Collecting, storing and processing data from real-life traffic scenarios is time-consuming, costly and consumes substantial amount of energy. Additionally, real-world data often exhibits errors, inaccuracies, or distortions that can adversely impact Machine Learning . Hazardous edge cases in traffic are infrequent in real data, but are crucial for verifying AI models. Synthetic data can be employed to complement and expand real data, creating relevant datasets for AI training.

Against this backdrop, novel technologies are required to extract highly accurate, characteristic and fair datasets of high quality from the extensive data generated during driving. The primary objective of the jbDATA project is to develop and implement AI-based methods and tools to efficiently collect data and with high accuracy. A novel aspect is that jbDATA situates data evaluation at the Edge, i.e. in the capturing vehicle. Instead of uploading all data to the cloud as done traditionally, the recorded data is preprocessed, evaluated, and sorted for further processing and storage using the Smart Data Loop, also to be developed in the project: Data is anonymized, analyzed for data gaps, and, where necessary, enriched with synthetic data using methods also to be developed in the project.

Test vehicles in the project are used to exemplary record a dataset, which is intended to serve as the foundation for a comprehensive, industrially usable dataset. Additionally, jbDATA plans to make the dataset collected in the project widely accessible for research beyond the consortium.