Remote human data collection
Capture humans wherever the work happens, and turn it into training-ready datasets for humanoid policies. No cameras, no markers, no line of sight, no lab.
Ground truth from the real world
Policies trained on lab-clean data meet a messier world at deployment. Inertial capture removes the lab from the equation.
Capture any movement, anywhere
No cameras, no markers, no line of sight. Indoor or outdoor, factory floor, warehouse, field or construction site. Walking, running, jumping, crawling, climbing stairs, lifting, reaching, crouching: the dataset reflects the world the humanoid will be deployed into.
Repeatable, validated quality
A standardized capture and QA pipeline keeps data consistent across operators, sessions, tasks and environments. The biomechanical model is scientifically validated, with native male and female anatomy in Biomech 2.0.
Never lose a day of capture
On-body recording runs a full working day untethered, no laptop or backpack PC in the loop. The session survives a dropped wireless link, and hot-swappable batteries keep capture going.
On-body capture rate
On-body recording
Wireless range
Battery, hot-swappable
From capture to training-ready dataset
Four stages. Whether you run the program in-house or we run it for you, the pipeline is the same.
Scope the dataset
Define the tasks, environments, operators and volume the policy needs. That maps onto a capture plan, a body configuration and the add-on modules that match the task.
Capture on site
Operators wear the suit and do the work. Setup takes minutes, recording runs on-body, and sessions continue through shift changes without a workstation in the loop.
Validate and QA
Every session runs through a QA pass: kinematic consistency, foot contact, drift and segment scaling. Recorded takes can be reprocessed at higher quality afterwards.
Export and integrate
One-click export into the formats your stack already reads, or stream straight into your pipeline. Retargeting onto your specific humanoid happens here.
Every movement a humanoid has to learn
Full-body kinematics at 240 Hz, with ground-truth foot contact and centre of mass. Not just the arms, and not just the demo.
Gait and terrain
Walking, running, turning, stair climbing, ramps, uneven ground and crouched movement. Foot contact phases and centre of pressure come out of the model, so balance and gait policies learn from measured contact rather than inferred contact.
Reach and dexterity
Reaching, grasping, placing, tool use and two-handed work. Add finger tracking and the dataset carries per-finger degrees of freedom inside the same capture loop.
Load handling
Lifting, carrying, pushing, pulling and crouched work under load. Segment scaling and anatomical modelling keep the kinematics valid across body types and payloads.
Interaction in context
Track tools, payloads and workspace objects alongside the operator, and lock the capture to a global reference frame. The dataset records the interaction, not only the body.
Hardware, software and integrations
Three layers that turn human captures into training-ready datasets.
Motion capture suit
Full-body humanoid-grade capture with on-body recording, so no session is ever lost. Works indoor or outdoor with no cameras, markers or line-of-sight requirement. Modular body set-ups, one-click sensor mounting and all-day wearable comfort.
Xsens Humanoid Software / SDK
Capture, validate and export. A scalable biomechanical model with built-in QA, available off-the-shelf for Linux and Windows or in SDK form. HD reprocessing lifts the quality of recorded takes after the fact.
Exports and streams
Native ROS 2 publisher and native exports into the simulators, middleware and tooling humanoid teams already run, with a native NVIDIA Isaac plugin rolling out. Datasets land in your training stack ready to use.
One dataset, many exports
Capture once. Export into whatever your training, simulation and analysis stack expects.
Straight into training
URDF-ready output, MJCF and a native ROS 2 publisher, with NVIDIA Isaac Sim and MuJoCo in the loop. Retargeting onto your humanoid is supported through XsensTargeter and research partners including GMR (Stanford), Westlake and MPI.
Into the DCC pipeline
BVH and FBX into MotionBuilder, Maya, Blender, Unreal Engine and Unity. Useful for review, annotation and building synthetic variation on top of measured motion.
Data, not just motion
C3D, MVNX, SMPL, CSV and XLS for biomechanical analysis and dataset inspection, with OpenSim, Visual3D, Siemens Tecnomatix and Dassault Delmia supported.
Expand your data collection
Layer the add-ons that match the dataset you want to build. All on the same capture, on the same clock.
Finger tracking
Manus or StretchSense gloves add full per-finger degrees of freedom inside the same capture loop. Required for dexterous manipulation, grasping and contact-rich tasks.
Object tracking
Track tools, payloads and workspace objects alongside the human, using HTC Ultimate Tracker or Vive. The dataset captures the interaction, not just the body.
Video sync
Time-locked multi-camera video streamed alongside motion. Vision and biomechanics on one clock, ready for VLA training and annotation.
HMD integration
Pair captures with VR headset data for first-person datasets and immersive review. Meta Quest and Pico VR support coming soon.
Absolute positioning
Lock the suit to a global reference frame so datasets become spatially grounded for interaction and scene-aware tasks.
Not sure what you need?
Tell us the task you want the humanoid to learn and we will map it onto the right capture configuration.
Build the program, or let us capture it
Same hardware, same model, same exports. The difference is who runs the operation.
Run it in-house
You own the suits, the software and the pipeline, and capture continuously at your own pace. Best when data collection is an ongoing part of the program, when captures happen on your own site, or when the data cannot leave your premises.
Data Collection as a Service
We install the infrastructure on site, run the sessions and deliver the dataset. Best when you want the data without building the operation, or when you want to validate quality before committing to a program.
Ready to build your dataset?
Tell us the tasks you want the humanoid to learn. We will configure the capture setup, the modules and the export path for your program.