Synthsis Labs

Skill

Skilled hands, made trainable.

First-person video, hand pose and process structure from people doing physical work on the job, for robots and world models.

See the data
A welder at work in a fabrication shop

One task, segmented.

  1. Load the garment onto the belt.

  2. Align the seam to the guide.

  3. Feed the panel through the stitcher.

  4. Inspect the wash under the lamp.

  5. Log the result and the cycle time.

An illustrative example, simplified.

Every recording holds

  • Egocentric video of hands, tools and workpiece.
  • Wrist and body pose, aligned to every frame.
  • Steps, tool events, faults and recoveries.
  • The outcome of each quality check.

A policy trained only in simulation learns the physics of the simulator.

Recorded at production speed.

Rigs go where the work already happens: production lines and trade crews, under signed agreements with each facility. Everyone recorded opts in and is paid. Faces and identifying details are anonymized before delivery.

Trades

  • Textile
  • Garment finishing
  • Metal fabrication
  • Plumbing
  • Electrical

Rigs

  • Head-mounted
  • Overhead
  • Wrist

What it trains

  • Manipulation policies
  • World models
  • Task segmentation and evaluation

In the marketplace

All of the marketplace

  • Textile and garment production

    First-person video of sewing, finishing and quality checks on working lines, with hand pose and process steps.

    egocentric video, hand pose, process steps

  • Loom setup and fault recovery

    Overhead and wrist cameras on loom threading, setup and recovery from faults.

    multi-camera video, fault recovery, tool events

  • Metal fabrication

    Cutting, welding and fit-up from the worker’s point of view, with protective equipment regions anonymized.

    egocentric video, tool events, safety events

  • Field trades

    Plumbing and electrical work on site, from diagnosis to repair and verification, recorded against your task list.

    egocentric video, diagnosis steps, instrument readings

Need a task we haven’t recorded?

Tell us the task, the environment and the volume.