vs. open data
Open data is your base. We sell the apex it lacks.
We don't compete on commodity volume — the free corpora already win there. We fill the gaps they structurally can't: your environments, rights-clean human subjects, force-labeled contact tasks, and the long-tail failure cases open sets don't cover.
| Dataset | Scale | Modalities | License | What it lacks vs. bespoke |
|---|---|---|---|---|
| Open X-Embodiment | 1M+ trajectories · 60 datasets | RGB + action (mixed) | Mixed / open | Your environments; consistent modalities; rights guarantees |
| DROID | 76k trajectories · ~350 h | 3× stereo RGB + action | Open (research) | Force/tactile; your embodiment; bespoke diversity |
| EgoDex (Apple) | 829 h · 194 tasks | Egocentric video + 3D hand pose | Open (research) | Robot action space; force; commercial rights clarity |
| BridgeData V2 | 60k+ trajectories | RGB-D + action | Open | Environmental diversity beyond its labs; contact data |
| Kinema (bespoke) | Made to order · pilot-scale | RGB-D + proprioception + 7-DoF action + hand pose + force/torque | Single-buyer commercial · consented | — by design, it's the apex |
Past a per-scene threshold, a policy's generalization scales with the diversity of environments and objects as a power law — far more than with raw demonstration count (ICLR 2025, arXiv:2410.18647). That's the whole argument for paying us over more of the same free data: diversity you can't self-produce moves your eval metric.
All figures public and verifiable · OXE = 1M+ (not 2M+)