Cases/06 · Productisation

Rope Robotics.Autonomous wind turbine maintenance.

Wind energy·Computer vision and autonomous robotics·Practitioner: Simon

Rope Robotics builds autonomous robots that maintain the blades of offshore and onshore wind turbines — replacing rope-access teams in the most dangerous and most expensive part of the maintenance window. The robots climb, inspect, and repair surfaces sixty to ninety metres above ground in conditions humans should not be doing this work in.

Rope Robotics blue-and-white rig clamped to the leading edge of a wind turbine blade, with another turbine and open farmland visible behind.
In the field · Rope Robotics rig on a turbine blade

Challenge

The robotics had to be autonomous enough to run with a single technician supervising from the ground. That meant the robot needed to see what it was doing — calibrate itself against the blade's curvature, detect defects in the surface, and, not least, recognise when it was about to fall off the blade's edge. Every one of those was a computer vision problem. Every one of those needed to be reliable enough that a person on the ground could trust it.

What was done

The computer vision and software layer of the autonomous servicing solution was delivered: blade calibration and navigation, defect detection, edge-detection (the robot doesn't fall off), and the operations layer that lets a single technician supervise multiple robots from the ground. The work was carried out alongside the engineering team that built the robot itself; the hardware was out of scope.

Edge detectionTwo side-by-side fisheye camera views from the rig, each annotated with a measured true distance to the blade edge and a model-predicted distance overlaid in red and green.
Real-time distance estimation — the robot's own measurement of how close it is to falling off the blade.
On-bladeClose-up of the Rope Robotics manipulator arm clamped to a turbine blade surface, ready to perform inspection.
The manipulator deployed on the blade surface during an inspection run.
SegmentationDiagram of a U-Net style segmentation pipeline: input fisheye image enters convolutional encoder/decoder layers and outputs a defect-segmented mask passed to post-processing.
U-Net segmentation pipeline — input frames in, defect masks out, post-processed for the operator.

Measurable results

67%

Operational cost reduction

3×

Faster deployment time

98.3%

Edge detection accuracy

Practitioner

Simon — Computer vision, AI implementation, Product.

Where this fits

This case maps to 'Our data could be new revenue. We don't know what to build.'Computer vision and real-time machine learning productised into a commercial offering — turning a hard technical problem into validated commercial revenue. The engagement was an Implementation — direct delivery, with the receiving engineering team integrated from day one.

See the situation →See engagement options →