DATORIC

Quality methodology

Quality.

Each collection specification includes checks for sensors, visibility, action labels, and outcomes.

Browse robotics data

The pipeline

How recordings are checked.

Each program begins with a written specification and modality-specific review criteria. The exact checks vary with the data.

  1. 01

    Specification-driven collection

    Requirements are written down first, so contributors and reviewers work from the same definition.

  2. 02

    Signal and metadata checks

    The specification identifies the technical fields used for the modality, such as video synchronization, depth and pose alignment, hand visibility, or timestamped robot actions.

  3. 03

    Redaction and safety review

    Where personal or sensitive material is in scope, the specification defines the applicable review and redaction controls.

  4. 04

    Labeling and consistency review

    Action, object, outcome, and annotation review rules are set against the fields and acceptance criteria for the program.

  5. 05

    Acceptance

    Accepted material is evaluated against the agreed criteria, with the documented quality fields included in the applicable delivery metadata.

Questions

Quality methodology

What does QA actually check?
Checks depend on the capture program. Robotics dataset specifications define signal quality, sensor synchronization, hand and object visibility, calibration, action labels, task outcomes, and privacy review.
How is data evaluated for a target robot?
Start with the dataset specification and a buyer review package. A pilot tests task coverage, sensor alignment, labels, and compatibility with the policy and action interface before scaling.
Can acceptance criteria be set per program?
For build-to-spec programs, acceptance criteria are agreed before collection begins.