The field is the real test environment
Sheet pile measurement is a routine part of construction site work, but doing it accurately and consistently across different sites is harder than it sounds. For this Finnish construction company, the manual process had become a persistent bottleneck.
Construction sites create difficult conditions for automated measurement. Lighting changes throughout the day, dust and mud accumulate on surfaces, pile orientations vary, and every site has its own configuration. Building a reliable system meant building it in those conditions.
The development process moved regularly between office and construction site over the course of the project. The team visited active worksites to collect footage and run live tests.
An AI system built to run on site
The solution we built runs directly on-site, processing camera footage locally without relying on a network connection. Measurement happens automatically as work proceeds, with no separate step required from operators.
The system went through multiple development cycles before the team arrived at an approach that held up consistently across different site conditions. Each iteration was shaped by footage and test results from real construction environments.
Accuracy is maintained through a calibration process tailored to each site configuration. The system can be updated remotely, so improvements reach the field without requiring on-site visits.
Iteration between lab and site drove the result
Camera selection, calibration procedures, and measurement methods were all refined through direct field experience. The team made repeated visits to active construction sites to collect new training data. Problems that appeared as edge cases in simulation were common in real deployments. Closing that gap was the central challenge of the development work.
Automated sheet pile measurement running on active construction sites
The customer runs an automated measurement system in production on real construction sites.
The system produces consistent measurements across different pile types and configurations and receives updates remotely. The architecture means the system can be extended to new requirements without rebuilding from scratch.
Read about AI solutions
Read about AI agents
