embodied-data 0.3.0

Bidirectional converter and validator for AgiBot World ↔ LeRobot v3 datasets.

embodied-data

Bidirectional converter and validator for AgiBot World ↔ LeRobot v3 datasets.

What it does

• Bidirectional conversion between AgiBot World (DigitalWorld sim + Beta/Alpha real hardware) and LeRobot v3.
• Schema-detect dispatcher — point convert at any AgiBot root and the right reader fires automatically.
• Five-check validator — schema conformance, fps consistency, timestamp monotonicity, action-dim consistency, frame ↔ video alignment.
• Batch + resume — --max-episodes for parallel conversion, meta/uuid_map.parquet for restartable jobs.
• Stdlib-first — h5py + pyarrow + av; no PyTorch dependency in the data path.

Quick start

LeRobot's pusht is the fastest end-to-end check (no HuggingFace gating, ~30 s):

pip install --upgrade embodied-data
huggingface-cli download lerobot/pusht --repo-type dataset --local-dir ./pusht

embodied-data preview  ./pusht
embodied-data validate ./pusht

preview prints a per-episode stats table; validate runs all five checks and exits non-zero on failure.

Real AgiBot data (HuggingFace gated)

AgiBot World Beta and Alpha live on HuggingFace under a gated license. Request access on the AgiBotWorld-Beta page first, then:

huggingface-cli login
huggingface-cli download agibot-world/AgiBotWorld-Beta \
    --repo-type dataset \
    --include "task_info_675.json" "observations/675/936938/**" "proprio_stats/675/936938.h5" \
    --local-dir ./agibot_beta_root

embodied-data convert \
    ./agibot_beta_root/675/936938 \
    /tmp/beta_v3 \
    --from agibot --to lerobot-v3

embodied-data validate /tmp/beta_v3

For batch conversion of a whole task, point convert at the task root and pass --max-episodes N. Streaming-extraction tips for partial Beta downloads are in docs/schema/beta.md.

Validation example

Why this exists

Robotics researchers spend days rewriting the same dataset conversion scripts. AgiBot World's official convert_to_lerobot.py has carried unresolved issues for months; LeRobot's v2.0 / v2.1 / v3.0 versions break each other; every lab writes its own timestamp alignment check. This tool is the layer that stops.

Concrete upstream issues this project addresses or works around:

• AgiBot-World #18 — task_info_*.json lookup ambiguity for sub-roots
• AgiBot-World #124 — Beta vs Alpha schema divergence
• AgiBot-World #149 — proprio HDF5 key drift across batches
• lerobot #2158 — v2 ↔ v3 episode-index incompatibility
• lerobot #2689 — fps/timestamp validation gap

Roadmap

• v0.3 (shipped on main, awaiting tag) — observation.images.head_color video for Beta / Alpha (single + batch) so v3 datasets are usable for VLA training end-to-end.
• v0.3.x patches — multi-camera (fisheye / hand / back), sparse action/*/index masks, end-pose flattening, reverse Beta path (see docs/v0.3.x-patches.md).
• v0.4+ — ALOHA HDF5 ingest, RLDS export, OpenX Embodiment alignment.

Cross-embodiment action-space retargeting and Chinese prompt embedding remain explicit non-goals.

Schema reference

• docs/schema/overview.md — AgiBot variant matrix
• docs/schema/digitalworld.md — DigitalWorld (sim) layout
• docs/schema/beta.md — Beta / Alpha (real hardware) layout
• docs/schema-lerobot-v3.md — LeRobot v3 target schema

Install

pip install embodied-data
embodied-data --help

Python 3.12+ required.

Development

git clone https://github.com/allenwu-blip/embodied-data.git
cd embodied-data
uv sync
uv run pytest

Coverage

• 56 commits, 3 PyPI releases (0.1.0 / 0.1.1 / 0.2.0); v0.3 staged on main
• 114 passing tests + 1 skipped (gated dataset)
• 4 upstream issue threads engaged
• 4 HuggingFace datasets exercised end-to-end (lerobot/pusht, AgiBotWorld-Beta, AgiBotWorld-Alpha, agibot-world/agibot_digital_world)

Acknowledgments

• HuggingFace LeRobot team for the v3 schema and reference datasets
• OpenDriveLab AgiBot World team for releasing Beta and Alpha under HF gating

License

MIT — see LICENSE.

Contact

Bug reports and feature requests: GitHub Issues.