A LiDAR or depth point cloud (a decoded ROS PointCloud2) drawn with raw WebGL2 and no 3D library: coloured by height or intensity on the accent's single-hue ramp with its scale, a recessive ground grid, the robot at the origin and an axes gizmo in words; orbit, pan and zoom by pointer, touch and keyboard; loading, empty, error, stale and no-WebGL states.
Drag to rotate, right-drag or two fingers to pan; the wheel or a pinch zooms once the view has focus. Arrow keys rotate, Shift with the arrows pans, plus and minus zoom, 0 resets the view.100,000 points. x 0.1 to 18.0 m forward, y -4.2 to 4.2 m left, z 0.0 to 4.5 m up. The robot is at the origin, facing x
100,000 points
Height0.0 m4.5 m
Point cloud by height
Install
npx shadcn add @syncrobotic/point-cloud Also installs: lucide-react, @syncrobotic/cn, @syncrobotic/button, @syncrobotic/button-group, @syncrobotic/skeleton
Overview
When to use
- What a robot's LiDAR sees right now, beside its camera on a robot's page or a teleoperation screen.
- Checking a depth camera's mounting or calibration: is the floor flat, are the shelves where they should be.
- Reviewing a saved scan after an incident or while building a map.
Use something else
- Where robots are on a site, in 2D: Robot map.
- What a camera sees: Camera feed, or Annotated image for one frame with detections.
- A mesh, a CAD model or a robot's URDF: a 3D-model item built on three.js, not this viewer.
More examples
States and real uses. Each example is also a Storybook story and a visual regression test.
Drag to rotate, right-drag or two fingers to pan; the wheel or a pinch zooms once the view has focus. Arrow keys rotate, Shift with the arrows pans, plus and minus zoom, 0 resets the view.100,000 points. x 0.1 to 18.0 m forward, y -4.2 to 4.2 m left, z 0.0 to 4.5 m up. The robot is at the origin, facing x
100,000 points
Intensity22236
Point cloud by intensity
Drag to rotate, right-drag or two fingers to pan; the wheel or a pinch zooms once the view has focus. Arrow keys rotate, Shift with the arrows pans, plus and minus zoom, 0 resets the view.Loading the point cloudLoading the point cloud
Point cloud loading
Drag to rotate, right-drag or two fingers to pan; the wheel or a pinch zooms once the view has focus. Arrow keys rotate, Shift with the arrows pans, plus and minus zoom, 0 resets the view.No points
No points
Point cloud empty
Drag to rotate, right-drag or two fingers to pan; the wheel or a pinch zooms once the view has focus. Arrow keys rotate, Shift with the arrows pans, plus and minus zoom, 0 resets the view.The scan file could not be read: it ended early
The scan file could not be read: it ended early.
Point cloud error
Drag to rotate, right-drag or two fingers to pan; the wheel or a pinch zooms once the view has focus. Arrow keys rotate, Shift with the arrows pans, plus and minus zoom, 0 resets the view.100,000 points. x 0.1 to 18.0 m forward, y -4.2 to 4.2 m left, z 0.0 to 4.5 m up. The robot is at the origin, facing x
100,000 points
Height0.0 m4.5 m
Point cloud stale
Drag to rotate, right-drag or two fingers to pan; the wheel or a pinch zooms once the view has focus. Arrow keys rotate, Shift with the arrows pans, plus and minus zoom, 0 resets the view.100,000 points. x 0.1 to 18.0 m forward, y -4.2 to 4.2 m left, z 0.0 to 4.5 m up. The robot is at the origin, facing x
100,000 points
Height0.0 m4.5 m
Point cloud no web gl
Usage rules
- For what a LiDAR or depth camera sees, in 3D. A site's floor plan with robots on it is
robot-map; a mesh or a CAD model is a different item. - Pass the points in the ROS frame they came in (REP 103: x forward, y left, z up, metres), as a
Float32Arrayof x, y, z. Decode the PointCloud2 in the app; NaN points are skipped. - Colour by height unless intensity answers the question (reflective tape, glass, a retro-reflector). Colour is the accent's single-hue ramp with its scale in the footer, never a rainbow;
colorBy="none"draws the grey of a map's scans. - For a live cloud, fix
colorRangeso a colour means the same height from one scan to the next, and setstaleAfterMs(withreceivedAtfrom the message's stamp when you have it) so a frozen cloud says Stale instead of passing for live. - Give
labelthe sensor and the robot ("Front lidar, Patrol dog 7F3A"); screen readers hear it as the view's name, with the point count and the extent in metres as its description. - Keep
pointSizeat 2 px for dense clouds; 3–4 px for sparse ones. Up to about 500,000 points stay smooth to turn: thin larger clouds before passing them (a voxel grid in the app). - Bottom left holds Reset view and Fit, so the emergency stop keeps the bottom right.
Changelog
- 0.24.12026-11-21A colorRange written inline no longer re-uploads the whole cloud on every render; resizing no longer flashes an empty frame; under Windows High Contrast the points stay visible (CanvasText on Canvas); a lost pointer capture ends the drag; the middle button no longer starts autoscroll; wheels that scroll by lines or pages zoom the right amount; a live cloud reuses its buffers instead of allocating new ones for each scan. Screen readers now hear a stable name (the label and “3D point cloud”) with the count and extent in the description, which changes at most every 5 s. New label
viewName; new exportspointCloudWheelZoomandpointCloudPaletteSources;packPointCloudtakes an optional array to pack into. - 0.24.02026-11-21First release.