Donnersdorf, 16/12/2021. The research project being carried out by the German Federal Ministry for Economic Affairs and Energy called KI Delta Learning is halfway completed. To mark the occasion, the project partners involved took stock of the progress made so far and presented it to the public at a virtual event on 07/10/2021.
The project is part of the flagship initiative by the German Association of the Automotive Industry (VDA) for autonomous and connected driving, which aims to help the automotive industry expand and develop particularly in the fields of AI and self-driving vehicles. 18 institutions and companies are cooperating to develop new methods and tools to improve the broad usage of innovative technologies and facilitate the implementation of AI.
The focus lies in resolving one of the current challenges of systems working with artificial intelligence: the changing environmental conditions, which entail fundamental changes to the equipment in the vehicle. A typical example of this is the country-specific differences in traffic rules, signs, and routes, which the applications have to cope while remaining fully reliable. Until now, handling this was labour-intensive and high-cost, since the changeover required retraining the AI system from the ground up.
That’s why the project aims to minimise training costs. The technology must be able to adapt to the different situations and develop a better understanding of the environment with no need for additional, complicated development work. This transferability is meant to be optimised, making the use of the technology more diverse and reliable.
In the course of the process so far, the participating partners are already developing procedures for learning the deltas in various fields of application. According to reference models and detailed data records in different environments, it was possible to perform an initial assessment. Using this information, complete methods for scaling the AI approaches for automated driving can now be developed by the time the project is completed.
InnoSenT GmbH is one of the technology providers for the project and is making its automotive radar sensors available for this purpose, which were installed in a test vehicle. Together with other built-in sensors (LIDAR, camera), the vehicle environment is recorded during defined journeys. As part of the project, the company is also tackling the technical challenges of generating artificial radar data using a simulator.
In order to lower costs, sensors with a small number of antennas are increasingly being installed in vehicles. Due to the reduced number of antennas, ambiguous returns often occur during target detection, which results in lacking and inaccurate detections. That’s why InnoSenT has set itself the task of artificially increasing the number of antennas by means of a neural network determining these additional channels. This enables better predictions about objects to be made and ambiguous returns to be reduced, despite using smaller, and thus more cost-effective, hardware. Both the artificially generated and the recorded radar data from the test drives can be used as training data.
For more detailed information about the functioning principle and the evaluation that has been carried out so far, there is a file from InnoSenT available on the project website. There you will also find posters showing the results from other project partners, further aspects of project set-up, and official press releases. They are available at: https://www.ki-deltalearning.de/mid-term-presentation-marketplace
##
Words: 551
Publication only with acknowledgement of source and consent of InnoSenT GmbH. For printable images or more information about this topic feel free to contact the InnoSenT Company. InnoSenT GmbH is pleased to receive a voucher copy.
About InnoSenT GmbH
InnoSenT GmbH from Donnersdorf was founded in 1999 and is one of the world's leading companies in the field of radar technology. As a manufacturer and developer, the company supplies the full bandwidth of engineering services - from customer-specific development to series production. Thanks to the strong focus on quality and innovation, InnoSenT GmbH has been following a course to global success for many years. More information can be found at www.InnoSenT.de