Synthetic Data for Autonomous Earthworks in Civil Construction

SynDAB is a validated technology that enables the generation of structured, synthetic training and test data for autonomous construction machinery. By simulating real-world conditions such as earthworks, slopes, and environmental clutter in Unreal Engine and ROS, SynDAB provides AI-ready datasets for system training, validation, and certification.

• Eliminates the need for costly, high-risk real-world testing
• Enables simulation of rare and hazardous corner cases
• Closed-loop validation pipeline with requirements traceability
• Supports standards-based evaluation and performance metrics

• Autonomous excavators, graders, and trenchers
• Earthmoving robots for civil infrastructure projects
• Robotics platforms for construction automation

The construction industry is rapidly adopting autonomous and semi-autonomous systems for tasks such as excavation, grading, and site surveying. However, these systems operate in dynamic and often hazardous environments where access to diverse, high-quality training and testing data is limited. Traditional field testing is time-consuming, expensive, and often fails to capture rare but critical edge cases. Furthermore, certifying these systems to meet related safety standards requires reproducible testing and traceable validation, which current methods struggle to deliver.

The invention originates from the Institute of Industrial Automation and Software Engineering (IAS) at the University of Stuttgart, under the direction of Prof. Michael Weyrich (Institute Director) and Prof. Christof Ebert. Their research focuses on AI-based testing and validation of autonomous and safety-critical systems in complex environments.

The project builds on prior work in the field of simulation-based validation for autonomous systems and software quality assurance. 

Validating autonomous systems for construction requires large-scale, scenario-rich datasets which are not readily available. Real-world data collection is dangerous, expensive, and often incomplete, especially for safety-critical or failure-prone situations. Certification processes remain lengthy due to a lack of reproducibility, traceability, and robust simulation tools.
Furthermore, construction sites involve constant environmental changes, multi-machine interactions, and unpredictable human factors, making it extremely difficult to replicate key scenarios for testing consistently. As a result, developers face high costs, long validation cycles, and limited regulatory clarity.

The University of Stuttgart has developed and patented an AI-based validation framework that enables the targeted simulation of operational and failure scenarios in construction environments. The system generates synthetic sensor data across various terrain types and dynamic site conditions, reducing the need for high-cost and high-risk physical testing.

By using artificial intelligence, the platform creates a Minimum Viable Test Set (MVTS) consisting of risk-prioritized scenarios, significantly reducing the number of real-world tests required for system validation. These scenarios replicate safety-critical interactions between autonomous machinery and unpredictable site conditions, enabling traceable and reproducible assessments.

The system supports integration with digital twins and test platforms and aligns with standards such as ISO 17757, ISO 13849, and PAS 1883. It accelerates development, supports regulatory compliance, and enables scalable validation of AI-driven construction technologies.

This technology can be transferred to industry through licensing, which provides a straightforward commercialization pathway. Alternatively, the inventors’ team at the University of Stuttgart has implemented several algorithms in software that can be adapted and extended through cooperation or consulting projects, allowing tailored applications for specific industrial use cases.

Ebert, C. et al. (2023). AI-Based Testing for Autonomous Vehicles. ResearchGate link
IAS (2022). Testing Software Systems. University of Stuttgart. Link
 

Fickinger, M. et al. (2023). Simulation-based Validation for Construction Robotics. Automatisierungstechnik, De Gruyter. Link