How we work
Define the work, agree the capture and review a concrete sample. A shared specification connects the first conversation to the dataset handoff.
COLLECTION / FROM BRIEF TO HANDOFF
For robotics data projects, the research objective should guide the field setup and the review process. The pilot is where those decisions become concrete.
01 / COLLECTION
Start with the behavior you need to study. We discuss target tasks, useful object interactions, environment coverage and the role the data will play in your workflow. Existing examples and failure cases help make the brief precise.
Bring to this step: Research objective, task list, reference clips and any existing data specification.
02 / COLLECTION
Match the brief to demonstrator skills, workplace access and a practical camera setup. Define what should appear in frame, what should remain outside it and which task variations need coverage. Agree participation, recording and usage arrangements before collection.
Bring to this step: Required settings, restricted content, contributor criteria and intended data uses.
03 / COLLECTION
Capture a focused sample rather than expanding immediately. Review whether the hands, tools, objects and task outcome are visible. Test the proposed file structure, metadata fields and any labels against your team's expectations.
Bring to this step: Feedback from the people who will inspect and use the data.
04 / COLLECTION
Run the collection against the approved brief. Track completed tasks and the requested variations so a growing folder of recordings also represents the coverage you asked for. Record exceptions while they can still be resolved in the field.
Bring to this step: An agreed task matrix and a process for resolving scope questions.
05 / COLLECTION
Check recordings against the pilot's acceptance criteria. Apply the agreed task descriptions and time-based labels where included. Separate accepted, revision-needed and excluded items, with reasons that make the review traceable.
Bring to this step: The approved label vocabulary and any project-specific review rules.
06 / COLLECTION
Package the data in the agreed structure. Walk through the manifest, naming conventions, annotation fields and any collection limitations. A defined review window gives both teams a clear route to resolve missing files or agreed revisions.
Bring to this step: Delivery destination, preferred structure and named reviewers.
REVIEW / BEFORE SCALE
Collection quality depends on a shared definition of a usable recording. We agree the checks and label rules during the pilot, then apply them to the larger collection. Annotation depth is scoped to the project.
| Review area | What to agree and inspect |
|---|---|
| Visibility | Can a reviewer see the relevant hands, objects and tool interactions? Check sustained occlusion, framing and actions outside the agreed view. |
| Continuity | Can the task sequence be followed from the agreed start to its outcome? Flag missing stages, interrupted recordings and unclear completion. |
| Technical integrity | Does the file meet the agreed recording specification? Check playback, file completeness and project-specific video settings. |
| Task fit | Does the clip show the requested task and variation? Compare the recording with the approved brief and coverage matrix. |
| Label consistency | Do the labels follow the agreed definitions and timing rules? Use pilot examples to resolve ambiguous boundaries and terminology. |
| Package completeness | Do the video, manifest and labels refer to the same recording? Check stable IDs, missing files and schema consistency before handover. |
PARTICIPATION & INTENDED USE
The collection brief should identify the people and workplaces involved, permitted recording conditions and the intended use of the resulting data.
Tell us what you want recorded. We can discuss the skill, environment, camera setup, annotation and budget in a scoping meeting.
