Short answer
Furniture and fixture data tells a humanoid where work happens and what it can move: desks, shelves, carts, doors, counters, rails, outlets. Facilities teams and furniture makers may hold images, floor plans and 3D models that support recognition and layout understanding. They do not show how a drawer sticks, a chair rolls or a door closes under a hand.
Why fixtures are their own category
Products come and go. Fixtures define the stage they pass through. A rack, a workbench, a reception desk or a fire door sets the reachable space, the heights at which things are stored and the paths a mobile body can take. Mistaking a fixed shelf for a movable cart, or a glass partition for open space, is a perception failure with physical consequences.
Most product and catalog datasets ignore this layer. A buyer deploying in an office, clinic, lab or back-of-house kitchen needs images of the fixtures that actually fill those rooms, labeled well enough to separate "bolted down", "movable with effort" and "meant to be moved". That distinction is rarely visible in a photograph alone and usually has to come from the owner's records.
This page covers sourcing recognition and layout data for fixtures. For the line between recognizing and handling objects, see object recognition versus manipulation data; the cluster overview is the product and object data hub.
What to ask for
Recognition imagery
Photos of fixtures in real rooms, taken at working heights, with the clutter people leave on them. Furniture-maker catalog photos are clean and consistent but show no context. Facilities photos, from move-in surveys, space audits or maintenance tickets, are messier and closer to what a robot sees. Ask for both if your model must cope with each.
Layout and 3D assets
Floor plans, room scans and manufacturer 3D models describe geometry that images cannot. They help with placement, clearance and sim-ready scenes. See manufacturer 3D models and CAD for rights and fidelity questions, and simulation-ready object assets for what makes an asset usable in a simulator.
Part and affordance labels
Handles, drawer fronts, hinges, latches, seat height adjusters and wheel locks are the parts a body would act on. Labels for them are covered in affordance and part labels, and furniture with moving parts is discussed under articulated object models.
A published reference, and where it stops
BEHAVIOR-1K defines 1,000 everyday activities in 50 scenes, run in the OmniGibson simulator, with more than 9,000 annotated objects (Li et al., 2024). It is a helpful reference for how scene objects can be annotated with properties relevant to activities.
Its scenes are simulated homes and similar environments, not real workplaces, so it is only a capture and labeling reference for this page. It does not tell you what furniture an office, laboratory or warehouse actually contains, and it should not be read as evidence that real-world fixture images transfer cleanly.
Illustrative fixture record
The schema below is illustrative, with synthetic values. It shows the kind of per-fixture attributes that make an image set filterable. It is not a standard and not a minimum.
| Field | Illustrative value | Purpose |
|---|---|---|
fixture_id | fx_02291 | Joins images, 3D files and floor-plan entries |
category | lab_bench | Class for detection |
mobility_class | fixed / movable_with_effort / designed_mobile | Tells planners what may be moved |
articulation | two_drawers, one_door | Flags parts that open |
dimensions_source | measured / manufacturer_nominal | Avoids mixing measured and catalog values |
room_type | clinic_exam | Context for domain splits |
site_id | site_011 | Held-out-by-site evaluation |
Keep mobility_class separate from category. Two identical-looking tables can differ: one welded to a wall, one on casters. Ask who assigned the class and from what evidence.
Supplier types and what each tends to hold
- Facilities and property managers. Space audits, move photos, asset registers, floor plans. Strong on context and mobility, uneven on photo quality.
- Furniture and fixture manufacturers. Catalog imagery, CAD files, spec sheets. Strong on geometry and variants, silent on wear and placement.
- Interior and fit-out contractors. Installation photos and as-built drawings. Useful for fixed elements.
- Operating businesses. Photos taken for other reasons, such as incident records or inspections. These carry people and sensitive areas, so rights review is unavoidable.
Whether any of these holds a particular type of fixture data is a question to put to each supplier, not an assumption to build a plan on.
What this data cannot support
Recognition imagery and 3D models do not describe how fixtures behave. They omit drawer friction, door closing force, caster resistance, stiffness of a loaded shelf and the torque a latch needs. A humanoid that must open a cabinet, push a cart or adjust a chair needs robot or teleoperation data with joint states, actions and contact signals. Photos, floor plans and manufacturer files cannot supply that.
Other limits to plan for:
- Nominal versus as-built. Manufacturer dimensions describe a model, not the unit installed in room 4 after a retrofit.
- Changes over time. Furniture is moved, replaced and modified. An archive of six months ago may not match today's room.
- Privacy and security. Images of workplaces can capture whiteboards, screens, badges, documents and restricted areas. Redaction reduces exposure but does not settle permission questions. This is not legal advice.
- Evaluation. Separate rooms or sites for testing so the model is not graded on a space it has seen.
Acceptance checks you can set
- Mobility class and articulation fields are populated, with the rule used to assign them documented.
- Images include working-height views, not only catalog angles, if your robot's cameras are at that height.
- Measured and nominal dimensions are never mixed in one column.
- Images are grouped by site and room, so splits can be made without leakage.
- Sensitive content (screens, documents, faces) is flagged and handled according to the license.
Where fixtures carry warnings, such as fire doors, eyewash stations and lockout points, the labels are covered separately in workplace signage and safety labels.
Questions to ask suppliers
- Which room types and fixture categories are represented, and which are absent?
- Is mobility (fixed, movable, designed mobile) recorded, and how was it determined?
- Are dimensions measured on site or taken from manufacturer data?
- What camera heights and viewpoints are present, and were they chosen deliberately?
- Do images show people, screens, documents or restricted areas, and was any review done before release?
- Who owns the images and 3D files: the business, a photographer, a contractor or the manufacturer?
- Are floor plans and CAD files covered by the same license as the photographs?
- How current are the records, and can they be refreshed after a renovation?
- Can some sites be held back for evaluation?
Explore your options
If your deployment depends on fixture recognition in a particular setting, describe the room types, assets and permitted uses to SourceX. SourceX looks for US businesses that hold suitable material, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. A request does not guarantee a match; SourceX follows up about potential sources. Start at the humanoid training data guide for the broader sourcing routes, or see the catalog entry for CAD and engineering drawings.
Explore sourcing and licensing options with SourceX
Sources
- BEHAVIOR-1K: A Human-Centered, Embodied AI Benchmark with 1,000 Everyday Activities and Realistic Simulation — simulated scenes with annotated objects; a labeling reference only, not workplace coverage.