DOMINO-Extended / README.md
HanningHanning's picture
Simplify data layout to task and scene
3f7021a verified
|
Raw
History Blame Contribute Delete
1.58 kB
---
pretty_name: DOMINO Extended Online-Follow
language:
- en
license: apache-2.0
task_categories:
- robotics
tags:
- robotics
- embodied-ai
- dynamic-manipulation
- vision-language-action
- domino
- aloha
- video
---
# DOMINO Extended: Online-Follow
This repository contains online-follow demonstrations for multiple robot tasks.
Because every entry uses the same follow-style collector, the data hierarchy only
distinguishes **task** and **scene**.
Current data: **1 task/scene entries, 50 episodes**.
| Task | Scene | Episodes | Speed (m/s) | Details |
|---|---|---:|---:|---|
| `adjust_bottle` | `dynamic-clean` | 50 | 0.040--0.117 | [metadata](./metadata/adjust_bottle_dynamic-clean.json) |
## Layout
```text
dataset/<task>/<embodiment>_<scene>_<episodes>.zip
metadata/<task>_<scene>.json
provenance/<task>_<scene>/
catalog.json
```
## Download the current entry
```bash
hf download HanningHanning/DOMINO-Extended dataset/adjust_bottle/aloha-agilex_dynamic-clean_50.zip --repo-type dataset --local-dir DOMINO-Extended
```
The bottle starts outside the camera view and moves independently. The expert
observes it online, approaches, follows, grasps, lifts, and completes the original
upright placement task without access to a future object trajectory or a precomputed
interception point. Raw HDF5 archives are canonical; converted LeRobot copies and
runtime caches are excluded.
## License and attribution
Released under Apache-2.0, following the upstream DOMINO repository and dataset.
See each metadata file for its integrity checks and provenance snapshot.