01 · Imagery
What the cameras saw
Raw fisheye and undistorted perspective frames from three cameras, with human and device masks so the operator never ends up in the scene
Real-world 3D data for training and simulation
Images, camera poses and measured geometry for frontier world models and robotic simulation
Coverage
10,000+ real-world scenes, and no two alike
30M+ camera observations tied to measured geometry and camera position, from real environments
10,000+real-world scenes
30M+camera observations
25 / 8cities / countries
verticals we operate in
Homes, kitchens, living rooms, hotel rooms and suites, commercial floors and construction sites
Where the scenes come fromhover a marker
Who it's for, and how they use it
Two kinds of team read the same library differently. One needs the measurements behind the pixels. The other needs a room a robot can be dropped into
Who it's forWorld Model CompaniesHow they use it ↓Who it's forRobotics and Simulation TeamsHow they use it ↓
For world-model teams
Train on explicit 3D data from real environments
Explicit 3D data is hard to source. Most public indoor datasets are research-only, and video has nothing measured behind it. We pair source imagery with camera poses and lidar geometry, so your model gets the connection between what a camera saw and the space around it
- Posed, not just pretty. Every frame carries the camera position it was taken from
- Metric, not inferred. Lidar geometry behind the imagery, in real units
- Commercially available. Captured under agreement, not on research-only terms
Four layers, one coordinate frame
SCN · KITCHEN
layer · imagery
layer · imagery
imageryposegeometrysemantics
drag to orbitSchematic stand-in drawn from the real layer stack. Real captures load the same layers, ask for the sample pack to run them yourself
02 · Pose
Where every frame was taken from
COLMAP sparse reconstruction for both image sets. Device trajectory at 10 Hz, IMU at 200 Hz. Delivered camera-to-world in TUM format
03 · Geometry
The room, measured
Gaussian splat PLY for photoreal rendering, lidar point cloud for structure, a decimated collision mesh and a per-scene navigation mesh
04 · Semantics in build
Every object, named and boxed
Per-frame instance labels: class, instance id, source image, 2D box, and a 3D oriented box with centre, size and quaternion, fused from imagery, poses and 3D coordinates
For robotics and simulation teams
Simulate the actual site before deployment
Start with the library. 10,000 captured environments, already more varied than anything a team would build by hand, ready to train in today. When the target site is the whole point, we go and capture that one instead
Train on the library
10,000 real environments, today
Homes, kitchens, hotel rooms and suites, commercial floors and construction sites, across 25 cities in 8 countries. Run the same task across hundreds of real rooms instead of one synthetic warehouse. Delivered as USD, so it loads into Isaac Sim or anything else that reads USD
Available now · nothing to captureCapture your site
The building you are deploying into
Recreating a facility by hand is weeks of modelling before a robot ever sees it, and the model is stale the moment the racking moves. We capture the building with cameras and lidar and deliver photoreal visuals with measured geometry, simulation-ready. Test against the real layout and clearances, rerun the task, recapture as the site changes
On site · one to two days per propertyRunning in simulation today
Example scene
A real space in the delivery format, splat, lidar, collision, poses and USD
Everything a scene comes with
Five layers, one coordinate frame, so a pixel in any image resolves to a point in the room. Hover a layer to see the files
Core
Imagery
Fisheye and undistorted frames from three cameras, operator and rig masked out
camera_0..2/fisheye + perspective framesmasks/human and device masks, per frame
Core
Pose
Where every frame was taken from, at 10 Hz, with IMU at 200 Hz
poses.csvcamera-to-world, TUM formathigh_frequency_poses.csvIMU, 200 Hzcolmap/cameras.bin · images.bin · points3D.bin
Core
Geometry
Photoreal splat for perception, lidar and collision proxy for contact, navmesh for the floor
splat.plyGaussian splatcloud.laslidar point cloud, metriccollision.objdecimated proxynavmesh.objwalkable floor
In build
Semantics
Every object named and boxed, in 2D and as an oriented 3D box
instances.jsonclass · instance id · source framebbox2d · bbox3dcentre, size, quaternion
Core
Sim export
Loads into Isaac Sim as delivered, splat for what the robot sees, proxy for what it touches
scene.usdzUSD scene, proxy split from splatcamera.jsonintrinsics + extrinsics
Spec sheetper scene, as delivered
- RGB trajectories
- Multi-view RGB sequences through real environments
- Camera calibration
- Intrinsics and timestamped 6-DoF camera poses
- Metric depth
- Per-frame depth aligned to RGB
- 3D scene
- Registered point cloud and/or Gaussian splat
- Coordinate system
- Metric, right-handed, Z-up world coordinates
- Scene scale
- Indoor and real-world environments, from rooms to large facilities
- Optional labels
- Human and device masks, semantic masks, 2D boxes, oriented 3D boxes
- Packaging
- One structured folder per scene, with metadata and documentation
How we work together
Start with the sample pack, the shortest path to knowing whether this data fits your stack
Sample pack
Twenty-five scenes in the delivery format. You pick which files you need, so we do not send you anything you will not open
Dataset licence
Bulk access to the library by scene type, with the full delivery schema: imagery, poses, splat, collision, semantics
Custom capture
Environments you need and we don't have yet. We run the capture; you get the same delivery format
Research access
For labs publishing on reconstruction, world models or sim-to-real. Scenes plus poses, on citation terms
Planned
Open research set
Roughly 1,500 scenes released publicly on citation terms, a successor to what ScanNet gave the field. Downloadable, no email
Coming soon
API
Query the library by room type, geography and modality, and pull scenes straight into your pipeline
Start here