Hugging FacePublic datasets
Dataset collection
PI-Dojo / datasets
- Egocentric
- Manipulation
- MCAP
Physical AI data infrastructure
Grounded, robot-native training data from real human demos
We bridge the robot data gap by transforming real-world human manipulation demonstrations into robotics training datasets.End-to-end data infrastructure—from human demonstration capture and 3D grounding to robot trajectory retargeting—powered by our MSME partner network for scalable, real-world data collection.
Multimodal Grounding
- vision
- ◦ RGB ◦ stereo pair ◦ stereo depth
- imu
- ◦ accelerometer ◦ gyroscope
- hands
- ◦ keypoints ◦ 3d-pose
- head tracking
- ◦ camera pose ◦ trajectory
- simulation
- ◦ robot embodiment ◦ motion replay
- robot-native trajectories
- ◦ time-indexed joint configurations (qpos)
- calibration
- ◦ intrinsics ◦ extrinsics
- task
- ◦ labels
Open data · Real-world demonstrations
Explore Our Data
We’ve made a sample of our datasets publicly available on Hugging Face. Real-world egocentric manipulation data, packaged in MCAP and structured for robotics and Physical AI research.
Contact us
Build With Us
Tell us about your target tasks and environments. We’ll take it from there.