Due to the high loads, the lifespan of a typical urban road is currently estimated at around 20 years. The service life of a road depends on how intensively it is used and how it is maintained, with a newly acquired road deteriorating after around 50 years without reliable maintenance. Furthermore, it was shown that the main goal of the municipalities when building new roads must be to minimize the follow-up costs, as today more than 50% of the municipalities cannot ensure the necessary road maintenance. This aspect of research is only made possible through the use of GPR technology. The basis for this is the automatic early detection of imminent damage in the deeper layers of the road structure (so-called "health monitoring").

This project focuses on the development of innovative AI-based autonomous driving systems to detect damages in the road substructure, as well as marking deep infra-substructure as fingerprints for robust and high-precision localization. To this end, a fully equipped autonomous test-vehicle is used, extending the conventional sensor set (front radar, short range radar, LiDAR and cameras), by a bi-frequency range ground penetrating radar (GPR). Deep learning based techniques will be used to develop algorithms to detect sub-surface damages in the collected radargrams. The damage detection algorithm will then be converted into a service that will be running on our server, fetching new radargrams from the database, detecting damages and pushing the damage details to our Digital Twin. Novel road-preserving autonomous driving algorithms can then be implemented by taking into account the inferred damage spots during a real-time path planning and a framework of hierarchical reinforcement learning, consisting of a multi-layer decision policy, is applied. The learning agents are able to make motion decisions, and also learn how to combine several objectives at a higher abstraction level. Moreover, all the data gathered by sensors will be used to build a digital twin, which maps the physics of the entire traffic on a road segment. The digital twin architecture serves as an uniform and extensible data layer, built using semantic technologies. It will be deployed in the cloud server, where deep neural networks will provide predictive maintenance of the road infra- and substructure.

The overall goal of the research project is the development of cloud-based traffic services for health monitoring for traffic routes and road-friendly automated driving. With the goal of road-friendly driving, we use an on-board ground penetrating radar sensors that serves as the basis for automatic early detection of impending damage in the deeper layers of the road. At the same time, the radar programs provide useful information for further localization of the vehicle.

• Design and implementation of a holistic Digital Twin for road transport infrastructure and traffic.
• Autonomous driving utilizing hierarchical AI techniques and modelpredictive control (MPC).
• Implementation of the necessary digital infrastructure, particularly a Car2Cloud interface on the protocol level (time and event based IoT protocols).
• Design and implementation of adaptive algorithms focusing on high precision localization and early recognition of emerging road damages via machine learning in the cloud.
• Health-Monitoring of the road substructure for the purpose of a Predictive-Maintenance-Service and identification and of mapping of utilities.

• Autonomous Connected Cars
• Cloud Services
• Reinforcement Learning
• Digital Twin
• Communication Protocols

Projektträger: TÜV Rheinland
 

Projektsteckbrief

Oct 2018 - Dec 2021

3D Mapping Solutions
CMORE Automotive
Dresden Elektronic
Geophysik GGD
RST Group

Prof. Dr.-Ing. Naim Bajcinca
Gottlieb-Daimler-Str. 42
67663, Kaiserslautern
+49 (0)631/205-3230
naim.bajcinca(at)mv.uni-kl.de