openbot-data 0.0.3 · updated 2026-09-02

Validate a LeRobot Dataset Before ACT or SmolVLA Training

Run a local, full-integrity LeRobot readiness check and preserve a machine-readable artifact before starting ACT or SmolVLA training.

Direct answer

Before an ACT or SmolVLA run, validate the exact LeRobot revision with a policy-specific profile and full media integrity. Keep the JSON artifact beside the training config. A clean sample audit is useful triage, but it does not establish full readiness or guarantee convergence.

Run the check

openbot-data readiness ./lerobot_dataset \
  --profile lerobot-act \
  --integrity full \
  --out ./act.readiness.json \
  --markdown ./act.readiness.md

Machine-readable output: act.readiness.json (`openbot.dataset_readiness.v1`)

What this proves

Which ACT profile requirements were evaluated by the installed tool version.
Whether the evaluated local revision is READY, BLOCKED, or PARTIAL.
Which checks ran, which capabilities were skipped, and which blockers were found.

What it does not prove

That the dataset license permits your intended commercial or research use.
That training will converge or reproduce a paper result.
That a later dataset revision has the same readiness result.

Decision checklist

01Pin or record the exact local/source revision before running the command.
02Use full integrity when you need frame and media coverage, not only metadata checks.
03Keep the JSON artifact with the training configuration and dataset revision.
04Repeat with `--profile lerobot-smolvla` when SmolVLA is the target policy.

Reviewed sources

Reviewed by OpenBot Catalog team

openbot-data getting startedCurrent CLI inputs, integrity modes, and output contracts.Hugging Face LeRobotUpstream dataset and policy ecosystem.