SWAM-Pretrain / README.md
FreDERicia's picture
Restore dataset card overview
2f79a64 verified
|
Raw History Blame Contribute Delete
7.16 kB
metadata
pretty_name: SWAM Pretraining Data
tags:
  - robotics
  - world-action-models
  - video
  - egocentric-video
  - lerobot
  - av1

SWAM-Pretrain

SWAM project cover

GitHub codeProject websitearXiv paper

SWAM pretraining data is organized into General, Ego, UMI, and Robotics, covering general videos, egocentric videos, UMI manipulation demonstrations, and robot demonstrations. We standardize the video format, preserve source-specific episode records and annotations, and add detailed annotations for manipulation data. This repository releases the General, Ego, and Robotics data listed below.

Uploads and verification are complete, with 27 subsets, 3,839 TAR archives, and approximately 3.98 TB of data. Videos are standardized to 320×256, AV1-encoded MP4. Browse all data

Data scope

Category Data sources Release scope
General OpenVid, Panda70M OpenVid and Panda70M released
Ego EgoVerse, Ego4D, EgoSuite EgoVerse released. Ego4D and EgoSuite excluded from this release
UMI UMI manipulation demonstrations Awaiting a future release by Noematrix
Robotics AgiBot World (Alpha, Beta, 2026 NOE), RoboCoin, RoboMIND, InternData A1, Galaxea, ABC-130K All listed subsets released

Ego4D and EgoSuite are excluded due to upstream licensing and redistribution restrictions. UMI data awaits a future release by Noematrix, with release arrangements to be announced through its official channels.

Research data statistics

The figures below describe the paper's full pretraining data pool of approximately 47,000 hours, including Ego4D, EgoSuite, and UMI, which are excluded from this release. The scope of this repository is listed in the table above.

Research data statistics

The left panel shows data sources and video hours. The right panel shows shares of total video hours and episodes, and video-duration distributions within each category.

Data format and directory structure

  • Videos are 320 pixels wide by 256 pixels high, AV1 encoded in MP4 containers. Source frame rates and temporal ordering are preserved.
  • Parquet files retain the actions, states, and language annotations provided by each source. Available fields vary across datasets.
  • Source-specific metadata is retained, with video-related dimensions and codec information updated for the processed files.
  • Data follows a LeRobot-style layout with data/, meta/, and videos/ directories and is distributed as TAR archives. See each source's meta/info.json for field definitions and action conventions.

Archives are organized by category and dataset. Released videos are standardized to 320×256.

data/
├── general/
│   ├── panda-70m/
│   └── openvid/
│       ├── source-512x512-24fps/
│       ├── source-832x448-24fps/
│       └── source-832x448-30fps/
├── ego/
│   └── egoverse/
│       ├── source-368x640/
│       └── source-480x640/
└── robotics/
    ├── agibot-world-alpha/
    ├── agibot-world-beta/
    ├── agibot-world-2026-noe/
    │   ├── il/
    │   ├── il-highres-wrist/
    │   ├── rich-interaction/
    │   └── rl/
    ├── robocoin/
    │   ├── cobot-magic-source-640x480/
    │   ├── cobot-magic-source-1280x720/
    │   ├── galaxea-r1-lite/
    │   └── galbot-g1/
    ├── robomind/
    │   ├── v1-agilex/
    │   ├── v2-agilex/
    │   ├── v2-ark/
    │   └── v2-franka/
    ├── interndata-a1/
    │   ├── agilex-split-aloha/
    │   ├── arx-lift-2/
    │   ├── genie-1/
    │   └── piper/
    ├── abc-130k/
    │   ├── train/
    │   └── val/
    └── galaxea/

Each subset directory contains metadata.tar and several chunk-*.tar archives. Extract all archives for a subset into the same target directory. For example, Galaxea uses the following layout.

archives/data/robotics/galaxea/
  metadata.tar
  chunk-000.tar
  ...
  chunk-020.tar

# After extraction
data/robotics/galaxea/
  meta/
  data/chunk-000/...
  videos/chunk-000/...
  ...

Download and extract

Install the Hugging Face CLI.

pip install -U huggingface_hub

Download the complete Galaxea subset.

hf download FreDERicia/SWAM-Pretrain \
  --repo-type dataset \
  --include "data/robotics/galaxea/*.tar" \
  --local-dir ./archives

Extract it on Linux or macOS.

mkdir -p ./data/robotics/galaxea
for archive in ./archives/data/robotics/galaxea/*.tar; do
  tar -xf "$archive" -C ./data/robotics/galaxea
done

For a smaller download, fetch metadata.tar and one chunk.

hf download FreDERicia/SWAM-Pretrain \
  data/robotics/galaxea/metadata.tar data/robotics/galaxea/chunk-019.tar \
  --repo-type dataset --local-dir ./sample-archives

Extract the sample archives into the same directory and load only the downloaded episodes. Video readers need AV1 decoding support.

Source terms and attribution

Each source's license, attribution requirements, and usage restrictions continue to apply, including non-commercial restrictions where specified. Please follow the relevant source terms and cite the corresponding work.

Code and citation

The accompanying SWAM code repository integrates RLinf, DreamZero, DMuon, and RoboTwin, with entry points for training, checkpoint conversion, and RoboTwin evaluation. The code repository is public. SWAM code repository

If you use this dataset, please cite the SWAM paper.

@misc{zhou2026swam,
  title = {SWAM: Scaling World Action Models with Cross-Modal Alignment},
  author = {Yu Zhou and Xinyuan Xia and Xuankun Yang and Yinghao Yang and Wendi Chen and Bowen Liu and Jianhan Ma and Weicheng Duan and Jun Lv and Cewu Lu and Yanwei Li},
  year = {2026},
  eprint = {2611.12870},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.12870}
}