What you get
Each dataset is continuous, unscripted footage of a professional doing their actual job — a carpenter, welder, chef, tailor, mechanic or builder — seen from their own point of view. Hands, tools, materials and the sequence of steps are all in frame.
Every session is delivered with a JSON metadata record: task, environment, duration and recording device. Every participant has signed a release, so the data can be used for model training and commercial products.
Who it is for
Robotics companies training manipulation policies and humanoid robots, research labs working on imitation learning and vision-language-action (VLA) models, and AI teams that need human demonstration data from real workplaces rather than staged lab tasks.
Buying by the hour
Egocentric data is priced and delivered per recorded hour of a single task. Start with a small pack to evaluate, then scale to tens or hundreds of hours of the same trade. If we don't have the volume yet, we record it for you.