Having so far relied on the synthetic data of its sister project KI Absicherung, KI Data Tooling will soon also produce its own synthetic data material. While the focus is currently still on tool optimization and lessons learned from KI Absicherung, a start will be made in November on determining the scene requirements for real and synthetic data on a Bosch test field.
This step is necessary in order to be able to better evaluate potential sensory challenges. In the next step, technically more challenging scenes, such as intersection scenarios, will be added from January 2022 of which an accurate digital twin will then be generated.
The aim is to generate approximately 100,000 frames synhetically within this project phase. In contrast to KI Absicherung, where a similar amount of frames was generated, in KI Data Tooling about half of the synthetic data will correspond to so-called Data Pairs. Those Data Pairs refer to synthetic data that replicates real data. The overall goal of this work is to verify how well the recorded real data can be reproduced synthetically and, above all, to learn what is crucial for an AI when it comes to interpreting this data correctly.
Moreover, in line with the VDA Leitinitiative, the synthetic data generated in this context will be defined as open source as soon as they are generated and thus made available to the public. This is intended to enable broad participation by other researchers.
Image: BMW / BIT Technology Solutions