Most 3D datasets are one of two things: scanned meshes that look real but behave like noise, or clean CAD models with nothing around them. WorldSimReady-Home, released by Yootta with furniture company Shangpin Home Collection, is neither. It’s 500 complete houses built to be run inside a simulator, and it entered our tracker at trending rank #15 this morning.
What’s in it
- 500 household scenes, 4,462 rooms, 31,711 placed objects, covering 36,601 m²
- Assets are CAD-based and “SimReady”: each carries configured physics properties (mass, friction, articulation) alongside visual materials with roughness and metallic parameters
- Everything ships as OpenUSD (
.usdz) - 1,000 robot task episodes: 1.3 million frames, over 12 hours of RGB-D recordings at 30Hz, 640×480, in the RLDS format used by Open X-Embodiment
”SimReady” is the part that matters
A model that only looks right is enough for a render and useless for a simulation. Drop it into a physics engine and it falls through the floor, weighs nothing, or has doors that don’t open. SimReady means each object has been prepared so it behaves: the drawer slides, the chair has mass, the surface has friction.
Preparing assets that way is slow, manual and expensive, which is why there’s so little of it available publicly. A furniture manufacturer being involved explains how 31,711 objects got done: the CAD models already existed as products.
Why a creative-technology site cares about a robotics dataset
The stated purpose is training robots and embodied AI in simulation. But the artifact is 500 furnished, physics-ready interiors in OpenUSD, and that’s directly useful outside robotics:
- Virtual production and previs need furnished interiors that hold up as camera environments and behave under simulation.
- OpenUSD is the interchange format the industry has settled on. These scenes open in Omniverse, Blender (via USD import), Houdini and Unreal.
- Games and XR can use them as greybox environments with real object density, which is usually what’s missing from procedurally generated interiors.
- The RGB-D episodes are a large source of indoor depth data, and depth is useful for far more than robot policies.
The license is the catch
CC BY-NC-SA 4.0, non-commercial, and access requires accepting contact-sharing terms. So: research, teaching, personal projects and prototypes, yes. Client work, a shipped game, or a commercial production, no. The ShareAlike clause also propagates to derivatives, which matters if you build on the scenes and want to release the result.
There’s no paper attached yet, just the dataset and a GitHub repo. Worth watching for how it’s evaluated.