Network
Physical-world robot data collection at operational quality
The capture network is the field side of every dataset we ship: sites, calibrated rigs, task protocols, and chain of custody. Nothing here is synthetic filler for your training mix.
Field workflow
How a session runs
- 01
Site selection
We choose capture sites for task diversity: kitchens, workshops, warehouses, and offices. Sites are scouted for lighting, clearance, and mobility constraints that mirror real robot deployments.
- 02
Rig deployment
Operators deploy standardised kits: head and wrist POV (RGB + depth), IMU logging, and optional controller telemetry. Rigs are calibrated before day one and rechecked between sessions.
- 03
Capture protocols
Structured task scripts (pick and place, handover, navigation under occlusion, bimanual work) keep coverage intentional. Deviations are logged. Scripts are versioned with each release.
- 04
Sync & handoff
Session bundles are encrypted and uploaded securely. Intake checks queue annotation. Operators receive a quality summary the next morning.
Quality & governance
Standards you can cite
Researchers and enterprises need data they can publish and ship. Field and data standards are built for that bar. See also Process and Security.
- Informed consent before every session
- Faces and identifiers handled per agreement before data leaves the capture site
- Chain-of-custody from device to release
- Scripts and calibration logs retained with each dataset version
- Raw footage access limited to named ROBOTRAIN engineers
Partners
Site, hardware, and annotation partners
Property owners & operators
Structured on-site sessions with compensation and consent agreed upfront. Releases are anonymised per agreement before they leave our control.
Hardware manufacturers
Integrate your sensors into our rig standard. Approved partners can receive pre-release samples for benchmarking.
Annotation partners
We work with teams experienced in robotics labelling for spatial and temporal tasks.
Commission a capture program
Buyers: scope a custom dataset. Partners: tell us about sites, hardware, or annotation capacity.