Case study KI Data Tooling
As part of a nationwide automotive consortium, fortiss and its partners developed key components for reliable, AI-based autonomous driving. The institute’s contribution focused on advanced data processing, the integration of synthetic data and the optimisation of machine learning. The result was a robust and scalable data pipeline that significantly improves the training, validation and safety of AI-based vehicle functions.
The AI Data Tooling project was funded by the Federal Ministry for Economic Affairs and Energy (BMWE) and brought together leading partners from industry and academia.
Challenge
Safe autonomous driving requires AI systems that function reliably even in rare and critical situations. However, existing data infrastructures often fail to provide comprehensive coverage across different sensor types and driving scenarios, making it difficult to train and validate these systems effectively. The consortium required a complete, scalable data toolkit – including data generation, quality assessment and efficient processing – to support robust AI development.
Solution
fortiss developed data-driven methods and deep learning techniques to support the end-to-end training of AI systems for autonomous driving. The focus was on challenges such as efficient data annotation, the integration of synthetic data, the detection of rare or unknown (corner case) scenarios, and the generation of contextual information for so-called long-tail scenarios. These elements were integrated into the consortium’s comprehensive tooling pipeline. The result was a standardised, modular data toolkit designed to accelerate development and validation processes in the automotive industry.
Result
- Successful integration of the fortiss solution into the project’s comprehensive AI tooling pipeline
- Enabling faster and higher-quality training of AI systems
- Improving reliability in the validation of AI-based driving functions
- Creation of a vital foundation for safe and intelligent mobility solutions in Germany
- High-quality, multimodal data as a key prerequisite for robust AI training in autonomous driving
- Use of synthetic data to expand training datasets
- Targeted modelling of long-tail scenarios to cover safety-critical exceptional cases
Outcome
The project demonstrates that a scalable, data-driven tooling pipeline significantly improves the training and validation of AI driving functions and can reliably cover critical and rare scenarios in particular. The approach thus provides a key foundation for the safe development and widespread deployment of autonomous driving functions.
Project partner
