RepliCar

To overcome the challenges of automatic environmental perception in public road traffic, sophisticated Artificial Intelligence (AI) methods are necessary. Currently, malfunctions of AI-driven functions are identified through multi-year analyses. Repeating development processes need efficient and reliable validation processes.

The RepliCar project aims to significantly advance the validation process with a novel reference sensor data platform, including a new type of reference radar. Based on the reference sensor technology, all partners are jointly developing new test bench concepts, methods, data pipelines, and simulation tools for a modular validation process for highly automated driving systems.

The FZI is working on the methodical and automated validation of machine perception, particularly concerning object recognition. Research is being conducted on how the principle of scenario-based testing can be concretized at the level of perception. To fulfill this purpose, the FZI is developing modular automated measurement processes of sensor data, contexts, and performance recordings to enable comprehensive statements between reference sensors and test objects. To generate the required sensor data, the FZI is also developing an antagonistic sensor demonstrator whose approaches contribute to the holistic “sensor-in-the-loop” concept.