# SoftVTBench — Evaluation Assets (`eval-assets/`)

USD assets required to rebuild the simulation scene for **closed-loop evaluation**
in Isaac Sim. Training does not need this bundle — only evaluation does.

## Layout

```
eval-assets/
└── USD/
    ├── basket/basket.usd + textures/          # LIBERO scene objects (32 dirs)
    ├── kitchen_table/ plate/ microwave/ ...
    ├── pastry001/pastry001.usd + resource/    # SoftVTBench deformable assets (11 dirs)
    ├── pastry002/ ... pastry006/ pastry010/
    ├── pastry005_visual_distractor/           # spatial-soft distractor variant
    └── stw_cube_hq/ stw_sphere_hq/ stw_cylinder_hq/
```

Every asset follows the `<name>/<name>.usd` convention that the evaluation
scripts resolve against.

## Usage

Point the evaluation scripts at this directory:

```bash
export SOFTVT_EVAL_USD_DIR=/path/to/eval-assets/USD
```

## Provenance & licensing

- **Scene objects** (`basket`, `kitchen_table`, `akita_black_bowl`, `ketchup`,
  `milk`, `plate`, `microwave`, and the other rigid props) originate from the
  [LIBERO](https://github.com/Lifelong-Robot-Learning/LIBERO) benchmark
  (MIT License), converted to USD for Isaac Sim by
  [Tabero](https://github.com/NathanWu7/Tabero) (Apache-2.0). They are
  redistributed here unmodified, purely so that closed-loop evaluation is
  reproducible without re-running the original conversion pipeline.
- **Deformable assets** (`pastry001`–`pastry011`, `stw_*_hq`,
  `pastry005_visual_distractor`) are contributed by SoftVTBench and are released
  under the same Apache-2.0 license as the main repository.

If you use these assets, please cite LIBERO and Tabero in addition to
SoftVTBench.

## Note on the `soft-assets/` folder

The separate `soft-assets/` folder in this dataset holds the *authoring-time*
copies of the deformable assets (named `model_pastryN.usd`). The copies here
under `eval-assets/USD/` are the *simulation-ready* variants (named
`pastryNNN.usd`) that the evaluation scene loader expects. They are not
interchangeable — use `eval-assets/USD/` for evaluation.