Authors

L. Schroven, N. Worzyk, Capgemini Engineering, Düsseldorf/Hamburg

Year

2026

Print Info

Production/Publication ÖVK

Summary

Safety, reliability, and trust are essential for the market acceptance of Advanced Driver-Assistance Systems (ADAS), where failures can cause financial loss, loss of trust into the product or company, and in the worst case harm to human life. As safety requirements grow, so do demands for data, compute, and time — driving up costs. To address this, we need precision: in identifying weaknesses in ADAS functions and obtaining the exact data needed for testing and improvement.

ADAS systems rely heavily on scenario perception, yet difficult situations such as text-covered cars or pedestrians in uncommon occluded crossings lead to false or missing detections. In our SAIfe project, we extract detailed scene understanding from traffic situations and correlate them with model performance, revealing weak spots and relate them to human understandable scenarios, which need to be improved upon.

Having identified situations which are difficult to handle for the model at test, our NXTAIM project synthetically generates the exact data necessary using Generative AI and simulation. We modify already labeled images and create new traffic scenarios with graph-based street layouts, enabling precise, targeted data generation without costly and time-consuming global data collection.

Finally, we combine these different capabilities mentioned before in our internal Machine Learning Center (MLC) which integrates all steps, from the identification of difficult scenarios, to the generation of these exact same situations via Generative AI, into a unified framework. This accelerates development, reduces costs, and strengthens ADAS safety and reliability in a continuously changing and demanding industry.
 

ISBN

978-3-9504969-5-6

DOI

https://doi.org/10.62626/swkq-snce Copy link

Order Members Login