The fifth consortium meeting of the just better DATA research project took place at the beginning of July – this time at the AVL TechCenter in Bietigheim-Bissingen. In addition to productive exchanges between all project partners, the program included a special highlight: extensive recording sessions with unusual traffic scenarios.
The focus was on situations that pose major challenges for current AI models: unusual pedestrian behavior, unexpected obstacles, or rare objects in the road space. Such anomalies and corner cases are crucial for making algorithms for automated driving even safer. After all, safe autonomous mobility means being able to react reliably even to rare and surprising events.
Three recording scenarios were carried out for this purpose:
Scenario 1: Pedestrian detection
Pedestrians are among the most vulnerable road users. Their movements are often unpredictable, especially in urban environments. Reliable detection is therefore essential for automated driving functions.
Scenario 2: Unknown objects
In real-world traffic, things or situations repeatedly arise that an automated vehicle cannot anticipate. This is because autonomous vehicles must not only master typical traffic situations, but also be prepared for the unexpected. The ability to recognize unusual or previously unknown objects and respond appropriately is a key component of safety and trust in automated mobility.
Scenario 3: Interaction with bicycles
Cyclists pose a particular challenge: they are fast, can change direction abruptly, and are sometimes obscured from view. They are also considered particularly vulnerable road users. AI must not only recognize them reliably, but also correctly assess their trajectories and maintain a safe distance—even in complex scenarios such as overtaking maneuvers or sudden turns.
The recordings will be used in the project to develop methods for the precise, context-related detection of road users and obstacles – even under varying conditions. The data obtained will help to identify weak points at an early stage and increase the robustness of the detection methods. The diversity of the project partners’ sensor setups, which provide different perspectives and data qualities, is particularly valuable in this context.
A brief insight into the second scenario:
Images: EICT & AVL
Video: EICT