Datasets:

projectsim
/

bottles-and-containers

Tasks:

Modalities:

Formats:

Size:

Tags:

Libraries:

License:

Dataset card Data Studio Files Files and versions

xet

Community

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Want assets of your own objects? See the projectsim API.

Bottles & Containers

projectsim Lab — this dataset, live in the browser

Try it before you download. Every item in this dataset is live in projectsim Lab: orbit each can, jar and mug at real scale, then grab the full USD (physics, PBR maps) or a GLB preview per asset. Jump straight to a class: Cans · Jars · Mugs.

Procedurally generated, sim-ready 3D tabletop containers - aluminium beverage cans, glass food jars, and ceramic coffee mugs - shipped as OpenUSD (.usda) with per-zone PBR maps, plus a static glTF (.glb) for tooling that doesn't read USD. Every variant carries real per-link mass, inertia and colliders authored from the class rig, with realistic packaging formats and standard sizes.

Motivation

Tabletop manipulation policies need containers that (a) look real, (b) sim-correctly in Isaac Sim / MuJoCo, and (c) span the distribution of sizes and materials a real robot encounters (cans in every common ml format, jars in mason / jam / pickle proportions, mugs in the diner-to-espresso range), not one shape re-textured. Public assets are either scanned singletons or category-generic props (no PBR, guessed masses). This dataset is our procedural pipeline's output: every variant has a spec that produces watertight geometry, sim-authored physics, and PBR textures.

What's included

Class	Count	Articulation	Materials
`beverage_can`	34	static (rolled seam + stay-tab fixed)	label (printed), lid, rim, bottom, tab
`glass_jar`	40	static (screw lid modelled fixed)	body (transparent glass), lid (painted steel), label (printed band)
`mug`	75	static (rigid handle)	body (ceramic), handle (ceramic), base

Total: 149 variants across 3 classes, ~3.1 GB.

Per-variant contents

assets/<class>/<vid>/
├── <vid>.usda            OpenUSD, textures referenced as ./textures/*.png
├── <vid>.glb             glTF binary, embedded textures, static pose
└── textures/*.png        PBR maps (base_color, normal, roughness, metallic)

Parametric variation covered

Every class is authored from a spec.json sampled deterministically from a (count, seed) pair; anchor rows are the class's presets and dimension extremes.

beverage_can - thin-walled aluminium can with a necked top, rolled double-seam rim, dome bottom, and stay-tab. A can_format enum drives real-world formats: 330 ml (Ø66 x 115 mm), 355 ml (Ø66 x 122 mm), 440 ml (Ø66 x 149 mm), 500 ml (Ø66 x 168 mm), 568 ml UK pint (Ø66 x 189 mm), and slim/sleek variants where applicable. Label print varies (soda, beer, energy drink styles).
glass_jar - cylindrical jar with a threaded neck and a metal screw lid. Real jam / pickle / mason-jar formats: 200 ml, 370 ml, 500 ml, 720 ml, 1000 ml. Slightly rounded shoulders, 2-3 shallow helical ridges at the thread. Painted-steel lid, printed label band.
mug - ceramic coffee mug with a C-shaped handle. Body Ø65-95 mm, height 80-130 mm; capacity roughly 200-500 ml. Handle geometry sampled from a small library of C-shape parameters; base can be flat or slightly countersunk. Photoreal retexturing on one half, shader palette (cream/blue/red/black glaze) on the remainder.

The exact per-variant parameter values are not shipped in v1.

Generate your own

This dataset is sample output from the projectsim pipeline. For sim-ready assets of your objects, use the projectsim API:

Standard asset - one image in, a sim-ready asset out
Measured asset - mm-accurate, from a ChArUco-board capture
Variation sets - procedural or high-fidelity families from a single object

Intended use

Training and evaluating manipulation policies (grasp handle, pick can from tray, uncap-simulate on jars via a scripted lid replacement)
Perception (segmentation, category ID, pose estimation) from synthetic renders
Domain-randomization sources for tabletop and kitchen robotics pipelines
Sim-based experimentation before scanning real containers
Prototyping in Isaac Sim / MuJoCo / any USD-aware DCC

Tested simulators

Verified end-to-end:

NVIDIA Isaac Sim 6.0.1 - loads <vid>.usda headless and GUI. Cans / jars / mugs stack, roll, and grasp under gravity with authored mass and friction.
MuJoCo 3.11.0 - USD is not native; use our sibling MJCF drop or the sim-ready usd2mjcf converter (matching physics for static props).
Blender 5.0 - USD and glTF import both work.

Physical properties - authored vs estimated

Property	How	Notes
Volume, COM, inertia	Computed from watertight mesh	Falls back to the convex hull for open-shell parts
Mass	Volume × density	Density per zone, classified by a material-analysis VLM. The VLM inspects the object's renders together with the zone names in context — e.g. reading a can body as aluminium, a jar body as soda-lime glass, a jar lid as painted tinplate, a mug body as glazed ceramic — and looks each zone's density up in a shared material table (aluminium ~2700, glass ~2500, tinplate ~7850, ceramic ~2400 kg/m³). Per-link mass = mesh volume × classified density.
Collision shape	Auto-classified per link	Author-time collision types per link, taken directly from the rig: • primitives (`cylinder`, `box`) where the rig is geometric — can bodies, jar bodies, mug bodies • `convexHull` for sealed volumes with a single graspable shape (jar lids, mug handles) • `convexDecomposition` with PhysX shrink-wrap for multi-part rigid props (mug + handle authored as one link but decomposed at load) • `sdf` (`sdfResolution 256`) for thin non-convex shells (jar shoulder, can neck)
Textures	4K PBR maps per zone (photoreal retexturing on one half; curated shader palette on the other)	Zone-mapped so a downstream engine can rebind materials per zone. Glass jar body renders with an authored OpenGlass MDL surface in Isaac; a translucent UsdPreviewSurface elsewhere.
`max_mass_kg`	Cavity volume × water density	Meta-only hint for pick-and-place - not the physics mass. Cans and jars carry their advertised capacity by size enum.

Limitations

Physics is in the .usda, not the .glb. glTF has no physics equivalent - the .glb is a visual copy (static pose, single mesh per link, no colliders). Use it for previews, tooling that can't read USD, Blender/Unity/three.js/model-viewer; use the .usda for simulation.
All three classes are static props - jar lids and can tabs are welded rigid, mug handles are rigid extensions of the body. Open / uncap behaviours are out of scope for v1.

License

Files: CC BY 4.0 - free to use, share, and adapt for any purpose, including commercial use; attribution required.

Citation

@dataset{kaedim_bottles_and_containers_2026,
  author = {{Kaedim} and {projectsim}},
  title = {Bottles \& Containers},
  year = {2026},
  publisher = {Hugging Face},
  version = {1.0},
  url = {https://huggingface.co/datasets/projectsim/bottles-and-containers}
}

Maintenance

New classes and variants are added as separate top-level assets/<class>/ folders; the on-disk <vid> prefix is stable across future commits - a v03_glass_jar_... today will be the same asset next release. Regressions are republished as an additive commit; assets are only ever removed with a version bump.

Issues / questions: open a Discussion on this dataset repo, or email support@kaedim3d.com.

Downloads last month: 96

Total file size:

4.49 GB