“TÜV” certification for Autonomous Cars - TÜV SÜD and DFKI develop platform for AI modules in autonomous vehicles

TÜV SÜD, a German organization for technical monitoring, and the German Research Center for Artificial Intelligence (DFKI) are joining forces to create Genesis - an open, non-proprietary platform for continuous validation of AI modules in autonomous vehicles. Just like the "body" of the vehicles (the bodywork, the engine, and all other physical components), the "brain" (the AI ​​modules) will be tested in the future, so that consumers can be sure that the products supplied by the industry are suitable for road traffic and safe.

Genesis LogoThe name stands for the genesis of a new generation of intelligent systems that will influence the daily lives of many people in the future as much as television or the Internet did.

On the way to the final platform for autonomous driving, some technical challenges still have to be mastered, which DFKI will address as part of the BMBF-funded research project REACT. TÜV SÜD supports research and development in REACT and, together with DFKI, prepares the introduction. Genesis will then be accessible to other partners who would like to participate in this innovation. The Genesis consortium is also involved in the newly established AI standardization committees of DIN and ISO.

The time to realize the "Safe Autonomous Driving" is pressing. Together with the DFKI, TÜV SÜD has now committed itself to the mission of certifying the necessary AI modules for the future vision of autonomous driving. The focus lies on the open platform Genesis. "DFKI, the largest artificial intelligence research center, is a strong partner in building Genesis. With its Competence Center Autonomous Driving (CCAD) and the activities in the standardization committees, we see ourselves together on the best way to secure autonomous mobility ", says Dr. Houssem Abdellatif, Global Head Autonomous Driving at TÜV SÜD.

In addition to providing validation scenes, Genesis also offers training material that includes high coverage of critical traffic situations. This is central to the protection of vulnerable road users (VRUs, pedestrians, cyclists, skaters, etc.). Only if this succeeds, autonomous driving can be realized as planned.

Of all AI methods, deep learning is currently most important for generating applications in autonomous driving. Essentially, these are neural networks with very many levels ("deep"), which are trained on a huge amount of examples. This is achieved by using synthetic data in addition to real data collected through test drives. Synthetic data is needed because critical traffic scenarios are too rare and too different to be covered by real data. This is the focus of research work at the Department of Agents and Simulated Reality of DFKI. Creating data requires research and development in the following areas:
• Modeling behavior based on intention (e.g., how pedestrians behave when willing to reach a bus that is already turning the corner);
• Modeling movement from behavior (e.g., how pedestrian move approaching the curb, knowing that they will stop);
• Rendering of sensor data from motion (for example, what does the scene described above look like from the perspective of a front radar?).

Beispielsszenerie auf der Basis synthetischer Daten zur Verbesserung und Validierung von KI-Algorithmen für das autonome Fahren-

Sample scenario based on synthetic data to improve and validate AI algorithms for autonomous driving.

Learning models from synthetic data, which are based on real data and allow testing in virtual space, is a specialty of the DFKI Research Unit Agents and Simulated Reality (ASR) under the direction of Prof. Philipp Slusallek. In this role, Slusallek has been working since 2008 to integrate Artificial Intelligence for relevant innovative issues into practice, such as autonomous systems, Industry 4.0, visualization, cooperative work, 3D worlds and applications, et cetera. His department also houses the new Competence Center Autonomous Driving, which bundles all DFKI activities in the field of AI technologies for autonomous vehicles. The automotive expert Dr. Christian Müller has recently been appointed as head of CCAD.

Within the framework of the recently launched BMBF-funded project REACT (autonomous driving: modeling, learning and simulation environment for pedestrian behavior in critical traffic situations) the basic foundation of Genesis will be built within the next three years. The overall objective of REACT is a systematic, secure and validatable approach to the development, training, and use of digital reality to ensure the safe and reliable operation of autonomous systems - especially in critical situations. For this purpose, methods and concepts of machine learning - in particular deep learning and (deep) reinforcement learning (RL) - are used to learn low-dimensional submodels of the real world. In this way, the whole range of existing critical situations should be identified so that they can be simulated in virtual space.

DFKI is, in addition to the content efforts to bring the AI ​​into practice, involved in defining standards, such as IEEE, DIN, and ISO. Also at the European level, DFKI participates in the call for proposals for an artificial intelligence-on-demand platform.


Prof. Dr. Philipp Slusallek
Head of Research Department Agents und Simulated Reality
German Research Center for Artificial Intelligence (DFKI)
Campus D3 2
66123 Saarbrücken, Germany
Email: Philipp.Slusallek@dfki.de
Phone: +49 681 85775 5276