Brief from the failure
Bring the failing eval episodes. We turn them into a task and object spec and collect demonstrations that target exactly that behavior.
Robotics, on demand
On-demand robotics training data, delivered fast.
When a policy fails on a grasp, a tool, or a new environment, brief the missing demonstrations and get them back fast, labelled for imitation learning and in the loader format your stack already uses.
- Use
- Imitation · VLA · world models
- Labels
- Grasps · contacts · actions
- First batch
- Median 48 h
- Formats
- RLDS · LeRobot
Definition
What is on-demand robotics training data?
On-demand robotics training data is human demonstration data collected when a robot policy needs it, scoped to the specific tasks, objects, or environments where it fails, and delivered labelled for imitation learning rather than bought as a fixed off-the-shelf dataset.
How it works
Built for the policy loop.
Labels your loss can use
Action segments, grasp types, contact onsets, and 3D hand trajectories, not just a caption on a clip.
Drops into your loader
RLDS and LeRobot out of the box, with readable converter source so you can retarget the schema.
Iterate every training run
Re-brief after each eval. Top-ups run against the same accept criteria, so new demonstrations stay consistent with the old ones.
At a glance
What you get.
- Demonstrations
- Egocentric + wrist, synchronized
- Labels
- Actions · grasps · contacts · 3D hands
- Spec to first batch
- a median of 48 hours (deep coverage)
- Formats
- RLDS · LeRobot · HDF5 · zarr
- License
- Perpetual, buyer-owned, exclusivity available
Every collection runs through the same seven-stage end-to-end custom collection pipeline. End-to-end custom data collection is a managed service that takes an AI data need from problem to owned dataset in a single accountable pipeline.
FAQ
On-demand robotics training data, answered.
- Do you collect robot teleoperation data?
- Our specialism is human egocentric demonstration. It transfers to robot policies and scales across homes and workplaces far faster than teleoperation fleets.
- Can data be exclusive to my team?
- Yes. Custom collections can be licensed exclusively so the demonstrations are never resold.
- How do I know the data helps my policy?
- Start with a pilot, train on it, and measure on your held-out suite. Our benchmarks page shows how we evaluate egocentric pretraining data ourselves.
Brief the gap. Get data back fast.
Tell us what your model is missing. We will quote reach, timeline, and price against a written spec, with a first validated batch in a median of 48 hours where coverage is deep.