Our research project is intended to develop a system that processes images and other vehicle information and uses this to derive the state of the vehicle in real time.

This holistic AI-based system depicts vehicle fleets as digital twins and enables fleet operators to transparently determine the condition in near real time without the use of specialized hardware. Vehicle damage is identified and categorized using computer vision. Among other things, novel convolutional neural network (CNN) models for object recognition are being researched and developed. Various use cases are considered and made possible:

- Before/after comparison and absolute status determination

- Routine examination

- Report of specific damage

The main research goal is a low-computation-intensive solution for AI models so that they can be operated directly on smartphones without sacrificing precision and thus avoid high loads in the cloud and resulting latencies.

Based on this, automated processing for customers and insurance companies should be possible through machine learning. The product is complemented by predictive maintenance. In the operating information collected from the vehicles, AI also uses previously unknown relationships to enable flexible maintenance intervals — including according to user requirements such as bundling activities vs. fleet availability.