Perceptron Egocentric 1.0 API labels first-person human and robot video for robot training at $29 per video-hour
Perceptron has released Egocentric 1.0, a self-service API that labels first-person video of people or robots so it can be used as robot training data. Customers submit up to 1,000 video links per batch. They get back timestamped subtask timelines plus optional per-hand activity labels and 2D or 3D hand tracking. Timeline, 3D hands and activity labels together cost $29 per video-hour; the timeline alone is $9. Perceptron says its timelines score within 0.01 of OpenAI's GPT-6 Astra on EgoHome, a new 137-clip household benchmark it built and plans to release, at about one-eighth the cost. It also claims the best segment-timing score on the public WGO-Bench, and says its 3D hand model needs no camera calibration. These are the company's own benchmark results. Perceptron points to research such as EgoMimic showing that first-person human video with 3D hand tracking can improve robot manipulation training.
More on OpenAI, GPT-6 Astra and robot ai.
About the footage: Not robot footage: model output (hand tracking and labels) overlaid on human first-person video from Perceptron's launch video. The robot clips on Perceptron's page come from the third-party Galaxea dataset and aren't used here.
The source post
Post by Humanoids Daily on X.
View the post on X
Sources
New to robots? Read our guide: How to tell if a robot video is real: CGI, sped up, or teleoperated?.
Summarized by Servospace from the sources above, drafted with AI help and checked against those sources before publishing. Claims are the companies' or outlets' unless we say otherwise. How we work.