Egocentric Data

First-person video data for embodied AI and robotics

Overview

Egocentric data is captured from the first-person perspective of a human performing real-world tasks.

This viewpoint preserves:

• Task intent
• Hand–object interactions
• Temporal structure of actions

For robotic systems, egocentric data more closely matches onboard sensor perspectives than third-person datasets.

---

Why it matters for robotics

Imitation learning

Learn manipulation policies directly from human demonstrations.

Perception alignment

Train vision models using viewpoints similar to robot-mounted cameras.

Long-horizon tasks

Capture full task sequences instead of short action clips.

Multimodal grounding

Align vision, language, and action in a shared context.