---
license: mit
language:
  - en
tags:
  - spatial-reasoning
  - embodied-ai
  - benchmark
  - 3d-simulation
pretty_name: SpatialWorld — Task Card Metadata
configs:
  - config_name: default
    data_files:
      - split: v1.0
        path: task_cards.parquet
---

# SpatialWorld — Task Card Metadata (v1.0)

**760 tasks** from [SpatialWorld](https://spatial-world.github.io/).

| Column | Notes |
|---|---|
| `task_id` | Task identifier |
| `task_name` | Short title (from `task.json` when available) |
| `instruction` | Natural-language instruction |
| `backend` | Simulator backend |
| `scene` | Scene / map identifier |
| `category` | Scenario category (`Daily`, `Work`, `Travel`, `Entertain`, `Social`, …) |
| `task_type` | `Navigation`, `Interaction`, or `Hybrid` |
| `data_path` | Path to task assets in `data/` |

Golden trajectories: indoor/outdoor tasks in each `task.json`; game tasks in `data/game/game_gt.csv`.

## Download

```bash
git clone https://huggingface.co/datasets/HongchengGao/SpatialWorld
```

```python
from huggingface_hub import snapshot_download
snapshot_download(repo_id="HongchengGao/SpatialWorld", repo_type="dataset", local_dir=".")
```