The final TUBERS demonstration in Athens brought inspection and repair activities together in an experimental pipe network. A new open dataset now preserves part of that work as 361 labelled endoscopic images that researchers can inspect and reuse.

Images from realistic test variation

The images were extracted from video recordings made during the robotic pipe-inspection demonstration. They cover PVC-labelled and transparent acrylic pipe sections in dry and water-filled conditions, with changes in pipe orientation, illumination, viewing angle and camera distance. The wider test network also included steel connectors, elbows, junctions and reducer couplings. Frames were selected manually to preserve visual diversity and remove near-duplicates; interface overlays were cropped out and all images were standardised as 512 x 512 PNG files.

Three labels, including the state after repair

Every image is listed in a central labels.csv file and assigned to one of three categories: defect, no_defect or after_repair. The post-repair subset shows repair tape applied over previously identified defects. This makes the record especially relevant: it documents not only what the inspection system saw before intervention, but also visual conditions after a repair action. The file structure preserves pipe material, water condition and defect status, while the README maps files back to their source recordings.

Keeping final validation useful

Created by the University of Essex, Bendabl and Alsymex, the 114 MB dataset can support research on defect classification, inspection-image analysis and post-repair verification. Its open publication under a Creative Commons Attribution 4.0 licence helps ensure that evidence from the final demonstration remains available for future comparison, testing and development.

Explore Soft Robot Image Inspection and Repair Data and download the files from Zenodo. Access the dataset on Zenodo