Embodied Intelligence,
Building an Open Multimodal Data Ecosystem

First-person capture from one industrial site: manual painting. Each episode ships two 1920×1080@30 streams, IMU, raw audio, frame index and on-device checker logs.

First-person capture from a restaurant kitchen: beef cutting. Each episode ships two 1920×1080@30 streams, IMU, raw audio, frame index and on-device checker logs.

First-person capture at a logistics site: parcel packing. Each episode ships two 1920×1080@30 streams, IMU, raw audio, frame index and on-device checker logs.

Stereo first-person video of five sewing tasks (buttoning, picking, sewing a button, sewing a sewing board, buttonholing), with task/left-right-hand semantic annotations and partial hand-pose annotations.

First-person capture from one industrial site: manual painting. Each episode ships two 1920×1080@30 streams, IMU, raw audio, frame index and on-device checker logs.

First-person capture from a restaurant kitchen: beef cutting. Each episode ships two 1920×1080@30 streams, IMU, raw audio, frame index and on-device checker logs.