Research partnership between BMW Group Plant Dingolfing, TH Deggendorf and b-plus, funded by the Bavarian Ministry of Economic Affairs
Artificial intelligence supports factory quality assurance of driver assistance systems +++ Analysis and simulation of driving data for more variance +++
Dingolfing. From Plattling via Aiterhofen to Pilsting and back: in the middle of Lower Bavaria, BMW Group Plant Dingolfing, together with the Deggendorf Institute of Technology and b-plus technologies, wants to set another milestone in automated driving. The research project is generating data and test kilometers on roads between Deggendorf, Straubing and Landshut as well as at BMW Group Plant Dingolfing in order to make automated driver assistance systems in BMW models even more attractive and to validate them at the factory. The special feature: Artificial intelligence extends the regionally experienced data mathematically many times over, making it possible to cover a very broad spectrum of driving situations and vehicle equipment/configurations - and thus support development in Munich. Many more test drives, both regionally and worldwide, can be dispensed with.
Insgesamt soll das Projekt rund 1,3 Mio. Euro kosten. Weil DaDriVe das Potenzial hat, den Standort Bayern zu stärken und die Digitalisierung vorantreibt, beteiligt sich auch der Freistaat Bayern mit einem Zuschuss von mehreren hunderttausend Euro. „Forschungsprojekte wie DaDriVe fördern die Digitalisierung, stärken die Kompetenz und das Knowhow im zielgerichteten, sinnvollen Einsatz Künstlicher Intelligenz. So sichern wir qualitativ hochwertige Arbeitsplätze in der Region und stärken unsere Wirtschaftskraft. Wir unterstützen diese Forschung daher sehr, denn sie ist essenziell für die Zukunft unseres Bundeslandes“, erklärt Tobias Gotthardt, Staatssekretär im Bayerischen Staatsministerium für Wirtschaft, Landesentwicklung und Energie.
Digital methods for the automated driving of the future
Data Driven Vehicle Validation, or DaDriVe for short, is the name of the project with which the BMW Group vehicle plant in Dingolfing, the Deggendorf Institute of Technology and the system specialist b-plus want to research and further develop the use of artificial intelligence for the quality assurance of driver assistance systems. Within three years, a major step towards the more efficient safeguarding of automated driving functions is to be achieved. "We are investing in research at the plant in order to further optimize the quality of the driver assistance systems in our premium vehicles using new digital methods," says Christoph Schröder, Head of BMW Group Plant Dingolfing, explaining the goal.
The project focuses on three key innovations that build on each other. The aim is to use modern AI methods to visualize and digitally record even minimal deviations in real driving, to artificially generate additional driving data on this basis using AI and to expand these methods to improve the protection of the systems. Furthermore, synthetic scenarios are to be generated in order to incorporate various weather and environmental conditions into the validation process. Schröder: "For us as a production plant, these are steps towards an even more digital future. For our customers, they primarily mean greater convenience and safety."
Key innovation 1: Anomaly detection using modern AI methods
It's a bit like having a highly sensitive passenger in the vehicle. A passenger who registers even the slightest movement of the steering wheel in an automated vehicle, directly senses a slightly abrupt braking movement and also perceives minimal acceleration. Who can always predict how the vehicle should behave in a certain situation - and therefore knows immediately if something is different. A co-driver who can then pass on all these perceptions in detail to make the vehicle's behavior even better in the future.
Of course, there is no other person in the vehicle during a research project for digital measurement methods. Instead, measuring devices record a wide range of vehicle movement data during test drives on normal roads - from speed and steering angle to braking behavior. The algorithm then uses this data to detect deviations and implausible vehicle behavior, i.e. anomalies. Just like the highly sensitive co-driver, only much more precise. And: based on existing parameters, the algorithm can predict how the automated driving vehicle should normally behave in the sequence of certain situations, so that it also registers the slightest deviations here. All of this data is extremely valuable for the analysis and start-up suitability of driver assistance systems at the BMW Group plant in Dingolfing. "We deliberately chose the test track so that a lane change, braking or acceleration sometimes doesn't run as smoothly and gently as the BMW customer would expect and in line with our self-image of premium quality," says project manager Gerald Sagmeister.
Key innovation 2: AI-based generation of artificial data and data augmentation
In order to obtain the widest possible range of results in addition to the findings from the real data and also to train the recognition algorithm even better, the DaDriVE project is developing methods to generate this data artificially. To do this, the AI extrapolates the measured data and constructs additional new data sets that deviate slightly from the original. In this way, the system creates any number of additional situations from a situation measured in reality, in which the driver assistance system brakes inexplicably abruptly, for example. The vehicle safety system can thus investigate driving behavior at different speeds or road conditions without the situation having to occur in real life. "Advanced data analysis with AI is particularly interesting for our research," says Prof. Thomas Limbrunner, TH Deggendorf, explaining the approach, which also saves many kilometers of driving. "We can't always rely on a special situation that we want to investigate actually occurring. Using AI, however, we can generate this situation mathematically - and then evaluate it optimally."
Key innovation 3: Use of synthetic scenarios for vehicle protection
The third key innovation also deals with special situations and scenarios. From the weather to road signs, the team can intervene in vehicle communication to simulate environmental conditions that are different from those actually present at the time of driving. For example, the team plans to feed data into the test vehicle that signals to the assistance system that it is in the USA. This will allow the correct functioning of the system to be tested with the design of traffic lights there, as well as the recognition of road markings. The system's behavior in heavy rain can also be tested, even though the sun is shining.
This form of simulation improves the predictability of test drives and is particularly important for export. The BMW Group's development departments can thus optimally adapt the driver assistance systems to country-specific approval requirements without having to run a test on site beforehand. "Our aim is to make it possible to validate defined environmental conditions more quickly and with less effort as part of the validation of startup suitability," explains Dr. Franziska Wutz, Data Scientist and DaDriVe project manager at b-plus technologies.
Proven partners
BMW Group Plant Dingolfing, Deggendorf Institute of Technology and b-plus technologies successfully completed a joint research project to safeguard automated driving functions back in 2021. "The knowledge we gained back then will of course help us enormously for the new project," explains Sagmeister. "In DaDriVe, however, we are taking a completely new approach by making greater use of AI and using the new methods specifically for our factory requirements due to the variety of vehicle variants and special equipment."
