@momehx X analytics

Momeh@momehx·13.4K followers·18d ago

PHYSICAL AI NEEDS MORE DATA AND SPATIAL CONTEXT FROM THE REAL WORLD. @vangrid_io is tackling a problem that often gets overlooked: robots and world models need fresh, ground-level information about the places they are expected to navigate and understand. Satellites and traditional mapping fleets help, but keeping physical environments updated at scale remains a challenge. VanGrid’s approach is to turn everyday smartphones into spatial data capture tools. HERE’S HOW IT WORKS: → Accept a funded bounty and capture a real-world location using your phone. → Faces and license plates are blurred on-device before video encoding. → The footage is reconstructed into a 3D spatial model. → Each capture receives a content fingerprint and is included in an EAS attestation on Base, making its provenance independently auditable. THE IMPORTANT PART IS THAT THIS IS MORE THAN A CONCEPT. VanGrid’s public explorer has reported more than 1.25M captures, 4,700 attestations, and $600K+ in USDC sett

Momeh@momehx·13.4K followers·21d ago

How Vangrid is solving the text data wall for Physical AI LLMs ran out of fresh human text to train on. Physical AI, which covers humanoid robots, self-driving fleets and spatial world models, has a tougher version of that problem, because there is no public web to scrape for dense, current 3D data of real streets, warehouses and sidewalks. @vangrid_io is building that missing layer on Base. I went through the flow end to end [add your real detail: what you captured, how long it took, what the app showed afterward]. 1Ground-level multi-view capture
→ Coverage gaps: satellites and car-mounted cameras miss doorway steps, loading docks and narrow pedestrian walkways
→ 3D reconstruction: crowdsourced multi-view phone video becomes dense spatial models that robotics simulators can use
→ Privacy at the edge: faces and license plates are blurred on the phone before a capture leaves the device
→ Verifiable provenance: each capture carries a content hash, coarse location and timestamp, anchor

Momeh@momehx·13.4K followers·25d ago

HOW VANGRID BUILDS A VERIFIABLE TRAIL FROM REAL WORLD CAPTURE TO ONCHAIN DATA Spatial data becomes far more useful when buyers can evaluate not only what was captured, but also where it came from and whether its provenance can be checked. For robotics, autonomous navigation, and physical AI, unreliable inputs can undermine the value of an otherwise sophisticated model. @vangrid_io is building a pipeline that connects physical world capture with cryptographic records and onchain verification. 1. Establishing Capture Provenance → Multi-Attribute Records: Capture metadata can combine content hashes, approximate location, timestamps, and device-related signals to support authenticity checks. → Onchain Attestations: Ethereum Attestation Service (EAS) schemas on Base can provide a public verification layer for capture records. → Motion Validation: Comparing camera perspective changes with device movement data can help flag suspicious recordings and replay attempts. → Auditable Histor

Momeh@momehx·13.4K followers·22d ago

YOUR NEXT WAREHOUSE ROBOT NEEDS A DIGITAL WORLD IT CAN NAVIGATE, NOT ANOTHER VIDEO. Here are four practical questions about how @vangrid_io turns real-world spaces into usable spatial data. 1. Why isn’t ordinary video enough? Video records a scene from a particular viewpoint. Robots need spatial information they can act on, including aisle widths, doorway clearance, ramp gradients, turning areas, and obstacles around corners. VanGrid transforms multiple views into 3D representations such as Gaussian splats, point clouds, and textured GLB meshes. 2. How is capture quality checked? Before reconstruction is paid for, the requester gets a preview to assess coverage. They fund a location in USDC, a nearby contributor records the site using a phone, and the preview helps identify missing areas or unstable footage before the process moves forward. 3. How is privacy protected? Faces and license plates are redacted locally on the contributor’s device before encoding. Raw video remains of