Synthetic Data for Autonomous Construction Machinery

Weiße Schrift

AI FOR THE GENERATION OF SYNTHETIC DATA FOR THE DEVELOPMENT AND TEST OF AUTONONOMOUS CONSTRUCTION MACHINERY

Autonomous construction machines are highly relevant for automation of the computational design and construction. Construction sites of the future demand for autonoums systems such as cranes, assembly robots, excavators, wheel loaders etc. due to labor shortages, efficiency requirements or hazardous environments. Questions of development and testing of automatic functions become very relevant for the usage of autonomous machines based on AI.

However, in this area of computational design and construction, it is difficult to generate real data sets for the complex scenarios of autonomous construction machines in order to develop and test the functionality. Synthetic data is required to create a sufficient number of scenarios, especially multiple test cases for the training of an AI and its functionality.

For example, in the development of autonomous construction machinery, scenarios with obstacles on the road outdoors or indoors, workers surrounding the machine, or yielding ground are typical situations. Rare events, which often represent critical situations, are hardly found in real data due to the limited operating time of prototypes and safety measures in operation. The generation of a large number of critical and rare situations in a real environment of autonomous construction machine is practically not feasible due to high costs and safety risks.

In this project, synthetic data is created by simulating rare scenarios in a Co-Simulation system consisting of an environment- and construction machine simulation resulting in training, development and system test data sets.

PARTICIPATING RESEARCHER

Prof. Dr.-Ing. Dr. h. c. Weyrich
Institute of Industrial Automation and Software Engineering (IAS)

RESEARCHERS

Andreas Löcklin
Alexander Schuster
Iman Sonji

PARTNERS

Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung (IOSB)
Robo-Test / Technologie-Transfer-Initiative GmbH an der Universität Stuttgart
Ingenieurgesellschaft Prof. Czurda und Partner mbH (ICP)

FUNDING

BMBF, grant number 01IS21057A