Physical intelligence — dataset catalogue
Robotics data, captured in the real world.
Egocentric video, teleoperation traces, tactile glove streams and annotated manipulation data — collected by trained workers across residential, commercial and industrial environments.
Off-the-shelf packs ship today; custom collections are scoped to your embodiment.
Mono Egocentric Data
ResidentialSouth AsiaView sample- 800
- 20,000
- 10,000
- iPhone 13 or newer
Mono Egocentric Data
CommercialSouth AsiaView sample- 500
- 18,000
- 15,000
- iPhone 13 or newer
Mono Egocentric Data
ResidentialLatin AmericaView sample- 100
- 15,000
- 8,000
- GoPro or any device
Ego with IMU
ResidentialAsiaView sample- 800
- 20,000
- 10,000
- GoPro or any device
Stereo Egocentric Data
ResidentialSouth AsiaView sample- 10
- 12
- 100
- GoPro or any device
Full dataset captured with left and right wrist views.
Stereo Ego
Industrial 60% / Residential 40%AsiaView sample- 50
- 500
- 5,000
- GoPro or any device
Full stereo calibration included — depth and disparity can be computed downstream via standard stereo matching.
UMI Gripper Data
Residential / CommercialAsiaView sample- 800
- 20,000
- 10,000
- GoPro or any device
Teleoperation
ResidentialAsiaView sample- 20
- 100
- 600
- GoPro or any device
Teleoperation — High Fidelity
ResidentialAsiaView sample- 20
- 8,000
- 5,000
- GoPro or any device
Optical Glove Data
Residential / CommercialAsiaView sample- Custom
- Custom
- Custom
- GoPro or any device
Tactile Glove Data
Industrial 60% / Household 40%AsiaView sample- 50
- 500
- 5,000
- GoPro or any device
Hand pose plus contact data for dexterous manipulation.
Annotated Data
CommercialSouth AsiaView sample- 500
- 18,000
- 15,000
- iPhone 13 or newer
Annotated Data
Residential / CommercialAsiaView sample- 800
- 20,000
- 10,000
- GoPro or any device
Streaming previews are compressed by Drive — download samples for full fidelity. North America collections are opening soon.
How it’s built
Collected by people, verified by process.
Trained collection network
Every contributor is onboarded, briefed per taxonomy, and monitored for capture quality across households, shops and factory floors.
Verified at every step
Multi-stage QA on framing, lighting, task completion and metadata before a single clip enters a delivery batch.
Consented & anonymised
All footage is collected with informed consent, PII-scrubbed, and referenced by anonymised codes end to end.