Case study · Autonomous mobile robots
The robot runs the survey. Engineers make the judgement calls.
Railway tunnels must be inspected regularly — when surveys are done manually, people walk kilometres through dark, confined infrastructure. For SNCF, we built and tested an autonomous robot that takes on the survey run.
The challenge
Tunnel inspection is essential, unavoidable — and one of the least automatable jobs in rail infrastructure.
- 01
Inspecting by hand means kilometres on foot through dark, confined infrastructure — slow, costly and hazardous for the people doing it.
- 02
Tunnels are GPS-denied, unlit and unstructured — exactly the conditions where off-the-shelf robots fail.
- 03
A defect report is only useful if engineers know precisely where the defect is. Manual surveys struggle to deliver that precision, kilometre after kilometre.
Autonomous survey without GPS or fixed infrastructure
The robot enters the tunnel and navigates it autonomously — no GPS, no pre-installed infrastructure. Multimodal perception builds a semantic map of the structure as it moves: not just geometry, but what each element is.
Anomaly candidates are flagged and precisely located on the map for engineers to review — the robot does the survey, engineers make the judgement calls. All processing runs on-board, so the robot operates autonomously in environments with no connectivity.
The results
Field-tested with SNCF in real, GPS-denied tunnel conditions.
In on-site testing, the robot surveyed at 2 m/s while building its semantic map, with 1 cm navigation precision and all processing on-board. Anomaly candidates were flagged and located on the map for engineer review.
Still surveying infrastructure on foot?
Book a demo to review the environment, localisation constraints and report format with our team.