Engineering validations
What our pipeline provably does
Every figure on this page is reproducible from an artifact committed in our repository — the annotation JSON, the rendered video, or the service that produces it. These are checks we ran on our own engine and published with the method. We have no customers yet, so there is nothing here to take on trust.
We deliver LeRobot v2.1, HDF5 (EgoDex-style paired files) and the open FH-Ego v1 JSON schema; Zarr v2 and TFRecord (RLDS-style) are produced on request. Every format is verified by an automated export bench before we quote it. We do not offer MCAP or WebDataset — MCAP is a robotics logging container rather than a training format, and labs shard WebDataset internally; we would rather tell you which is which.
You will not find named customers or projected training-speedup numbers here. When we deliver our first production dataset, the results will be published with the Commissioning lab's written consent and the underlying evaluation. Until then, judge us on the engine — inspect the schema and the sample annotation output directly.