Vangrid 让 AI agent 悬赏采集真实世界数据
AI agent 可发布 USDC 悬赏，由附近用户用手机拍摄 3D 场景，链上验收后自动付款。
Gm gVangrid A robot can learn from millions of online images and still have no idea what a loading dock looks like today. That’s the gap @vangrid_io is working on. Here’s the interesting part: an AI agent doesn’t have to wait for someone to build a dataset for it. It can request the data it needs. The agent posts a bounty and locks USDC in escrow. Someone nearby captures the location using a smartphone. The agent reviews the submission and accepts the 3D model. Payment settles onchain. If the job is cancelled or expires, the funds return to the payer. Captures are anchored through EAS on Base, with faces and license plates blurred on the contributor’s device before upload. The flow is live on Arc and Base. What I like about this approach is how it connects three different users: AI builders who need real-world data, contributors who can collect it, and agents that handle requests and payments. That creates a clear path from demand to capture to verified delivery. If physical AI is going to operate in the real world, it needs data from that world. Vangrid is building a way to source it on demand, using smartphones as the capture layer.