Nessco Vision · local verification

Uses the workbench Vision handlers and the real OpenCV runtime. No camera permission is needed for the sample.

UI responsivePhysical AI preview
On-device OpenCV

Nessco Vision

Use a live camera or upload PNG, JPEG, and WebP images. Analysis runs in OpenCV on this device. Models load only when needed; no account is needed for local analysis. Camera access starts only when you ask, and stops when this panel closes or the page is hidden.

Camera is offStart the camera or upload an image.

Analysis stays in this browser. Capture & attach adds a chat attachment, which is sent only when you send your message.

Reference object

Upload an image cropped to a flat, textured object, then save it as the reference. Upload another scene or start the camera to find it.

No reference saved. It stays in this tab until cleared.
Live resultWaiting for an image
Start the camera or upload an image to inspect it with OpenCV, here on this device.
3D pose estimationOpenCV solvePnP · entirely on this device
Open printable ArUco marker #7
Camera calibration

Optional JSON: width, height, fx, fy, cx, cy, distortion [k1,k2,p1,p2,k3]. Use the same camera, resolution and crop as the calibration.

Approximate camera: field of view estimate.
Use the full black width of an ArUco marker, or the real width and height of the saved flat reference. A plain square has no identity or stable in-plane orientation. Axes show X red, Y green, Z blue. This estimates a known object's pose; it does not reconstruct a scene or a human skeleton.
Physical AI

Import a robot, simulate it, train a policy against it, and export to Omniverse and Isaac Lab.

Drop a URDF, STL, STEP, or USDA file

URDF is what RoboDK and ROS export. From SolidWorks, export STL or URDF — STEP gives bounds only.

Joint targets

Reinforcement learningheld-out · population mean

SLAMmap · drift vs ground truth

Inverse kinematicsDamped least squares, as RoboDK and MoveIt solve it