Google Scanned Objects
Provider：Google Research
Google Scanned Objects has 1,030 scanned household items (17 categories, 13 GB), each already an SDF model with mass, friction and inertia.
At a glance
- Readiness
- L1 · Rigid-body ready
- Rationale
- Manipulation L1: 1,030 genuinely scanned household items, each shipped as an SDF simulation model with estimated mass, friction, inertia and collision volumes; textures are diffuse-only and there is no articulation or semantic material data, hence L1.
- Scale
- 1,030 scanned objects across 17 categories, 13 GB total; OBJ meshes average 1.4 MB and PNG diffuse textures average 11.2 MB (Figures from the GSO paper (arXiv 2204.11918) and the Google Research blog: 1,030 objects, 13 GB, 17 categories. Note: the direct Fuel collection link app.gazebosim.org/GoogleResearch returned 404 when checked on 2026-10-25; search the 'Google Scanned Objects' collection (owner: GoogleResearch) inside Gazebo Fuel instead. The Google Research blog has moved to research.google/blog. Checked 2026-10-25.)
- License
- CC BY — CC BY 4.0 with attribution for commercial use; models are hosted as SDF packages on Ignition Fuel (Open Robotics).
- Access
- Open
- Formats
- obj · dict
- Runtimes
- gazebo · pybullet · other
- Physics properties
- Collision geometryMass / inertiaFrictionInertia tensor
- Domain
- Other
- Published
- 2026-10-25
- Updated
- 2026-10-25
Why it is listed
Most 3D asset libraries are modelled by hand and therefore idealized. GSO is scanned — structured light through a glass platform so even the base is captured — which means the meshes carry the dents, asymmetries and imperfect silhouettes of real objects, exactly the variance that makes sim-to-real transfer hard.
Readiness assessment
Each item arrives as an SDF model rather than a raw mesh: an automated pipeline filters invalid scans, estimates mass and moment of inertia from shape and volume, builds collision volumes and downsamples the mesh for simulation. That makes it an L1 drop-in for Gazebo and Bullet rather than something you have to convert first.
Practical notes
The scanner's capture volume tops out around a breadbox, so chairs and larger furniture are absent, and textures are diffuse-only, so highly specular or transparent objects are poorly represented. The set is 1,030 objects in 17 categories at 13 GB under CC BY 4.0.
Disclaimer: license and usage terms follow the official source — verify them yourself before use. This site indexes and links only; it does not host or mirror any asset file.
License and usage terms are those of the official source — verify before use. No asset file is hosted or mirrored on this site.