RoboX Technologies

Crowdsourced egocentric video for robotics imitation learning.

RoboX is building a diverse first-person video dataset for training the next generation of robots. Contributors around the world record short clips through the RoboX mobile app, capturing how humans grasp, move, navigate, and interact with everyday environments. Each clip flows through an automated annotation pipeline that produces hand keypoints, object tracks, action segments, sensor data, and spatial context, ready for imitation learning research.

Datasets

Dataset	Description	Clips
RoboX-EgoGrasp-v0.1	Single grasp actions on everyday objects	10 sample / 1,800+ full
RoboX-Egocentric-Collection-v0.2	Combined collection across four campaigns: EgoGrasp, EgoDaily, EgoScene, EgoNav	7,342 episodes

Campaigns

EgoGrasp: Grasping interactions with 1036+ object categories
EgoDaily: Routine household and workplace activities
EgoScene: Scene-level environmental captures
EgoNav: First-person navigation through indoor and outdoor spaces

How It Works

Contributors record first-person video through the RoboX mobile app following structured task prompts
Clips are quality-filtered and face-blurred on-device before upload
An automated pipeline produces hand keypoints, object detection and tracking, action segmentation, and spatial context labels
Verified clips are published as research datasets and made available for full access on request

Full Dataset Access

Sample clips are published openly on Hugging Face. Full dataset access is available on request for research and commercial robotics teams.

Visit robox.to to request access or learn more.

License

All RoboX datasets are released under CC-BY-NC-4.0 for research and non-commercial use. For commercial licensing, contact the RoboX team via robox.to.