Back

egocentric data collection

Egocentric data collection is the practice of recording data from the first‑person point of view of an acting agent (a human or a robot) while they perform tasks.

In this setup, cameras and other sensors (for example, head‑ or chest‑mounted RGB cameras, eye trackers, IMUs, or robot‑mounted cameras) are placed on the agent so the data reflects exactly what they see and how they move and interact with objects during real activities.

This first‑person stream typically captures fine‑grained details such as hand trajectories, object contact events, gaze shifts, and the temporal order of actions that are often missed by fixed third‑person cameras. In robotics and embodied AI, egocentric datasets aggregate many such recordings across tasks and environments (for example, cooking, assembly, navigation) to learn policies and perception models that better align perception with action from the robot’s own viewpoint.