Mapping with Fleet Data

 
The conventional creation of high precision digital maps require substantial investment in operating measuring vehicles. Constant changes in the environment, however, limit the usefulness of such maps if not kept updated equally frequently.

Many modern production vehicles are already equipped with some degree of driver assist and automation features that do rely on the detection of road properties like traffic lines, signs, lights and others. Additionally, radar sensors are common in such vehicles and can be used to perceive both non-road features and road features. Together with location data, fleet data can be obtained by the manufacturer.

Our work focuses on overcoming the lower precision of measurement obtained by production vehicles compared to dedicated measuring vehicles.

We aggregate individual trips using alignment either through radar point cloud or on a lane boundary detection level. Through this we are capable on providing both radar occupancy as well as actual lane boundary maps.

Publications

Alexander Blumberg, Jonas Merkert, and Christoph Stiller. Radar-based Pose Optimization for HD Map Generation from Noisy Multi-Drive Vehicle Fleet Data. arXiv preprint arXiv:2603.03453 (2026).

F. Immel, R. Fehler, M. M. Ghanaat, F. Ries, M. Haueis and C. Stiller. HD Map Generation from Noisy Multi-Route Vehicle Fleet Data on Highways with Expectation Maximization, 2023 IEEE Intelligent Vehicles Symposium (IV), Anchorage, AK, USA, 2023, pp. 1-7, doi: 10.1109/IV55152.2023.10186773.