Egocentric Video Data for VLA & World Models
License egocentric human activity video for VLA pretraining and world models — head-mounted RGB/4K capture with task, action and object annotation. Request a sample from Shaip.
Dataset Specifications
| Target Buyer Teams | Head of Robot Learning; Director of Embodied AI; AI Data Lead; Foundation Model Research |
| Representative Tasks | Cooking, cleaning, organizing, tool handling, packaging, office and warehouse activities |
| Capture / Modalities | Head-mounted RGB/4K video; optional stereo, audio, IMU, gaze and hand tracking |
| Annotations / Metadata | Task and subtask boundaries, action segments, object interactions, narration, hand pose, success and quality flags |
| Delivery Formats | MP4 + JSON/Parquet metadata; optional dataset card and train/validation/test split |
| Scale (Illustrative)* | 2,000–10,000 capture hours |
| Participants / Operators | 500–4,000 participants |
| Tasks / Scenarios | 50–250 tasks |
| Objects / Sites / Views | 10–100 environments |
| Recommended Engagement Scope | 100–300 hours across 10–25 tasks and 3–10 environments |
Shaip's Physical AI program spans 30,000 hours of egocentric video and 4,000+ participants.
View the related case study →Where This Dataset Is Used
This egocentric video dataset is built for teams pretraining vision-language-action (VLA) models, world models and robot learning policies from first-person human experience. Foundation model labs, humanoid robotics companies and embodied AI research teams use this first-person video dataset as human activity data for action recognition, task understanding, hand-object grounding and human-to-robot skill transfer. If you’re sourcing embodied AI training data, head-mounted camera data or a robot learning dataset captured in real homes, kitchens and workplaces, this dataset delivers annotated egocentric video at pretraining scale.
Request a Sample of the Egocentric Human Activity Video Dataset
Tell us about the robot product or Physical AI system you’re building and we’ll respond with sample access, specifications and a proposal matched to your task.
"*" indicates required fields
