Semantic & material ready Embodied / Robotics Mixed (check per asset) Open

RoboCasa365 Scenes and Object Assets

Provider:UT Austin x NVIDIA Research

RoboCasa365 ships 2,500+ kitchen scenes, 3,200+ interactive objects and articulated furniture, with 600+ hours of human and 1,600+ hours of synthetic demonstrations.

Get it from the source →

At a glance

Readiness
L2 · Semantic & material ready
Rationale
Built on RoboSuite/MuJoCo with articulated furniture and appliances and collision/mass properties; 3D assets come from Objaverse, LightWheel and Luma generation with per-asset licenses, hence L2 and mixed.
Scale
2,500+ kitchen scenes / 3,200+ objects (RoboCasa365 (ICLR 2026) reports 2,500 scenes and 3,200+ objects, plus 2,200+ hours of demonstrations (600+ human, 1,600+ synthetic). Checked 2026-10-24.)
License
Mixed (check per asset) — Framework code is MIT (LICENSE, copyright 2026 the RoboCasa Team, noting embedded Apache-2.0 portions of DeepMind MuJoCo); the README declares CC BY 4.0 for assets and datasets; object meshes additionally come from Objaverse 1.0, LightWheel AI and Luma AI generation with their own per-asset licenses — verify each before commercial use.
Access
Open
Formats
mjcf · obj
Runtimes
mujoco · isaac-sim · sapien
Physics properties
Collision geometryMass / inertiaFrictionJoint structureSemantic labels
Domain
Solid / Structure
Published
2026-10-24
Updated
2026-11-15
Evidence: Primary sourceVerified: 2026-11-153 sourcesView evidenceHide evidence
Scope and limits

Verification covers the licence ruling made on 14 November 2026 against the repository LICENSE and README, plus the scene, object and demonstration figures reported by the RoboCasa365 paper and project page. The L2 readiness grade and the mixed licence label are this site's assessment of those materials; the upstream project does not publish such a grade. Object meshes come from several upstream sources whose per-asset licences were not checked individually in this round. This change adds structured sources only; the existing ruling text is unchanged.

#SourceTypePublished / Accessed
1RoboCasa repository LICENSE and READMERoboCasa TeamRepositoryNot stated / 2026-11-14
2RoboCasa project siteRoboCasaOfficialNot stated / 2026-11-15
3RoboCasa documentation: overviewRoboCasaDocumentationNot stated / 2026-11-15

The verification date covers only what was checked in that pass; it does not mean every figure on this page was independently verified.

Why it is listed

RoboCasa is a large-scale simulation framework for everyday manipulation tasks; the 365 release scales to 2,500 kitchen environments and 365 tasks.

Readiness assessment

Scenes are built from real-world layouts and styles, then fitted with interactable furniture and appliances; many are articulated, so doors open, knobs twist and burners switch on.

Practical notes

Object assets are curated from Objaverse 1.0, LightWheel AI and text-to-3D generation, and textures can be swapped for domain randomisation.

For the framework, the task suite and the benchmark protocol, see the RoboCasa framework entry; this entry indexes scene and object assets and their simulation-readiness grade only.

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.
Reconciliation note (2026-11-14): the repository LICENSE declares MIT for the code and notes embedded Apache-2.0 portions of DeepMind MuJoCo, while the README separately declares CC BY 4.0 for assets and datasets. Earlier text in this entry labelled the framework code as Apache-2.0, which was a misstatement; Apache-2.0 applies to the embedded MuJoCo portions only.

License and usage terms are those of the official source — verify before use. No asset file is hosted or mirrored on this site.