Exclusivity and Field-of-Use Terms for Robotics Data

Two different levers

Buyers often ask for "exclusive rights" when they mean one of three things. Separating them can shorten the negotiation.

A license can combine them: non-exclusive in general, exclusive within one field for a fixed term, with a defined list of permitted uses. The licensing terms overview for robotics training data covers the full clause set; this page covers the exclusivity and field-of-use pieces only.

Why robotics data raises specific questions

Robotics data has properties that make "exclusive" less obvious than it sounds for text or tabular data.

• One recording, many uses. A single workplace video can feed visual pretraining, step recognition, reward estimation and evaluation. A field-of-use clause that says "robot training" does not say which of those is covered.
• Embodiment is a field. A buyer building a humanoid may not care if the same footage is licensed to a company building fixed-arm cells. Another buyer may. Define it by embodiment class, not by company name.
• The supplier's own use. An operating business may want to keep using its recordings for internal training or safety review. An exclusive that bars this may be refused, or priced accordingly.
• Consent limits the supplier's grant. If workers or customers appear in the footage, what they agreed to caps what any exclusive can promise. See worker consent and identifiable people.
• Collection can be exclusive more easily than archives. For newly commissioned recordings, exclusivity can be written in before capture. For archives, other parties may already hold copies.

What exclusivity can and cannot do

Exclusivity can stop the supplier from granting the same recordings to others for a stated field and term. It can require the supplier to delete or segregate its copies for that field, and it can set remedies if the supplier breaches.

It cannot make the data unique in substance. Another supplier's recordings of the same task at a different site are not covered. It cannot bind parties who obtained the material before the license, and it cannot bind third-party content owners who appear in the footage. It does not stop a competitor from recording the same task themselves, and it does not control public datasets that contain similar material.

It also reduces your options. Suppliers who would otherwise license to several buyers may decline, or ask for a higher fee, because exclusivity removes their other revenue. Question whether you need exclusivity on the raw data at all, or only on the labeled and curated package you pay to create.

Field-of-use terms: where definitions fail

Field-of-use wording often fails when it is either too broad ("any AI purpose") or too vague ("robotics research"). Both leave disputes about edge cases: Is a vision-language model that later controls a robot in scope? Is internal evaluation a "use"? Is fine-tuning for a customer's deployment covered?

A common reference point is how open licenses draw lines. Creative Commons describes NonCommercial as use not primarily intended for or directed towards commercial advantage or monetary compensation, judged by the use rather than the user, and its NoDerivatives term bars sharing adaptations publicly (Creative Commons). The same page treats licenses carrying NC or ND terms as non-free. Even this widely used vocabulary needs interpretation, so a bespoke commercial license should spell out purposes instead of borrowing labels. For public datasets, run the checks in public dataset commercial license checks.

The U.S. Copyright Office's report on generative AI training examines fair use and licensing options for AI training (U.S. Copyright Office, 2025). Treat it as background for your counsel, not as a substitute for them.

Illustrative field-of-use and exclusivity schedule

The values below are synthetic. A buyer's counsel would adapt every line.

# Illustrative - not a template to sign
license_id: LIC-EXAMPLE-001
exclusivity:
  scope: "named recordings listed in Schedule A (batch_017 only)"
  field: "training and evaluation of mobile-manipulation policies for bipedal humanoid embodiments"
  term_months: 24
  territory: "worldwide"
  supplier_retained_uses: ["internal safety review", "internal staff training"]
  after_term: "converts to non-exclusive; supplier may license outside field"
field_of_use:
  permitted: ["policy training", "pretraining", "evaluation", "internal research"]
  excluded: ["resale or redistribution of recordings", "use to identify individuals", "training of general-purpose face or voice recognition"]
derived_outputs:
  labels_created_by_buyer: "buyer-owned"
  model_weights: "buyer-owned; no obligation to delete on expiry unless Schedule C says otherwise"
breach:
  remedy: "notice and cure period; deletion certification; fee adjustment"

Pitfalls to check

• Term and conversion. What happens when exclusivity ends? One option is reversion to non-exclusive; reversion of all rights is usually not what a buyer wants.
• Sublicensing and affiliates. Can your contract manufacturers, evaluators or cloud processors touch the data? An exclusive without a sublicense route can stall your own pipeline.
• Definition of "same data". A re-recording of the same task at the same site by the same supplier may or may not be "the same data".
• Retention after term. If you hold derived artifacts, decide what must be deleted. See retention, deletion and audit rights.
• Model derivatives. Check that field-of-use limits on data do not accidentally restrict deployment of weights. See model weights, derivatives and redistribution.

Limitations and what exclusivity cannot repair

Exclusivity does not fix poor data. Footage with the wrong viewpoint, no actions and weak labels stays weak when only you hold it. Human video does not contain robot joint states or forces, so exclusivity over it does not substitute for robot data; see what human demonstration video can and cannot do. It also does not cure a permission defect: if the supplier cannot grant a right, an exclusive on that right has little value. Redaction or blurring does not resolve consent questions on its own.

Questions to ask suppliers

1. Which recordings are covered, and can you list them by batch or ID in a schedule?
2. Has any of this material been licensed, shared or published elsewhere already?
3. Will you accept exclusivity limited to a named field and term, and what do you want in return?
4. Which uses do you need to keep for yourself?
5. Who appears in the recordings, and what did they agree to about licensing and exclusive use?
6. Does third-party content appear (product packaging, screens, vendor materials) that you cannot license exclusively?
7. If you hold other recordings of the same tasks, are they excluded from the exclusive?
8. What are your deletion and segregation obligations for your own copies?
9. Will labels and metadata we create be ours alone?
10. Can exclusivity convert to non-exclusive after the term without a new negotiation?

Discuss your requirements

If exclusivity matters to your program, say so when you describe the data you need to SourceX. SourceX looks for US businesses that hold suitable material, checks the data and the supplier's licensing permissions, and coordinates the license and delivery. Exclusive terms are agreed with the supplier case by case, and a request does not guarantee a match. For wider context, see the humanoid training data guide, the procurement hub, and SourceX's licensing terms page.

Discuss your humanoid training-data requirements with SourceX

Sources

• About CC Licenses — NonCommercial and NoDerivatives terms in Creative Commons licenses restrict use by purpose and adaptation, an example of field-of-use style limits.
• Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version) — Licensing options for AI training are discussed in the U.S. Copyright Office report.