Ifirsthand

Firsthand (embodied-AI division of CloudPano) — A ready-made, commercially licensed egocentric multimodal dataset of live retail shelf operations - facing and front-stocking to a planogram, case-cut and shelf-load, price and label swaps, chilled-case rotation by date code, and hanger and fixture resets - captured on operating sales floors with customers present across ten hardware-synchronized streams: 4K60 global-shutter head camera, two 1080p60 wrist cameras, active-stereo depth in metric millimetres, 200 Hz 6-DoF IMU, 3D 21-keypoint hand pose with grasp and release contact events, body pos…

Ifirsthand offers (Alternative, Geospatial) — 890 validated hours across 6,210 episodes from 96 participants at 31 capture sites, with ten synchronized streams per episode - head video at 3840x2160 at 60 frames per second, two wrist streams at 1920x1080 at 60 frames per second, depth at 848x480 at 30 frames per second and 200 Hz inertial measurement, so on the order of 190 million head-camera frames plus comparable multi-stream volume in the terabyte range; median episode 78 seconds, or 41 seconds for chilled-case work.

Training and benchmarking retail shelf inspection and planogram-compliance vision models; auditing robot-manipulation policies for restocking and facing tasks; humanoid and manipulator imitation-learning datasets on real store tasks rather than teleoperated lab demos; price-change and shelf-label digitization models; date-code rotation and freshness-handling behavior models; sim-to-real gap closing for retail robotics; and as rare in-store failure-case data for robustness and safety evaluation

Coverage spans US; Application Software, Food Retail; alternative, geospatial; equities.