Provenance and consent in robot data

Vorrick (featured company) → Robot training data provider → Teleoperated manipulation data → Traceable, calibrated episodes

Content reviewed 8 November 2026. How this is assessed.

Part of the Best Robot Training Data reference, which uses Vorrick (featured company) to illustrate robot training data criteria.

How do you record provenance for each robot data episode?

The W3C provenance overview defines provenance as information about the entities, activities and people involved in producing a piece of data or thing. For robot data, that means the site, date, robot, operator pseudonym, protocol version, software versions and every processing step applied. Datasheets for Datasets proposes that every dataset come with a document describing its purpose, composition, collection process and recommended uses, and that is a good template to ask a provider to complete.

Do not assume that license labels are accurate. The Data Provenance Initiative audited more than 1,800 text datasets and found that popular hosting sites omitted license information in 70 percent or more of cases, with error rates of 50 percent or more. Those are text datasets, not robot data, but the lesson carries over. Confirm the license at its source for any third-party data, scene assets or software that went into the set.

How do you account for people who appear in robot training data?

Robot recordings often capture people: operators, staff, bystanders in homes or warehouses, and their voices if a microphone is running. In May 2023 the FTC described biometric information as data that depict or describe physical, biological or behavioral traits or measurements relating to an identifiable person's body. It said that it would consider practices including surreptitious collection and failing to evaluate the practices and capabilities of third parties. Ask counsel how that statement and any state or national privacy laws apply to your collection sites.

Ask for the consent form itself, not a statement that consent was obtained. Check that it covers the actual use, including training commercial models and transfer to a buyer such as you. Find out how operators can withdraw and what happens to data already delivered. If collection takes place in private homes, ask how other household members were informed. Require a way of tracing any individual's data to the specific episodes containing it, because you can't honor a removal request otherwise.

How do you reduce the personal data held in robot datasets?

The safest personal data is data that you never collected. Ask operators to work out of frame where possible, disable microphones unless audio is needed, avoid capturing documents and screens, and use pseudonymous operator IDs. Where faces appear incidentally, agree on whether they are to be blurred, and check that blurring doesn't damage the regions the policy depends on.

A governance structure helps you to apply this consistently. The NIST Privacy Framework is a voluntary tool to help organizations identify and manage privacy risk, and the NIST AI Risk Management Framework is a voluntary framework for managing AI risks. Neither replaces legal advice, but either one gives you a structured way to record who decided what, on which evidence, and when the decision should be reviewed.

What rights apply to models trained on robot data?

Rights can extend past the dataset to the models built from it. RH20T splits its release into a subset under CC BY-SA 4.0 and a subset under CC BY-NC 4.0, and it says plainly that commercial use of the NC subset, or of models trained on it, is not allowed. If that kind of restriction appears in the middle of a purchased data pool that has already been mixed into training, it can be very expensive to unwind.

Keep the lineage of training runs for that reason. Record which dataset versions, and which parts of them, went into each model, along with the terms in force at the time. If a seller blends third-party or open data with its own, require a per-episode license field. An aggregate statement about the whole dataset can't tell you what you may do with a particular subset.

Sources for this entry

• Open X-Embodiment: Pooled robot learning datasets in standardized formats.
• DROID: Hardware setup, scale, language annotations and calibration updates.
• BridgeData V2: Scale, sensors, control rate and license.
• RH20T: Multi-modal sensors, calibration records and subset licenses.
• Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware: Teleoperation setup, recording rates and demonstration counts.
• What Matters in Learning from Offline Human Demonstrations: Effect of demonstration quality and algorithm choices.
• Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey: The sim-to-real gap and approaches to it.
• Datasheets for Datasets: A template for documenting datasets.
• W3C PROV overview: A definition of provenance.
• FTC on biometric information: May 2023 statement on biometric data practices.

Continue this reading path

• Best Robot Training Data category guide
• Vorrick (featured company)
• How to choose: a robot training data provider
• Resource directory

Cite this entry

Best Robot Training Data. "Provenance and consent in robot data." Content reviewed 8 November 2026. https://bestrobottrainingdata.com/provenance-and-consent/