Collection brief
Signed off before capture
- Data
- Egocentric task video
- Tasks
- 40 household tasks
- Target
- 2,000 episodes
- Format
- 1080p 30 fps, IMU
Collect
Data your model has never seen, collected to a brief.
Robot and egocentric episodes, speech, images, video and documents captured to your specification, with consent on file and quotas tracked by group.
Follow one collection, from brief to delivery
A sample egocentric collection for robot learning: operators wearing head cameras record household tasks, followed through our workspace.
- 1Brief
Agree what to record, before anything is captured
Which tasks, how many episodes of each, in what kind of home, with which camera and sensors. Quotas are set per task group so the data covers what your robot will be asked to do.
- 2Recruit
Find operators, screen them and train them on every task
Operators come from our own team and from specialists recruited through Rubricate, our expert network. Each is screened by home layout and handedness, trained on the task list, and signs consent before recording.
- 3Capture
Record each task from the operator’s point of view
Operators wear a head mounted camera and work through the task list one episode at a time, at home or on a set. A checklist confirms consent, camera position, sensor calibration and that both hands are in frame.
- 4Check
Every episode checked on intake
Each episode is checked for a completed task, hands in frame, dropped frames, camera shake and anything private in view before it counts toward a quota. Rejected episodes are recorded again, never counted.
- 5Deliver
Delivered with consent and provenance for every episode
Episodes arrive in your storage with synced sensor data, a manifest, the task for each episode and a consent index, ready for action episodes and task descriptions.
What lands on your desk
Four documents run through every collection, so you always know what was agreed, who took part and what you received.
Consent record
One per operator
- Operator
- O-118
- Use
- Robot learning
- Signed
- In app, form v2.1
- Withdrawal
- Any time
Weekly quota report
Every week of the project
- Accepted
- 1,240 of 2,000
- Behind
- Tools and appliances
- Rejected
- 7.2%, recorded again
- Next week
- 6 operators added
Delivery manifest
With every delivery
- Episodes
- 640
- Hours
- 7.8
- Includes
- IMU, tasks, consent index
- To
- Your S3 bucket
What we collect
Five kinds of data, each recorded by people trained for it and delivered in the format your pipeline expects.
Robot and egocentric episodes
Head camera and multi camera recordings of people doing tasks, and teleoperated robot episodes
- Recorded by
- Trained operators following a task list
- Delivered as
- Episodes with synced camera and sensor files
Speech and conversations
Read scripts, spontaneous speech, call style dialogues and wake words
- Recorded by
- Speakers by language, accent, age and region, including Indian languages
- Delivered as
- WAV or FLAC, verbatim transcripts
Images and photos
Products, scenes, objects in context and people with consent
- Recorded by
- Contributors on phones, in the settings and lighting the brief sets
- Delivered as
- JPEG or PNG with capture metadata
Video
Actions, gestures, driving, warehouse and site footage
- Recorded by
- Contributors and field staff following a shot list
- Delivered as
- MP4 with timestamps and scene tags
Documents and handwriting
Forms, receipts, handwritten notes and signatures
- Recorded by
- Contributors writing or photographing to a template
- Delivered as
- Scans or photos with a field map
Contributors come from the Dataclap team and from specialists recruited through Rubricate, our expert network.
Consent, privacy and ownership
The question every collection raises first. Here is how we answer it.
- Consent on file
- Every contributor signs a consent form for the stated use before recording. You receive the consent index with the data.
- Right to withdraw
- Contributors can withdraw at any time. Their files are deleted and you are told which files to remove.
- Personal details kept out
- Names, faces and numbers that are not part of the brief are redacted or never captured.
- You own the data
- Collected data and its rights pass to you on delivery. We keep nothing after the project closes.
Delivered to
S3GCSAzure BlobSFTP
Priced
Per itemper hour of audioor per programme
Starts with
A free pilot batchbrief agreed on a call
Questions about collection
Can you reach a specific language, region or age group?
Yes. We recruit by language, accent, region, age and device through our own team and Rubricate, and track every group against its quota each week.
Do you collect egocentric data for robot learning?
Yes. Operators wear head mounted cameras and record household and workplace tasks one episode at a time, with sensor data synced and every episode checked before delivery.
Who owns the collected data?
You do. Rights pass to you on delivery, and each file comes with the consent record that covers its use.
What happens if a contributor withdraws?
Their files are deleted from our side and we tell you exactly which files to remove from yours.
Do you collect in Indian languages?
Yes. Tamil, Hindi and other Indian languages, as well as Indian English and code switched speech.
How fast can a collection start?
Once the brief is signed off, recruiting starts straight away. A small pilot batch comes back early, so you can check it before the full run.
Tell us the data your model has never seen
Send a small batch and get it back done, free, usually within 24 to 48 hours. You judge the work before you pay for anything.