rmsreis/characterization-ready-assets

CharacterizationReady Assets

Simulation-ready OpenUSD assets for materials characterization laboratories, packaged in the same way as NVIDIA SimReady Assets: one folder per asset, a root USD file over geometry and physics payloads, semantic labels, thumbnails, and a .metadata/ folder with a bill of materials, hashes and conformance results.

There is one addition. Some assets are functional instruments: the TEM declares its controls, states and constraints, and its measurements come from a simulator backend (TEM-view). Grids, holders and specimens carry the information that backend needs, so moving a grid into a holder and the holder into the microscope changes what the microscope measures.

Browse the assets in 3D, drive their joints, and run the site calculator in CharacterizationReady Central, a static Space that runs in your browser.

✓ present · ◐ partial · ✗ absent. The NVIDIA column comes from inspecting nvidia/simready-assets/bin_waste_a01; a ✗ there means "not in that asset", not "absent from the whole collection".

Feature	NVIDIA SimReady	CharacterizationReady v0.1
Geometry	✓ production meshes, instanced	◐ primitive shapes at true scale (metres, Z-up)
Visual materials	✓ shader materials + PBR textures (albedo, normal, ORM)	✗ display colour only
Rigid bodies, mass, colliders	✓ per part, with centre of mass and inertia	✓ per part: mass, centre of mass, principal inertia and axes computed from geometry (checked against Monte Carlo integration to within 0.3 %)
Physics materials (density, friction, restitution)	✓ PhysicsMaterialAPI	✓ PhysicsMaterialAPI, bound to every collider (typical handbook values)
Articulated joints	✓ revolute joints with limits (lid, pedal)	✓ revolute joints with limits: holder α-tilt (±30° / ±25°, eucentric), grid-box lid (0–110°), tweezers jaw (0 to −4°); named states map to joint angles
Physics-engine variants	✓ PhysX, MuJoCo, Newton (enabled / disabled)	◐ one engine-neutral physics variant
Semantic labels	✓ Wikidata class + Q-code; labelled physics materials	✓ project class + Wikidata class/Q-code in NVIDIA's layout for 11 of 44 assets; labelled physics materials. ◐ Holders, grid boxes, stubs, prep equipment and mitigation have no exact Wikidata item, and some lookups were rate-limited; TEM grid (Q115570755) is a stub
Grasp annotations	✓ Grasp scope with grasp-identifier frames	✓ the same Grasp scope layout (closing line, contact point and normal), plus the allowed tool and a link to the manifest frame
Typed interfaces (grid fits holder fits TEM)	✗	✓ provides / accepts, with capacities
Packaging BOM + hashes	✓	✓ same structure
Conformance report	✓ feature checks, failures recorded openly	✓ CR.001–CR.008
Readiness	◐ one conformance profile per asset	✓ three labels (scene / interaction / measurement) with evidence
Thumbnails	✓ rendered, three lighting variants	◐ simple-shape render, two sizes
Instrument state machine + operating constraints	✗	✓
Link to a measurement simulator	✗	✓ TEM-view
Environment coupling (vibration and field sources, mitigation, sensitivity)	✗	✓ S06 (illustrative values)
Application scenarios shipped with the data	✗	✓ S01–S06
Collection size	2,575 assets, 472 GB	44 assets + 3 scenes, 9.6MB

Assets (v0.1)

44 assets in 12 groups. All geometry is original and generic; nothing is derived from vendor CAD.

Instruments (4)

Asset	Name	Interfaces	Joints / simulator / site
equip_stereomicroscope_a01	Stereo inspection microscope	accepts sem.stub.pin, tem.chip.mems, tem.grid.3mm	—
instrument_fibsem_generic_a01	Generic FIB-SEM dual-beam	accepts sem.stub.pin, sem.stubholder, tem.grid.3mm	fibview (planned)
instrument_sem_generic_a01	Generic field-emission SEM	accepts sem.stub.pin, sem.stubholder	semview (planned)
instrument_tem_generic_a01	Generic TEM/STEM (80-300 kV), TEM-view backend	accepts tem.holder.side-entry	TEM-view; site sensitivity

TEM holders (6)

Asset	Name	Interfaces	Joints / simulator / site
tem_holder_cryoln2_a01	Side-entry LN2 cryo holder	provides tem.holder.side-entry; accepts tem.grid.3mm	alpha_tilt -25…25°; TEM-view holder=ln2
tem_holder_doubletilt_a01	Side-entry double-tilt holder	provides tem.holder.side-entry; accepts tem.grid.3mm	alpha_tilt -25…25°, beta_tilt -25…25°; TEM-view holder=room
tem_holder_heating_a01	MEMS heating holder	provides tem.holder.side-entry; accepts tem.chip.mems	alpha_tilt -20…20°; TEM-view holder=heat
tem_holder_lhe_a01	Liquid-helium cooling holder	provides tem.holder.side-entry; accepts tem.grid.3mm	alpha_tilt -20…20°; TEM-view holder=lhe
tem_holder_singletilt_a01	Side-entry single-tilt holder (room temperature)	provides tem.holder.side-entry; accepts tem.grid.3mm	alpha_tilt -30…30°; TEM-view holder=room
tem_holder_tomography_a01	High-tilt tomography holder	provides tem.holder.side-entry; accepts tem.grid.3mm	alpha_tilt -70…70°; TEM-view holder=room

TEM specimen supports (6)

Asset	Name	Interfaces	Joints / simulator / site
tem_chip_memsheating_a01	MEMS heating chip	provides tem.chip.mems	—
tem_chip_sinwindow_a01	SiN membrane window chip, 3 mm	provides tem.grid.3mm	—
tem_disc_3mm_a01	Specimen disc, 3 mm, dimpled	provides tem.disc.3mm, tem.grid.3mm	—
tem_grid_au300mesh_a01	TEM grid, Au, 300 mesh, holey carbon, 3.05 mm	provides tem.grid.3mm	—
tem_grid_cu200mesh_a01	TEM grid, Cu, 200 mesh, 3.05 mm	provides tem.grid.3mm	—
tem_grid_liftout_a01	FIB lift-out half grid, Cu, with posts	provides fib.liftout.post, tem.grid.3mm	—

Grid storage (3)

Asset	Name	Interfaces	Joints / simulator / site
tem_cassette_12slot_a01	Grid cassette, 12 slots	provides tem.cassette.12; accepts 12 × tem.grid.3mm	—
tem_gridbox_4slot_a01	Grid storage box, 4 slots	provides tem.gridbox.standard; accepts 4 × tem.grid.3mm	lid_hinge 0…110°
tem_gridbox_cryo_a01	Cryo grid box, 4 slots, rotating lid	provides tem.gridbox.cryo; accepts 4 × tem.grid.3mm	lid_rotation 0…90°

SEM specimen handling (4)

Asset	Name	Interfaces	Joints / simulator / site
sem_stub_pin12mm_a01	SEM pin stub, 12.7 mm	provides sem.stub.pin	—
sem_stub_pin25mm_a01	SEM pin stub, 25 mm	provides sem.stub.pin	—
sem_stubbox_a01	SEM stub storage box, 12 slots	provides sem.stubbox; accepts 12 × sem.stub.pin	lid_hinge 0…100°
sem_stubholder_6slot_a01	SEM stub holder, 6 slots	provides sem.stubholder; accepts 6 × sem.stub.pin	—

Cryo-EM (3)

Asset	Name	Interfaces	Joints / simulator / site
cryo_ln2dewar_a01	LN2 transport dewar, 4 L	—	—
cryo_plungefreezer_a01	Plunge freezer with humidity chamber	accepts tem.grid.3mm	—
cryo_workstation_a01	Cryo grid loading workstation	accepts tem.cassette.12, 4 × tem.gridbox.cryo	—

Sample preparation (4)

Asset	Name	Interfaces	Joints / simulator / site
prep_glowdischarge_a01	Glow-discharge unit for grids	accepts 12 × tem.grid.3mm	—
prep_ionmill_a01	Precision ion mill for 3 mm discs	accepts tem.disc.3mm	—
prep_plasmacleaner_a01	Plasma cleaner for TEM holders	accepts 2 × tem.holder.side-entry	—
prep_sputtercoater_a01	Sputter coater for SEM stubs	accepts 6 × sem.stub.pin, sem.stubholder	—

Tools (1)

Asset	Name	Interfaces	Joints / simulator / site
tool_tweezers_grid_a01	Grid tweezers, self-closing	provides grasp.tweezers	jaw_hinge -4…0°

Storage (2)

Asset	Name	Interfaces	Joints / simulator / site
storage_desiccator_a01	Vacuum desiccator	accepts 2 × sem.stubbox, tem.gridbox.cryo, 6 × tem.gridbox.standard	—
storage_holderstand_a01	TEM holder storage stand, 4 positions	accepts 4 × tem.holder.side-entry	—

Lab furniture (4)

Asset	Name	Interfaces	Joints / simulator / site
furniture_fumehood_a01	Fume hood, 1.5 m	—	—
furniture_glovebox_a01	Inert-atmosphere glovebox	accepts 2 × tem.gridbox.standard, tem.holder.side-entry	—
furniture_labbench_a01	Lab bench, 1.8 m	—	—
furniture_storagecabinet_a01	Storage cabinet, 2 m	—	—

Site sources (5)

Asset	Name	Interfaces	Joints / simulator / site
env_source_chiller_a01	Recirculating water chiller	—	site source
env_source_hvacduct_a01	Ceiling HVAC duct with fan coupling	—	site source
env_source_powerpanel_a01	Electrical distribution panel	—	site source
env_source_ups_a01	UPS cabinet	—	site source
env_source_vacuumpump_a01	Rotary-vane roughing pump	—	site source

Mitigation (2)

Asset	Name	Interfaces	Joints / simulator / site
mitigation_fieldcancel_a01	Active field cancellation system	—	site mitigation
mitigation_isolationplatform_a01	Pneumatic isolation platform	—	site mitigation

Site assets carry a site_physics block: what they emit, what they attenuate, or how sensitive an instrument is. The TEM also declares a sensitivity block: column resonance and damping, floor-to-specimen coupling, nm/µT at each kV, intrinsic information limits (1.0 Å uncorrected, 0.6 Å corrected), and an acceptance criterion. All site numbers are illustrative. They are physically plausible but not measured. Replace them with survey and vendor data before drawing conclusions about a real room.

Specimens (specimens/*.json; each binds to a TEM-view specimen): Au nanoparticles on carbon · Si / SiO₂ interface · apoferritin in vitreous ice (needs < 135 K, 60 e/Ų budget) · FeGe lamella (Lorentz modes).

catalog.jsonl has one row per asset (class, tags, mass, interfaces, readiness, conformance, paths) and is what the dataset viewer shows.

All geometry is original and generic. Nothing is derived from vendor CAD.

Package layout

tem_holder_singletilt_a01/
├── manifest.json                                     CharacterizationReady contract
├── com.characterizationready.packaging.json          package id, license, content + package hash
├── .metadata/
│   ├── LICENSE.txt
│   ├── com.characterizationready.root_usds.json      entry point(s)
│   ├── com.characterizationready.packaging.bom.json  every file: size + sha256
│   └── com.characterizationready.conformance.Package@0.1.0.json
└── usd/
    ├── tem_holder_singletilt_a01.usda                root: references payloads, `physics` variant
    ├── payloads/base.usda                            geometry, frames, semantic labels
    ├── payloads/Physics/physics.usda                 RigidBodyAPI, MassAPI, CollisionAPI
    └── .thumbs/  (256x256/…png, …_thumbnail.png)

Other top-level folders: scenes/ (assembled rooms), specimens/, scenarios/, spec/manifest.schema.json, tools/ (validator, packager, thumbnail renderer), adapters/ (TEM-view bridge).

The manifest

Each asset's manifest.json holds what the USD file alone does not:

• frames: named attachment Xforms. A holder's InsertionAxis aligns to the TEM's SpecimenPort, and a grid's SpecimenFrame to a holder's GridSeat.
• interfaces: typed provides / accepts slots. Something fits only if the types match.
• grasp_regions: where a hand, gripper or tweezers may hold it.
• states: allowed actions per state (for the TEM: empty → airlock → holder_in → ready).
• backend: simulator parameters this asset contributes (holder type, temperature range, kV, modes, which analysis runs in which mode).
• readiness: three independent labels, each not_applicable | none | claimed | validated:
  • scene: scale, frames, labels, usable geometry
  • interaction: tested motions, collisions, interfaces, state transitions
  • measurement: named forward model, documented assumptions, reference tests
• limitations: what the asset does not support, in plain words.

Conformance checks

tools/validate.py runs these on every asset. The results are written to each package's conformance file:

Check	Requirement
CR.001	manifest matches the schema and its folder name
CR.002	root layer composes with no unresolved layers; units / up axis / default prim match
CR.003	every prim named in frames, interfaces and grasp regions exists
CR.004	default prim has a SemanticsLabelsAPI:class label equal to the manifest class
CR.005	colliders present with physics enabled, absent when disabled; USD mass = manifest mass
CR.006	no readiness level is validated without evidence
CR.007	instruments declare a backend binding and a state machine
CR.008	site_physics: sources have an emitter frame and a spectrum; mitigations have their parameters and a frame; instrument sensitivity covers every kV the instrument runs at
CR.009	every manifest joint exists as a USD revolute joint with the same axis, limits and bodies; bodies are rigid bodies; holder tilt limits equal the TEM-view binding; named state angles are within limits
CR.010	every rigid body has positive mass, a centre of mass and principal inertia obeying the triangle inequality; every collider is bound to a physics material with valid density, friction and restitution
CR.011	a Grasp scope with one grasp_identifier (closing-line curve, tool, frame link) per manifest grasp region

CR.004 also checks Wikidata labels where the manifest declares them. Every v0.1 asset passes all eleven. Every readiness level is at most claimed. Nothing has been checked against a physical instrument.

Application scenarios

Each scenario is a runnable script in scenarios/ with its outputs checked in under out/.

S01 · Lab layout and clearance planning

Does a proposed room layout keep the TEM usable? Checks the holder insertion corridor (the volume swept by a 0.34 m holder plus hand room along the specimen-port axis), the operator zone at the console, and that nothing sinks below the floor. It uses only the assets' frames and bounds. Result: the shipped layout passes. A proposal that moves the bench next to the column fails: the bench, grid box and tweezers fall inside the operator zone.

S02 · Robotic grid transfer

A pick-and-place task: tweezers take a grid from its box slot and place it on the holder's grid seat. task.json gives the success criteria (≤ 50 µm placement error, compatible interfaces, allowed grasp tool, grid stays above the bench). waypoints.jsonl is a reference trajectory. A scripted kinematic baseline solves it in 10 steps, moving the grid 0.86 m. It uses the joints: the closed grid-box lid blocks the tweezers' approach until the lid hinge is opened to 110°; the jaws open to −4°, an 8.4 mm gap at the tip; and the holder cradle tilts to ±30° with the grid seat moving 0 µm. The tilt axis passes through the seat by design, so this check flags any future geometry change that breaks eucentricity. The task is meant for learned policies or physics rollouts: load the scene with the physics variant enabled. The baseline itself does no contact simulation.

S03 · Autonomous acquisition benchmark

An agent gets the Si / SiO₂ specimen, an empty TEM and a 1500 e/Ų dose budget. It must report a lattice spacing within 2% of a true Si reflection. All actions go through the instrument's state machine and constraints, and each acquisition becomes a TEM-view request. Baseline result (TEM-view): a CTF focus check (fitted −53.1 nm vs −50 nm set) followed by SAED indexed (111) 3.115 Å, (220) 1.917 Å, (311) 1.638 Å and (400) 1.360 Å, all within 0.7%. Total dose was 300 / 1500 e/Ų, with 0 refused actions. A first version of the baseline also ran lattice analysis on an HRTEM image. TEM-view only indexes spacings in diffraction mode, so that step wasted 800 e/Ų. The instrument manifest now declares which analysis runs in which mode, and the adapter refuses such a request before any dose is spent.

S04 · Cryo-EM procedure training

An operator's sequence for a vitreous-ice specimen is replayed against the assets. Four of its 13 steps are refused, each with the reason a real lab would give:

1. loading the ice grid in the room-temperature holder ("needs 4–135 K but … holds 293.1 K")
2. turning the beam on with no holder inserted
3. warming the stage to 150 K
4. a fourth 20 e/Ų exposure once the 60 e/Ų budget is used up

The constraints come from asset and specimen metadata, not from the script. The scenario can serve as a procedure trainer or as guardrails an AI operator must respect.

S05 · Labelled synthetic data

A defocus × thickness sweep (Au on C, 300 kV, realistic detector). Each sample's labels come from the asset chain: instrument, holder, grid and specimen versions, the exact TEM-view request, and the instrument state. The same pattern scales to training sets for CTF estimation, thickness regression or denoising.

S06 · Site assessment: vibration, stray fields and fault diagnosis

Scene scenes/tem_site.usda holds a TEM installation (with its roughing pump), a chiller and an electrical panel. Two USD variant sets define the options: placement (A near the panel wall, B across the room) and mitigation (none / isolation / cancellation / both). Every what-if is a variant selection, so there is one scene file and no copies.

For each option, adapters/site_physics.py takes the world positions of every source, mitigation and the instrument from the composed stage, then propagates each disturbance:

• Vibration: geometric spreading and material loss → isolation transmissibility → column relative-motion response → coupling.
• Fields: falloff → cancellation → nm/µT.

The result is a specimen-plane displacement covariance Σ. Motion during an exposure multiplies the image spectrum by E(q) = exp(−2π² qᵀΣq). The information limit from the environment is π·√λmax(Σ), combined in quadrature with the instrument's intrinsic limit.

1 · Placement × mitigation (corrected TEM, 300 kV; acceptance ≤ 0.66 Å):

	none	cancellation	isolation	both
A (near panel)	3.33 Å	0.89 Å	3.20 Å	0.62 Å ✓
B (across room)	1.32 Å	0.89 Å	1.15 Å	0.60 Å ✓

At A, the panel's 60 Hz field dominates, so isolation alone buys almost nothing. Cancel the field, and the pump becomes the limit. Only both together pass, with little margin at A and more at B. The floor beside the column rates VC-B because of the installation's own pump, which suggests moving the pump as a cheaper fix worth costing.

2 · What each option records: the TEM-view reference image (corrected, Au, 300 kV) blurred by each option's Σ. Stray fields give directional streaks and a spectrum squashed along one axis. Vibration gives isotropic loss. The Au lattice reflections disappear once the information limit passes about 2 Å.

3 · Fault diagnosis. After commissioning at A with both mitigations, one fault is injected at a time:

• the pump's isolation mount fails (5× emission)
• the isolation platform loses air
• field cancellation is switched off

The diagnosis receives only the degraded image and a floor-accelerometer plus column-magnetometer survey. It estimates Σ from the image and scores three explanations: survey-scaled sources with all mitigations working, with the platform failed, or with cancellation failed. Each explanation is scored on how well it reproduces the image blur, plus a penalty for needing large changes in source amplitude.

Injected fault	Finding	Information limit
pump mount failed	pump emission 4.7× commissioning; mitigations intact	0.63 Å (still within acceptance; isolation absorbs it)
platform lost air	platform not working	0.84 Å (outside)
cancellation off	field cancellation not working	2.99 Å (outside)

All three were diagnosed correctly in 60 of 60 runs (20 noise seeds each). This separates "a source got worse" from "a mitigation stopped working", which is the distinction that tells facility staff whether to fix the pump mount or the platform's air supply.

Using real survey data

The background survey (scenes/tem_site.background_survey.json) and the fault surveys (scenarios/s06_site_assessment/out/survey_*.json) are plain JSON:

{
  "floor_velocity_bands": {"bands_hz": [1, 1.25, ..., 200], "velocity_um_s": [1.5, ...]},
  "accelerometer_tones":  [{"f_hz": 24.6, "velocity_um_s": 21.0}],
  "magnetometer_tones":   [{"f_hz": 60.0, "b_uT": 0.017}]
}

To assess a real room:

1. Put your one-third-octave floor survey into the background file.
2. Replace the spectra in the source manifests with measured ones.
3. Measure the instrument's sensitivity block.
4. Run python scenarios/s06_site_assessment/run.py --survey measured.json.

Quick start

pip install -r requirements.txt
hf download rmsreis/characterization-ready-assets --repo-type dataset --local-dir characterization-ready
cd characterization-ready
python tools/validate.py .
python scenarios/s01_lab_layout_clearance/run.py
python scenarios/s04_cryo_workflow/run.py

Download a single asset:

hf download rmsreis/characterization-ready-assets --repo-type dataset --include "tem_holder_cryoln2_a01/*" --local-dir .

Use an asset in your own stage by referencing its root file. Pick the physics variant you need:

def "Holder" (
    prepend references = @./tem_holder_cryoln2_a01/usd/tem_holder_cryoln2_a01.usda@
    variants = { string physics = "enabled" }
) {}

TEM-view backend

TEM-view is a physics-based TEM simulator and is not bundled here. adapters/temview_adapter.py turns the loaded instrument, holder and specimen into a TEM-view simulate request, after checking states, interfaces, temperatures, analysis/mode pairs and dose. Scenarios S03–S05 write these requests to out/requests.jsonl. The checked-in .npz outputs and responses were produced by running those requests through TEM-view.

Limitations

• Geometry is built from simple shapes for layout, collision and handling; it is not photoreal and not vendor-accurate. There are no articulated joints yet: the grid-box lid, tweezers jaws and holder tilt are logical states.
• Holder insertion is a frame alignment. It says nothing about whether robotic loading of a real microscope is safe.
• Measurement fidelity is TEM-view's. The room holder pins 293.1 K. TEM-view reported 105.8 K for a 100 K request on the LN2 holder.
• Site physics (S06) is a frequency-domain budget model, not a structural or field solver. Isolation is single-degree-of-freedom; the column is one resonance; fields are radial from point sources; lens image rotation, acoustic coupling and thermal drift are not modelled; background floor vibration is uniform. The degraded images are TEM-view images blurred after simulation, which is exact for specimen motion during an exposure but does not capture effects inside the column such as beam-tilt modulation.
• Do not treat success in simulation as authorization to operate real hardware.

Roadmap

• SEM (SEM-view) and FIB (FIB-view) instruments with stub and lamella interfaces, plus a FIB lift-out → TEM grid hand-off
• Articulated USD physics joints for the holder tilt, grid-box lid and tweezers
• Per-engine physics variants, as in SimReady
• Corrector-tuning benchmark from Ronchigrams. This needs TEM-view to accept individual aberration coefficients.
• Acoustic and thermal-drift channels for S06; finite-element column modes attached to the instrument asset
• Validation of S06 against a real site survey and real images (this is what would earn measurement: validated)
• Backend output rendered onto the TEM monitor prim

Working from source

The tools need OpenUSD (usd-core). Create a project environment with python3 -m venv .venv && .venv/bin/pip install -r requirements.txt, then run the tools with .venv/bin/python. The source tree keeps assets under assets/<asset_id>/. python tools/enrich_assets.py regenerates each asset's physics payload (rigid bodies, mass properties, colliders, physics materials, joints) and its semantic and grasp annotations from the geometry plus the manifest. python tools/package_hf.py dist/hf builds this release layout: it adds .metadata/, the packaging files and catalog.jsonl, and rewrites scene references. python tools/render_thumbnails.py regenerates the previews.

Citation

DOI: 10.57967/hf/11308. Machine-readable metadata is in CITATION.cff.

@misc{dosreis2026characterizationready,
  author       = {dos Reis, Roberto},
  title        = {{CharacterizationReady Assets}: simulation-ready {OpenUSD} assets for materials characterization laboratories},
  year         = {2026},
  version      = {0.1.0},
  publisher    = {Hugging Face},
  doi          = {10.57967/hf/11308},
  url          = {https://huggingface.co/datasets/rmsreis/characterization-ready-assets}
}

License

CC BY 4.0. Author: Roberto dos Reis.