Define it. Calibrate it. Collect it.
Clear requirements make quality review practical. Each stage has a decision and a documented output.
- 01
Define the brief.
Set task coverage, action taxonomy, capture modalities and the acceptance rubric around your training objective.
AGREED OUTPUTA capture protocol and data schema: task → subtask → action taxonomy, object vocabulary, episode boundaries, time conventions, modalities, permissions and acceptance thresholds.
- 02
Calibrate together.
Review a small batch for capture quality, temporal labels and loader compatibility. Resolve gaps before scaling.
AGREED OUTPUTA calibration batch reviewed in your loader. Confirm annotation rules, sensor alignment and participant/site split policy; proceed, adjust or stop.
- 03
Collect. Review. Refine.
Check task completion, visibility, synchronization, duplicates and label consistency. Record exclusions and rework.
AGREED OUTPUTAn episode inventory with integrity checks, duplicate review, coverage and annotation QA, plus rejection reasons and rework records.
- 04
Deliver with context.
Receive versioned episodes, annotations, provenance and QA results in the agreed schema. Run your acceptance checks.
AGREED OUTPUTVersioned MP4 + metadata, checksums, provenance and permissions references. Buyer-specific formats such as LeRobot or MCAP require separate schema and import validation.
Quality needs
a shared definition.
Agree what counts as usable activity before capture. Review should distinguish technical quality, task coverage, label quality and rights to use the material.
Make the edge cases explicit.
Define how idle time, repeated takes, camera failures, occlusion and rejected episodes are handled. Rework, review windows and acceptance responsibilities should be written into the pilot scope.
Start with the brief templateDiscuss your
data needs.
Bring a brief. Or start with a task. We’ll define the next step together.