cooperates not only with the above-mentioned German industrial partners from the automotive sector, but also with leading German academic groups in computer vision and AI research, such as Prof. Brox's group at the University of Freiburg and Prof. Ommer's group at the Ludwig Maximilian University of Munich.
Using the unique infrastructure of the FZJ and exchanging ideas with AI authorities and key industry partners, innovative processes can be developed that will advance highly automated driving in Germany and Europe and potentially also find application in numerous other fields.
The event was streamed and is available online: https://www.youtube.com/live/9wqTUQj3kKE
Contact: Andreas Lintermann (JSC), Stefan Kesselheim (JSC)