SAMPLE Egocentric POV human-motion video in industrial environments

This dataset provides egocentric (first-person POV) video recordings of human motion in real-world and staged industrial environments, designed specifically for training robotics and physical AI systems.

The data captures hands-in-frame interactions, object manipulation, and multi-step workflows across a wide range of industrial and operational scenarios, including:

Picking, placing, sorting, and stacking
Assembly and tool usage
Cleaning and maintenance tasks
Packaging and material handling
Inventory and warehouse operations

All recordings are captured using head-mounted or wearable cameras, ensuring realistic perspective alignment for embodied AI and robot learning use cases.

Key characteristics:

Egocentric POV (human perspective) with hands and objects in view
High-resolution video (Full HD / 4K depending on setup)
Scalable production pipeline (hundreds to thousands of hours)
Metadata included (task type, environment, motion type, object categories, participant info)

The dataset is particularly suitable for:

Robot manipulation learning
Imitation learning (IL) and behavior cloning
Vision-based control systems
Multimodal AI (video + action understanding)
Foundation models for physical intelligence

Custom data collection is available, including task design, environment setup, and domain-specific workflows tailored to client requirements.