Skip to content

Garden

Language models learned from the internetRobots need to learn from the physical worldEvery day, people build, repair, grow and make things. Most of that work is never recorded.

How Garden collects data

  1. Capture

    Data gardeners record their work as they do it, using a lightweight wearable device.

  2. Reconstruct

    Garden reconstructs the scene in 3D, including depth and the movement of both hands.

  3. Enrich

    Each recording is paired with descriptions of the actions and context behind the work.

  4. Grow

    From workshops to farms, every new gardener brings another part of human skills into the library.

Recorded on our deviceHands in 3DPick up the red cube · place it in the box

It becomes a leaf on the branch of its skill

Next to every other recording of the same work.

Garden creates the engine that turns real-world work into enriched data for physical AI

Side camera housingMain cameraIMUForehead mountSide cameraStereo boardSecond camera
  • Side camera housing
  • Main camera
  • IMU
  • Forehead mount
  • Side camera
  • Stereo board
  • Second camera

Open-source device

A 3D-printable enclosure, an open parts list and open-source firmware. The device can be assembled anywhere or ordered ready-made.

Marketplace

From wiring to harvesting, find data by task, trade and environment. License recordings from the library or commission the work you need captured.

Data gardeners earn

People and companies share in the revenue when their data is licensed. Garden handles the processing and annotation.

Garden today

Built from everyday parts

Off-the-shelf cameras, a sensor and a Raspberry Pi in a 3D-printed frame.

≈ $375*for the whole device* Estimated calculation based on retail prices, shipping not included

The device’s parts and what they cost
Stereo cameraGXIVISION, 2 × 1080p, synced$76
Two side camerasOV2735, 1080p USB$71
IMUWitMotion WT901C, 9-axis$53
Raspberry Pi 58 GB$175
  • Gardeners signed

    Our first participants are ready to record their work. Among them is a Spanish renovation company, with an agreement to capture work at one of its properties.

  • Open dataset

    We're preparing a test dataset for Hugging Face, including recordings from the renovation project.

  • Technical whitepaper

    We're preparing a whitepaper on Garden's approach to capturing and reconstructing physical work.

Invest in Garden

We’re raising to bring Garden to the next stage.

Get in touch

About

Opens your mail app, ready to send. Or write to savva.riisman@gmail.com