---
pretty_name: Isaac Material Property Seed Dataset
license: other
task_categories:
  - other
tags:
  - materials-science
  - robotics
  - isaac-sim
  - physics-simulation
  - pbr
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train.jsonl
---

# Isaac Material Property Seed Dataset

A lightweight material-property catalogue intended to support material
retrieval and automatic material assignment in NVIDIA Isaac Sim and related USD
simulation workflows.

The initial release contains 150 major material subgroups. 

## Schema

Identity fields:

- `material_description`
- `material_class`
- `material_family`

Physical parameters:

- `rho`: density, canonical unit `kg/m^3`
- `E`: Young's modulus, canonical unit `Pa`
- `nu`: Poisson's ratio, dimensionless
- `mu_s`: static-friction coefficient, dimensionless
- `mu_d`: dynamic-friction coefficient, dimensionless
- `e`: coefficient of restitution, dimensionless
- `G`: shear modulus, canonical unit `Pa`
- `K`: bulk modulus, canonical unit `Pa`

Each physical parameter has exactly three fields:

- `<parameter>_value`: nullable floating-point value in the canonical unit
- `<parameter>_source`: nullable URL, citation, or stable source identifier
- `<parameter>_dt_obtained`: nullable ISO 8601 timestamp recording when the value was obtained

`schema.json` is the machine-readable schema and unit contract. `data/train.jsonl`
is the canonical Hub data file; `data/train.csv` is included for convenient manual
editing and spreadsheet workflows.

## Curation rules

1. Never populate a numeric value without its source and acquisition timestamp.
2. Convert all values into canonical units before storage.
3. We prefer representative subgroup values or midpoints over false grade-level precision.
4. Preserve ranges and test conditions in the source record during later enrichment.
5. Treat friction and restitution as simulation priors unless their counterface and
   measurement conditions are known.
6. Values derived from other fields must identify the derivation in the source field.

Useful isotropic derivations, when `E` and `nu` are grounded, are:

```text
G = E / (2 * (1 + nu))
K = E / (3 * (1 - 2 * nu))
```