Building the data supply chain for physical AI

Real people, real environments, real work. We capture egocentric video of domestic tasks, warehouses and factory floors, starting with pilots in the Philippines. We turn that footage into training-ready data: consented at source, quality-checked and annotated.

Book a call

Samples

Explore our data

TaskBag folded pillowcasesOpen the bag
00:00
Open bag00:04.5
Insert bundle00:12.5
Fold wrapping00:17
Tape closed00:28.5
Turn bundle
01 Egocentric video
Real people, real work, real environments.
02 Task & action labels
Timed annotations at both levels.
03 Quality control
Human review of the final output.
04 Pose & depth
Hand keypoints and depth estimates.
05 Your delivery format
Packaged to the agreed specification.
Request a sample

How we deliver

From brief to dataset

  1. Define the capture

    We agree on tasks, settings and hardware: from mono cameras to stereo and sensor streams.

    A shared specification, including consent and quality criteria.

  2. Review a pilot

    We collect a small batch. You review the footage, poses, depth and task- and action-level labels.

    Refine the setup before scaling.

  3. Validate, then deliver

    We collect in batches, manually review the final output and export in your agreed format.

    Synchronized data, with metadata and documentation.

Environments

Where real work happens

Developing collection partnerships across Southeast Asia and West Africa.

01

Factories

Assembly, textiles & production

02

Warehouses

Picking, packing & handling

03

Industrial plants

Specialist processes & operations

04

Homes

Cooking, cleaning & everyday tasks

05

Restaurants

Preparation, kitchens & service

06

Airlines & freight

Cargo & ground handling

Let’s talk

What are you
teaching your model?

If you see a way we could work together, from shaping what we collect to partnering on data collection, we’d like to talk.