Project Directory
Decentralized data pipeline for physical AI: contributors film real-world hand tasks on their phones for robotics training datasets.
CapturGO is a decentralized data pipeline for physical AI. Robots and other embodied AI systems learn from first-person video of people doing real tasks with their hands, and that data is scarce. CapturGO pays a global network of contributors to capture it on the phones they already own, then annotates and cross-verifies it with standardized capture protocols and human-in-the-loop quality checks.
The result is consented, first-person manipulation datasets delivered in formats robotics teams already use, such as MCAP and LeRobot, for training and evaluating models. CapturGO says more than 16,000 contributors have already registered on its network, and its iOS and Android apps are about to go live. CapturGO is part of Superteam Japan's Colosseum hackathon cohort.