# Placement Task Asset Notes

This repository is intended to support placement-oriented robotics tasks such as:

- placing objects into containers
- placing tableware or food on plates, trays, counters, and shelves
- loading objects into articulated fixtures such as dishwashers, drawers, and cabinets
- composing tabletop or kitchen scenes from reusable object, receptacle, fixture, robot, and environment assets

## Currently available local assets

For any imported source, the upstream-to-local directory mapping is recorded in:

```text
docs/structure_mapping.md
```

### RoboCasa

RoboCasa currently contributes dishwasher fixtures:

```text
assets/robocasa/raw/fixtures/dishwashers/
assets/robocasa/raw/fixtures/fixture_registry/dishwasher.yaml
```

These are useful for placement tasks involving loading dishes or objects into a dishwasher.

### DISCOVERSE

DISCOVERSE is imported as a collection-level asset:

```text
assets/discoverse/raw/collections/models/discoverse_models/
├── mjcf/
├── meshes/
└── urdf/
```

Placement-relevant DISCOVERSE MJCF task/environment files include:

```text
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/place_block.xml
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/place_coffeecup.xml
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/place_jujube.xml
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/place_jujube_coffeecup.xml
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/place_kiwi_fruit.xml
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/place_spoon.xml
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/coffeecup_plate.xml
assets/discoverse/raw/collections/models/discoverse_models/mjcf/task_environments/stack_block.xml
```

Placement-relevant DISCOVERSE object assets include bowls, plates, cups, fruit, boxes, baskets, pans, spoons, towels, trashbins, and cabinets under:

```text
assets/discoverse/raw/collections/models/discoverse_models/mjcf/object/
assets/discoverse/raw/collections/models/discoverse_models/meshes/object/
```

## Candidate external sources

The following sources are recorded as link-only candidates. Do not copy assets from them into this repository until a downstream task actually uses them and the license has been reviewed.

| Source | Link | Placement relevance | Intake rule |
| --- | --- | --- | --- |
| MuJoCo Menagerie | https://github.com/google-deepmind/mujoco_menagerie | Robot grippers, robot arms, hands, and robot-ready MJCF assets | Import exact used robot/object model with license and upstream commit. |
| PartNet-Mobility | https://sapien.ucsd.edu/browse | Articulated cabinets, drawers, doors, and household objects | Import only selected object IDs after checking license and conversion notes. |
| Objaverse / Objaverse-XL | https://objaverse.allenai.org | Large-scale everyday objects for object placement and scene variety | Import exact used asset IDs with original UID, license, source URL, and conversion pipeline. |
| Google Scanned Objects | https://app.ignitionrobotics.org/GoogleResearch/fuel/collections/Google%20Scanned%20Objects | Scanned household objects suitable for manipulation and placement | Import exact used model with provider URL and license metadata. |
| RoboCasa | https://robocasa.ai | Kitchen fixtures, objects, scenes, textures, and placement-style tasks | Import normalized assets under `assets/robocasa/...`. |
| DISCOVERSE | https://github.com/discoverse-dev/DISCOVERSE | MJCF task environments, manipulation objects, robots, and scenes | Already imported as `assets/discoverse/raw/collections/models/discoverse_models/`. |

## Third-party intake checklist

When a placement task uses an external asset:

1. Copy the exact used asset into this repository.
2. Preserve upstream filenames and relative paths where practical.
3. Add `metadata.yaml` in the asset directory.
4. Add one JSONL row to `manifest/assets.jsonl`.
5. Record license, upstream URL, retrieval date, source asset ID, and conversion steps.
6. Run:

```bash
python scripts/validate_asset.py --all
```