KI Data Tooling

Autonomous driving is intended to increase road safety and reduce pollutant emissions from vehicles. To enable the development of reliable systems, extensive and relevant data bases are necessary.

The joint project “KI Data Tooling” develops methods and tools for an efficient extension of the data basis for training, testing and validation of AI functions for automated driving. Overall, new methods and tools should be able to better describe, process, select, generate, refine, extend, compress, provide and transfer data.

The joint project is divided into 5 project parts:
 

• Part 1 (TP1): Validation of synthetic sensor data
• Part 2 (TP2): Quality requirements and efficiency potentials of data generation and provision
• Part 3 (TP3): Methods and tools for real data acquisition and processing
• Part 4 (TP4): Cross-sectional aspects
• Part 5 (TP5): Operational, tactical and strategic control of the overall project level

Research topics at Aschaffenburg UAS are the development of algorithms for the abstraction of different sensor data (TP2) and the enrichment of sensor data with real 3D pedestrian poses (TP3).

In TP2, new Deep Learning methods are used to transform all relevant information from different sensor data into a sensor-independent representation. In TP3, the existing sensor data / sequences are extended by additional pedestrians, based on the real 3D poses of pedestrians.

Furthermore, Aschaffenburg UAS provides a research intersection for measurement campaigns (TP3), which generate real data for project partners.