projectsim/yam-real-to-sim-assets

Real-to-Sim YAM Benchmark: authored assets

The sixteen sim-ready objects of the authored reconstruction in Measuring Asset and Scene Reconstruction Effects in Real-to-Sim Robot Evaluation (Verma, Cilio, Christodoulou, Fermelis, Ng, Chung, Liang, Kaedim, 2026; arXiv:2610.00731). They are the exact files loaded by the twenty simulated runs the paper reports, exported to text USD alongside the original binary.

Each object was built by the projectsim pipeline from one photograph of the real object on the evaluator's table, with metric scale estimated from a printed ChArUco board of known size captured with the object. Mass, friction and restitution are authored per object from a material classification and a per-class property table; mass is that density times the mesh volume. Colliders are chosen by the pipeline per object (signed distance field, convex hull or convex decomposition). Centre of mass and inertia are left to the engine.

The tasks, the real per-trial scores, the scoring code and the run artefacts are in the harness repository: Kaedim/yam-sim-harness-oss.

What's included

object	task	size (cm)	collider	mass (kg)	material	friction s/d	restitution
Heineken bottle (sm_bottle_glassheineken_8d6)	bottles in bin	5.9 x 5.9 x 22.3	convex hull	0.214	glass	0.55 / 0.40	0.30
Sapporo bottle (sm_beerbottle_glasssapporo_d06)	bottles in bin	5.9 x 5.9 x 22.7	convex hull	0.193	glass	0.55 / 0.40	0.08
Pellegrino bottle (sm_bottle_glasspellegrino_d5d)	bottles in bin	7.8 x 7.8 x 28.7	convex hull	0.449	glass	0.55 / 0.40	0.08
blue storage bin (sm_storagebin_plasticopen_f43)	bottles in bin, clear table	27.6 x 27.2 x 12.5	convex decomposition	0.434	plastic	0.35 / 0.28	0.25
red wooden block (sm_block_woodreda_e0a)	stack blocks	4.0 x 4.1 x 4.3	convex hull	0.040	wood	0.45 / 0.35	0.35
blue wooden block (sm_block_woodblue_d8b)	stack blocks	4.2 x 4.1 x 4.2	convex hull	0.045	wood	0.45 / 0.35	0.25
ribbed plastic bowl (sm_bowl_plasticribbed_07f)	stack bowls (x3)	14.4 x 14.4 x 7.1	signed distance field	0.093	plastic	0.35 / 0.28	0.50
steel latte art cup (sm_pitcher_metalbrushed_046)	move latte cup	11.5 x 7.6 x 9.2	signed distance field	0.266	metal steel	0.60 / 0.45	0.42
bamboo cutting board (sm_cuttingboard_bamboofunctional_c14)	move latte cup	23.6 x 35.2 x 2.4	convex hull	1.130	wood	0.45 / 0.35	0.35
drinks can (sm_cokecan_aluminumcylindrical_5ba)	move latte cup	6.3 x 6.3 x 12.3	signed distance field	0.354	metal aluminum	0.45 / 0.35	0.25
spray bottle (sm_cleaningbottle_plasticlabel_3fc)	move latte cup	10.7 x 5.9 x 20.1	signed distance field	0.713	plastic	0.35 / 0.28	0.18
puzzle cube (sm_rubikscube_plasticpuzzle_64e)	clear table	5.7 x 5.8 x 5.6	signed distance field	0.085	plastic	0.35 / 0.28	0.50
baseball (sm_baseball_leatherstitched_713)	clear table	7.3 x 7.3 x 7.3	bounding sphere	0.145	leather	0.55 / 0.45	0.54
plastic grapes (sm_grapes_thinplastichollow_a7d)	clear table	16.7 x 9.9 x 6.2	convex decomposition	0.217	plastic	0.35 / 0.28	0.30
red mug (sm_mug_plasticred_02c)	clear table	10.9 x 8.2 x 9.7	signed distance field	0.076	plastic	0.35 / 0.28	0.60
small plate (sm_plate_plasticspeckled_3b0)	clear table	14.9 x 14.9 x 1.5	signed distance field	0.039	plastic	0.35 / 0.28	0.45

The three bowls in stack bowls are three instances of the same asset. The blue bin is a static collider in both tasks that use it. Sizes are the axis-aligned bounding box of the mesh in the asset's own frame, Z up, metres per unit 1.0. Friction values are pair values on dry steel with combine mode average; restitution values are drop-test pair values on a hard floor with combine mode max.

Total: 16 assets, ~550 MB (textures are 4K PNGs).

Per-asset contents

assets/<asset_id>/
├── <asset_id>.usd        the binary OpenUSD file the benchmark runs used
├── <asset_id>.usda       the same stage exported to text USD, textures referenced as ./textures/*.png
├── <asset_id>.json       the pipeline's physics record: material class, density, friction,
│                         restitution, mass, centre of mass, inertia, combine modes
└── textures/*.png        PBR maps (base colour, normal, roughness, metallic)

objects.csv at the root is the table above as data.

Provenance and fidelity notes

• These are the assets as the reported runs used them: the colliders in the table are what PhysX collided with, and the masses printed in every run log match the values in the table. The .usda is an export of the same stage for readers who want to diff or edit in text.
• The default reconstruction of the paper (the same photographs through TRELLIS with hand-typed scale and engine-default physics) is a separate release. Do not mix the two when reproducing.
• The photographs the assets were built from, the ChArUco clips and the scene splat are released separately with the capture data.

Use with the harness

Place the assets/ folder where the runner expects the authored arm ($HARNESS_ROOT/assets, see the harness README) and run any cell:

./run_cell.sh configs/bowls_pi05.env kaedim

To evaluate your own reconstruction of the same objects, build an asset folder with the same <asset_id>/<asset_id>.usd names and point ASSET_ROOT at it; the scoring compares it against the real trials with the paper's paired bootstrap.

Generate your own

These assets are output of 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 (the method used here)

Licence and citation

CC BY 4.0. The real-robot trials these objects were evaluated against were run by Robocurve under a benchmarking agreement with Kaedim; see the paper for the full disclosure.

@misc{verma2026realtosim,
  title  = {Measuring Asset and Scene Reconstruction Effects in Real-to-Sim Robot Evaluation},
  author = {Verma, Sanya and Cilio, Luca and Christodoulou, Velissarios and Fermelis, Tyler and Ng, Qi Ting and Chung, Rachel and Liang, Sofia},
  year   = {2026},
  eprint = {2610.00731},
  archivePrefix = {arXiv},
  note   = {Code: https://github.com/Kaedim/yam-sim-harness-oss}
}