One protocol, every site
Tasks, scene variables, devices and calibration are specified the same way at every capture site.
ROBOT LEARNING, CONNECTED.
Human skill in,
robot skill out.
Wearable capture, robot workstations and a data engine
that turn human demonstrations into robot training data.
University labs / Corporate R&D / Data capture teams
FROM HUMAN TO ROBOT
DATAENGINE
CaptureBuild datasetsLearn and validate
Use the arrow keys or drag the cards to switch. Autoplay pauses on hover or focus.
THREE WAYS TO START
Start where you are, not from scratch.
Capture a single task, stand up a robot lab or run multimodal research. Buy devices on their own, or get a complete system built around your goal.
THE LEARNING LOOP
Every manipulation
feeds the next round of learning.
Devices to data, models back to tasks.
Follow a single recording through the full research loop.
Failure attribution → targeted re-capture → next data version
01/ 04
First, record what happened.
Capture egocentric video, demonstrations and contact signals in one shared task context. Pick Ego, UMI, gloves, armbands or real-robot capture to fit the scene.
Task definition / device records / raw dataEach stage is configured and validated per project.
BUILT FOR DATA OPERATIONS
More than a lab setup.
Designed for capture at scale.
Enterprises don't need more video. They need data that fits the task definition, passes QA, stays traceable and comes with clear usage rights.
See the data delivery processAccept only valid data
Define what counts as a valid episode, and keep rejection reasons, spot-check records and batch reports.
Own your data
Versioned datasets with field dictionaries and access control, not scattered folders.
THE HARDWARE, CONNECTED
One system. Every part has a job.
Robots and dexterous hands act in the real world. Ego cameras, grippers, gloves and armbands each capture a different layer of human demonstration.
Modular platform
Atlas R1
A7 Dual-Arm Modular Research Platform
2 cameras + IMUMultiple configurations
Thor-Duo
Stereo visual-inertial capture device
- Cameras
- 2 × 1080P
- Device weight
- ~180 g
- IMU sampling rate
- 300 Hz
UMI GripperMultiple configurations
UMI Gripper
Handheld manipulation demonstration kit
Tactile Data GloveMultiple configurations
Tactile Data Glove
Tactile mapping with electromagnetic tracking
DATA → POLICY → EVALUATION
TRAIN ON YOUR TERMS
Compute sized to the experiment,
models evaluated on real tasks.
Capture workstations, on-robot inference and cloud training are planned separately. Size compute to your policy, data scale and deployment constraints, or plug into your existing environment.
A CLEAR PATH TO DELIVERY
Define success before you build.
From pilot to full delivery, every step has a clear deliverable and a sign-off.
Define the task
Confirm the research question, existing equipment, data rights, timeline and acceptance criteria.
Validate with a pilot
Validate equipment, synchronization, interfaces and baselines on a small batch or a single task.
Configure and deploy
Freeze the hardware and software scope, then deploy, capture or process data in batches.
Accept and iterate
Deliver the configuration, data and reports, then use failure records to plan the next round.
LET’S BUILD YOUR NEXT EXPERIMENT
Contact us.
Every partnership starts with a task.
Whether you're choosing devices, setting up a lab or running enterprise capture, start by telling us your goals and current setup.
YOUR RESEARCH BRIEF
Tell us about your project.
Name, contact details and project needs are required. Everything else is optional.