Egocentric POV — everyday tasks, hands in frame
The brief
"Egocentric video" just means first-person footage with your hands doing real work — and it's one of the most in-demand data types in AI right now, because it's how robots learn to manipulate the world. The generic ask is vague; the specific tasks below are what actually gets licensed.
The recipe: mount a camera on your head or chest (GoPro, Insta360, or a phone in a strap), frame so your hands and the workspace fill the shot, then do the task for real, start to finish, in one continuous take. Don't perform for the camera — natural pace and natural mistakes are part of the value.
What gets licensed first:
- Complete tasks (start state → done), 5–30 minutes, no cuts or speed-ups
- Hands clearly visible and in focus for essentially the whole take
- Bright, even lighting and a reasonably steady head — check a 30-second test clip first
Keep it clean: only you in frame (no family members or bystanders), no TV or music playing in the background, and nothing personal readable on mail, screens or documents.
Requirements
- Head- or chest-mounted camera (GoPro / Insta360 / phone in a mount) — not handheld
- 1080p or higher at 30fps+, landscape
- One continuous, unedited take per task — 5 to 30 minutes
- Hands and workspace in frame and in focus throughout
- Only you on camera; no bystanders, pets are fine
- No TV, music or podcasts audible in the background (copyright)
What your footage trains
First-person manipulation footage is the raw material for robot learning — models map what your hands do to what a robot gripper should do. Complete, uncut everyday tasks are exactly what labs are buying.