Curatrix
Blog

What “rights-cleared” training data actually means

Consent, location releases, GDPR and licence scope: the four layers that separate usable robotics data from footage you cannot ship a product on.

2 min readBy Curatrix Research

“Rights-cleared” is easy to print on a slide and hard to prove in diligence. When a robotics team licenses demonstration data, four separate questions need a documented answer. Who performed the work, and did they agree to this use? Whose premises are visible? What personal data was captured incidentally? And what exactly does the licence let you do with the result?

Layer one: contributor consent

Every Curatrix recording is tied to a signed contributor release that names the purpose: training and evaluation of machine-learning models, including commercial ones. The release is stored against the capture session, so a given episode can always be traced back to the agreement that covers it. Contributors are paid for accepted hours, which is also why the consent holds up: it is part of a transparent commercial relationship, not a clause buried in an app’s terms.

Layer two: site authorisation

A workshop, a construction site or a client’s home is somebody’s property. Capture only happens at approved sites with a written location release from whoever controls access. This matters more than it sounds: a surprising share of otherwise excellent footage is unusable because nobody asked the site owner.

Layer three: privacy processing

Head-mounted cameras record whatever is in front of the worker: a colleague walking past, a name on a delivery note, a licence plate through the window. Automated detection flags faces, text and screens; a human reviewer then checks every clip before it enters a release. Data stays in EU-region storage throughout, which keeps the GDPR analysis simple for European customers and predictable for everyone else.

Layer four: licence scope

Finally, the licence itself. Curatrix releases ship under evaluation, research, commercial or enterprise terms, and the scope is recorded in the dataset manifest next to the provenance data. If a team later needs broader rights, the upgrade path is explicit rather than a renegotiation from zero. The FAQ covers the most common licensing questions; for bespoke terms, the enterprise page is the place to start.

Related articles

All articles
1 min read

Five metrics for judging a robotics dataset before you buy it

Hours of video is the least informative number on a datasheet. Contributor diversity, environment spread, episode completeness, annotation density and rights coverage tell you more.

  • data-quality
  • evaluation
  • physical-ai
2 min read

Why Physical AI needs field data, not lab benchmarks

Robots trained on tidy lab demonstrations fail on real job sites. A practical look at what field-captured manipulation data adds and how to judge it.

  • physical-ai
  • data-quality
  • field-capture
1 min read

From raw footage to training episodes: inside the Curatrix pipeline

Segmentation, annotation schema, manifests and formats. What happens between a tradesperson pressing record and a versioned dataset landing in your training run.

  • pipeline
  • formats
  • annotation
Next step

Start with a scoped pilot

Bring a task. On the call we scope the capture protocol, the episode volume, the annotation schema and the delivery format.

Pilots typically start from approximately €25,000. This is an indicative figure only—final scope and pricing are set after a technical qualification call.