Company and program review
Graviti review for workplace video and multimodal training data
Graviti advertises catalog data and custom collection for physical AI and language models.
- Commercial basis
- Catalog or custom collection
- Review updated
- Provider website
- graviti.com ↗
The verdict
Consider it when comparing catalog and commissioned multimodal data. Check whether views, labels and identifiers survive the same export.
What to check: Verify the relationship between clips and annotations, and how consent or removal requests apply.
Review basis. Fiund reviewed the cited public materials. Fiund also operates an AI data licensing business. This review includes no paid engagement or product rating.
Graviti advertises catalog data and custom collection for physical AI and language models. One workplace specification describes head-mounted and fixed cameras used together, with time-aligned action labels and estimated pose. The fit is strongest for a buyer testing whether synchronized views reveal an action the first-person view misses. Inspect a representative sample before relying on listed volume. Graviti catalog and specifications
Verify what is delivered together
Ask whether listed hours are unique activity hours or accumulated camera hours. Compare timestamps, calibration notes, scene coverage and missing views in a short pilot. The site labels some products early access. Confirm the availability of the exact data, annotations, and formats you need. Request separate pricing for raw and annotated outputs and any minimum order.
Clarify consent, removal and downstream use
The company says contributors sign a revocable written agreement and that commercial licenses cover model training and deployment. Ask how contributor withdrawal interacts with the license for delivered files and already-trained models. Review role, site permission, access, retention, and redistributability against the signed agreement. A public statement about rights helps identify the right questions for the specific collection. Graviti rights overview
Graviti retail back-of-house capture
Graviti’s public skilled-work specification describes retail back-of-house as a capture environment and lists head-mounted RGB, inertial data and procedural annotations. Ask for a retail-specific sample, how hand–object contact is labeled and whether any accompanying inventory evidence comes from the same take. Request usable unique hours separately from headline collection targets and confirm the rights package for each store and worker. See the retail workflow guide. Graviti skilled work dataset
An original synchronized-view review
Consider a practice task where an operator puts a carton on a shelf. The head view is blocked for two seconds, while a fixed camera still shows the placement. Ask whether the shared task identifier and timestamps connect the two streams and whether the labels cite the view that actually supports the result.
Next, offset the supplied label interval in a copy used only for your own review. An evaluator should detect the mismatch rather than accepting an action label because its text sounds plausible. Do not alter a provider’s source and present the altered copy as its delivery. This is an independent test design, with no claimed test result.
Map availability and permission to each source layer
Keep a specimen manifest for the exact product and release. It should identify original views, processed files, motion signals, estimated poses and reviewed annotations. Ask whether missing views remain explicit and how an early-access feature affects delivery support. Availability of one workplace specification does not establish every possible scenario.
If a contributor agreement is described as revocable, ask what changes for future collection, distributed source files and existing model rights. Request the applicable written terms and evidence for the actual workers and sites. A general rights description cannot settle every item in a mixed workplace archive.
Sources and further reading
Related reviews and guides
Original examples
Industry directory
Find a guide by industry, task or six-digit NAICS code.
Related reading
Dataset documentation tool
Robotics and physical AI
Dexset review: robotics capture, samples and acceptance
Start with one warehouse exception and ask for the relevant views, timestamps and labels. A short trial reveals which parts of the decision the export can establish.
Robotics and physical AI
DXTRAL review for first-person workplace and robotics data
Compare it when you need recordings from a defined workplace and the delivery can preserve the relevant sensor context.
Robotics and physical AI
Hermai review for factory demonstrations and workplace data
A plausible candidate for industrial demonstrations where action timing and motion context matter.
Robotics and physical AI
RoboDATA.AI review for embodied AI data and contributor pilots
Evaluate one wearable-capture and labeling workflow before planning either a dataset order or a contributor pilot.