stereo-hero
Stereo Depth

Half a millimetre,
from stereo cameras alone

No expensive 3D sensor sits on the floor.
A model trained on the disparity between two views rebuilds the depth and shape of the workpiece.

Method

Precision sensors to learn,
cameras alone to work

Inference on site runs on stereo camera images alone. Precision comes from the software, not from the hardware bill.

Coil workpiece evaluation · mean error against Alsontech measurements0.5–0.6mm
Sensor used for on-site inferenceStereo camera

Shared internally on 10 September 2026. "Half a millimetre" refers to this comparison.

Measuring on the coil point cloud
Measured straight on the reconstruction. The 11.95 mm on screen is the wire diameter of the coil.
Reconstruction

The weld member,
rebuilt in three dimensions

These are real screens from our own labelling tool, showing the stereo inference result. The joint and the surface of the plate can be inspected from any angle.

The plate and its joint, reconstructed from stereo images
The plate and its joint, reconstructed from stereo images.
Point cloud zoomed in on the curvature and surface around the weld
Point cloud zoomed in on the curvature and surface around the weld.
Rotating the reconstructed shape
Demo 1 - the reconstruction, turned to be seen from every side.
Zooming into the joint
Demo 2 - the joint magnified to show the surface.
Use

From precise sensing
to the robot at work

  • Shape recognition Recovers the 3D information the task needs — gaps, steps and curved surfaces.
  • Dimensional measurement The reconstruction is used to measure the workpiece and its relative position.
  • Handoff to the robot Supplies what is needed to fix the weld datum and correct the workpiece position.
Data flow background

Find out what it recovers
on your own parts

Send us the drawings and conditions of the workpiece, and we will come back with the precision you can expect.