Caspius

Caspius is an embodied-AI data collection network on Base that turns first-person footage needed for robot training into assets that can be collected, verified and licensed. Contributors record manipulation video on a phone or wearable and upload motion trajectories, environment interactions and sensor readings alongside it; the network checks quality, discarding shaky, occluded or duplicated segments, and accepted data enters datasets used for model training, with contributors rewarded according to how much of their data is used.

It addresses the data bottleneck in embodied AI, where real-world manipulation samples are far scarcer than text or images, costly to gather and hard to authenticate. Separating collection, quality control and licensing lets one recording be licensed repeatedly without losing its provenance. The platform launched in April 2026, tasks are grouped by scenario across household manipulation, warehouse handling and tool use, and data types, annotation rules and acceptance criteria are fixed per task, with acceptance and licence records written on-chain.