GM gVAN
@vangrid_io framed the problem well: when the signal exists but the context does not, embodied systems stay limited.
Robots and world models already process plenty of data. What they lack is fresh, ground-level spatial context from the places they operate in, and satellites or mapping fleets only reach so far. I ran through the capture flow myself. You accept a bounty, walk your phone around the location, faces and license plates are blurred on device, and the video is reconstructed into a 3D spatial model. Each capture gets a content fingerprint and is batched into an EAS attestation on Base, so buyers can audit provenance themselves.
What convinced me is that it is live. Buyers fund bounties in USDC, funds stay in escrow until a capture is accepted, and the public explorer already shows 1.25M+ captures, 4,700+ attestations and $600K+ USDC settled.
The bottleneck for physical AI is less about another model and more about verified ground truth from the places agents actually need to operate.
A distributed grid of phones is a practical way to keep that data current.