AI tames test data volumes for automated driving
During tests of automated vehicles, large amounts of data are generated. Reducing these data volumes to save storage space, power and evaluation effort, but at the same time condensing the information to make the vehicles safer is the goal of the recently launched KIsSME project. Algorithms based on artificial intelligence (AI) select the data during driving and sort them into scenario catalogues.
Automated vehicles have numerous sensors with which they record information about their own status and from their environment. Based on this information, they must make reliable driving decisions in the shortest possible time. During their testing, each vehicle variant has to cover millions of kilometres and master many different scenarios that combine infrastructure, weather and other road users and their behaviour. This generates data volumes of up to 8 terabytes per vehicle and day. Handling such data volumes poses great challenges for the test and development engineers of the vehicle manufacturers. One approach to reducing these data volumes is to create catalogues of driving scenarios and to sort in newly occurring scenarios during vehicle testing.
This approach allows testers to record only those data during driving that actually add value. Scientists from various research institutions and companies are investigating how this could work in the joint project KIsSME. The acronym stands (in English translation) for “Artificial Intelligence for the selective near-real-time recording of scenario and manoeuvre data during the testing of highly automated vehicles”. For this purpose, scientists are developing AI algorithms that select the data already during driving operation. The project project relates to automated driving of levels four to five – i.e. autonomous cars for specific driving scenarios as well as fully autonomous robot taxis, which no longer even have the usual control elements. “KIsSME aims to expand the scenario catalogue and at the same time reduce data volumes,” explains Michael Frey, deputy institute director at the Institute of Vehicle Systems Engineering (FAST), at the Kalrsruhe Institute of Technology (KIT). According to the scientist, this saves storage space and electricity and reduces the effort for evaluation and data protection.
The KIT researchers provide data from real driving tests as well as from simulations for the project. For this purpose, measurement runs are being carried out in public urban traffic and on the Autonomous Driving Baden-Württemberg (TAF BW) test field in Karlsruhe, as well as closed vehicle-in-the-loop simulations on a complete vehicle test bench at KIT. In addition, the researchers are testing the AI models and AI selectors developed in the project by applying the algorithms developed by the collaborative partners to the data from tests and simulations.
The KIsSME joint project is coordinated by AVL Deutschland GmbH. In addition to KIT, partners include the Fraunhofer Institute for High-Speed Dynamics, the Ernst Mach Institute, the FZI Research Centre for Information Technology and Robert Bosch GmbH. The German Federal Ministry for Economic Affairs and Energy (BMWi) is funding the project with a total of 6.5 million euros. KIsSME started at the beginning of 2021. The results are to be presented in three years.
More information: https://www.mobilitaetssysteme.kit.edu/
Related articles:
AVL, Rohde & Schwarz join forces for XiL tests systems
Jointly researching AI for automated driving in cities
Study quantifies value of data from the connected car
Artificial Intelligence and future directions for ETSI
“Software will become the decisive factor for the coming years”
Deep learning method improves environment perception of self-driving cars
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
