Vangrid crowdsources 3D spatial data for robots via phone clips
Vangrid has users record short phone clips that it reconstructs into point clouds and Gaussian splats, attests onchain, and delivers to robot companies through bounties.
a robot can see a photo. that doesn’t mean it understands the room. that was the idea behind @vangrid_io early thesis in April. Vangrid argues that Physical AI needs more than 2D images and text. robots need continuously updated 3D ground truth to understand things like distances, obstacles, floor layouts and what has changed since the last observation. that’s what they mean by “the spatial cortex.” the phone is the capture layer. people use normal phones to record short clips. Vangrid reconstructs those clips into point clouds and Gaussian splats, while faces and license plates are blurred on-device. each capture also gets provenance data, including a content hash, timestamp and location. batch roots are then attested on Base so the resulting data can be verified. then bounties connect demand with supply. a company or AI agent can request a specific location, contributors capture it, and the structured 3D data is delivered through the network. capture → reconstruct → verify → update → deliver the April post was the thesis, not a product launch. since then, Vangrid has moved toward the actual infrastructure with its capture app, bounty marketplace and $9M seed. now the important test is simple: can crowdsourced phone captures become fresh and reliable enough for robots to actually depend on?