KI Data Tooling

In the KI Data Tooling research project, a data solution for the training and validation of artificial intelligence (AI)-based automated driving functions is being developed. For a successful deployment of AI-based functions in vehicles, a well-founded database is essential, which must include as many scenarios as possible as well as data of different sensor modalities. So far, only isolated solutions exist for the data of individual sensor types, a small number of scenarios or parts of the training and evaluation chain. Therefore, the research project examines tools and methods that automatically process, supplement and combine data from different sensor modalities (camera, lidar, radar) and subsequently make them available for AI-based functions. Real data as well as synthetic data are considered in an integrated way and the efficiency potential arising from the combination is used.

In the KI Data Tooling project real, synthetic and augmented data of different sensor modalities shall be generated. Furthermore, methods for the analysis, quality assessment, and efficient as well as resource-saving storage and transmission of these data will be developed. An optimized AI training strategy will be built based on the data and on methods.

In the KI Data Tooling research project, the FZI Research Center for Information Technology is responsible for the real data augmentation as well as for evaluating and demonstrating the project results. The methodological focus of the FZI researchers are novel approaches for context identification and corner case detection as well as methods for augmentation and automated labeling of real data. In addition, research is being conducted on methods for the synthetic generation of realistic radar data, which currently represent a particularly great challenge in synthetic data generation.