Memories.AI launches Memories Upload, a gig-work platform for robot training data

Memories.ai wants to pay you to chop vegetables. The catch is that you have to film it, so a robot can eventually do it instead.

The San Francisco startup has launched Memories Upload, a global gig-work platform that pays contributors to record first-person video of physical tasks. That footage becomes training data for robots, which still need to be shown how humans handle the most ordinary jobs.

How Memories Upload works

Contributors accept paid recording assignments, clip on a camera, and capture themselves doing the work from their own point of view.

The tasks are refreshingly unglamorous: preparing food, stocking shelves, repairing equipment. Payouts run from $6 to $18 per task, with examples that include vegetable prep and bike repairs.

The camera is Memories.ai’s own hardware, called the LUCI Pin. It weighs under 45 grams and attaches magnetically to mounts worn on the head, chest or wrist.

The pin records 4K video with audio and offers a 109-degree field of view. Two pins can be paired for stereo capture, which produces views that help estimate depth.

Memories Upload launched on October 28, 2026. As of late October 2026, more than 1 million hours of first-person video had been collected using the platform, and the company has deployed more than 5,000 LUCI Pins internationally.

The footage does not just sit in a folder. Memories.ai enriches it with detailed annotations and supplies it to robotics and model-building labs through what it calls a Video Datalake API.

The company behind the camera

Memories.ai was founded in 2024. Its founders, Shawn Shen and Ben Zhou, are former researchers from Meta Reality Labs, the division behind Meta’s work on wearables and headsets.

The startup has raised around $16 million in seed funding. That includes an $8 million round led by Susa Ventures in July 2025.

The company describes itself as the “video data infrastructure layer for physical AI.” Physical AI refers to systems that act in the real world, such as robots, rather than ones that only generate text or images on a screen.

What this means for the robot data race

The market for robot-training data has expanded approximately 10 times year over year, pulling gig platforms and data suppliers into the robotics sector.

In 2026, data-collection competitors reported payouts of nearly $10 million, a sign that paying ordinary people to generate training footage is becoming its own small industry.

What separates Memories.ai, according to the research, is the bundle. It controls the hardware, runs the contributor platform, and handles the annotation, which gives it a say over consistency from the moment a video is recorded to the moment a lab receives it.

There is also an irony baked into the model that the industry will keep bumping into. Workers earning $6 to $18 per task are, in a literal sense, demonstrating their jobs to the machines being trained to perform them.