Humanoid data collection covers whole-body trajectories, dexterous hand manipulation, bipedal locomotion, and loco-manipulation — the full stack of behaviors a humanoid policy needs to learn. We collect across 24 platforms so your model transfers across embodiments.
Humanoid data is hard
High DoF, full-body coordination, and platform-specific kinematics make humanoid collection the most complex data work in robotics. We’ve built the operator bench and rig fleet to handle it.
8 active humanoid programs.
24 platforms supported.
50K+ trajectories delivered.
Where we collect
41+ delivery centers across 12 countries. Every program runs from a Roborax hub near your target time zone.
Asia Pacific
India · Philippines
Americas
USA · Canada · Colombia · Jamaica · El Salvador · Belize
EMEA
UK · Albania · Kosovo · Morocco
Four data classes that bipedal foundation models need but generic data pipelines rarely produce.
Coordinated upper + lower body kinematics, 30Hz, in your robot frame.
Same task captured across multiple humanoid morphologies for transfer learning.
Per-finger joint logs and grasp state for complex object handling.
Walking while carrying, reaching while balancing — the chains that break generic policies.
Four stages that produce humanoid-ready data, not adapted-from-arm data.
Bring up your humanoid or use ours. Joint calibration captured. URDF locked.
Whole-body teleoperators trained on your kinematic envelope and balance constraints.
Trajectories captured against your acceptance bar. Stability and contact monitored.
Same trajectory rendered in multiple morphologies where the platform supports retargeting.
Production humanoids, research bipeds, and the rigs that retarget between them.
Production
Production
Production
Production
Research
Production
Four verticals. One data partner.
Long-horizon tasks in real environments.
Pick-pack-place across real SKU diversity.
Procedure-grade demos on surgical platforms.
Remote operator-driven data collection.
In-person task demos for imitation learning.
RGB-D, LiDAR, force, and tactile streams.
FAQ
Across the spectrum — from pre-alpha hardware needing initial motion primitive data to commercial-stage teams scaling up training data for a specific deployment environment. There is no minimum stage.
The gap between “our policy works in the lab” and “our policy works reliably in varied real-world conditions”. We close that gap by adding diverse, high-quality real-world demonstration data.
Yes. We provide training data for teams building general-purpose robot foundation models that need broad task coverage across multiple platforms and environments.
From the blog
Lessons from 50,000 Humanoid TrajectoriesWhat 50,000 trajectories taught us about humanoid data collection.
From the blog
Humanoid Robot Training Data: How Much Do You Need?Volume requirements for humanoid foundation model training.
Tell us the platform, the tasks, and the morphology mix. Six weeks to first cross-embodiment batch.
FROM THE FIELD
Warehouse robots underperform in production not because of the model, but because training data missed the edge cases that matter.
Sub-millimeter precision, HIPAA compliance, and credentialed operators — surgical robot data has requirements general robotics programs cannot meet.
Robotics data quality is not a review meeting. At production scale it is automated validation, per-operator metrics, and same-day feedback loops.
Robot annotation is not image labeling with a new name. Temporal structure and task semantics demand distinct tooling and annotator qualification.
Simulation offers unlimited training data at zero cost. The sim-to-real gap is a structural problem, not a rendering one.
Language models scaled on internet data. Embodied AI must build its data from the physical world — and that changes everything.
