Start a data program →
See the fusion ▸
Adding pose, force and intent labels that
cameras cannot produce.
13 Patents
7 granted Patents + 6 pending Patents
surface EMG · neural input
12+ yrs
surface EMG R&D
Wearable Devices Ltd.
6
shipped devices
consumer-proven hardware
Training robotics’ dexterous hand manipulation is hard! By now it is captured with computer vision only - egocentric and palm cameras which suffer from:
01
Self-occlusion
Tools, hands, and equipment frequently block camera views during real-world tasks, causing critical loss of tracking data.
02
Closed-hand occlusion
Fingers hide fingers. A closed grip is the most common hand state and the least observable one.
03
Force is invisible
No camera measures newtons. Grip, pressure and effort simply are not in the frame at any resolution.
04
Fast motion
Vision can only register a movement once it exists. The command that caused it has already happened.
Fuse our EMG-based neural wristband with any egocentric camera array. No gloves, no line-of-sight dependency, and no teleoperation rig.
Start a data program
Less Data. Less Time. Higher Quality.
Recorded hours vs. Data quality comparing the two paradigms of camera vision only data, and the addition of the Mudra neural layer.
Mudra Neural Layer - Data Collection Efficiency
*All figures are illustrative and for concept presentation only
0h
RECORDED HOURS
DATA QUALITY
Camera + Mudra Neural Layer
Camera Vision Only
Cost 01
You pay full price for partial data
An hour of camera-only capture costs the same operator time, the same annotation and the same QA as a fused hour. It arrives missing critical manipulation info and suffers from diminishing returns.
Cost 02
Recollection means paying twice
The subjects, the sites, the protocol and the annotation all run again. The second pass costs what the first one did - and the calendar you spend on it is not recoverable either.
Cost 03
The fleets are training now
Humanoid robots are shipping to homes throughout 2026 and learning from human demonstration. Nobody re-runs last year's dataset. Whatever gets recorded now is what those policies learn from.
We don't replace your camera stack. We upgrade it using a non line-of-sight neural layer - surface EMG array, IMU and PPG at the wrist, fused with any egocentric or fixed camera, indoors or out.
Output 01
Hand pose & joint angles
Glove-grade kinematics from wrist sEMG fused with any camera - glove estimation without the glove, and robust when the hand leaves the frame.
Output 02
Pressure & force
Fingertip and grip force that cameras cannot see, including sEMG-based pose dependent force estimation.
Output 03
Point-of-no-return
The exact moment manipulation begins. The neural command fires before contact is visible, so we timestamp intent itself.
EGOCENTRIC CAMERA
MUDRA WRISTBAND
sEMG · IMU · PPG
FUSION
MODELS
3-D POSE
FORCE
ONSET
MUDRA PRO\Ultimate - PRODUCTION sEMG BAND
ALREADY RUNNING IN AI AND XR PROGRAMS
Any camera + the band · three outputs vision can't produce
Start a data program
Mudra Ultimate is our research-grade capture band at the core of every partnership. Mudra Pro is a cost effective solution for Academia and exploration.
3 EMG channels · launched May 2026
Mudra Pro
8 EMG channels · research-grade
Mudra Ultimate
Full forearm coverage for spatial decomposition, WiFi streaming, multi-sensor synchronization out of the box. The band at the centre of every research partnership.
Two ways to partner
with us.
For physical-AI companies
Add the wearable layer to your data program
You are already paying for annotated human recordings. Adding pose, force and intent labels that cameras cannot produce makes each hour of that dataset materially more valuable, we synchronize the capture alongside your existing stack rather than replacing it, via LSL.
For research labs
Mudra as a research kit
Mudra Pro and Ultimate as instruments for academic groups working on manipulation, prosthetics, HCI and neuromotor interfaces. Publications and benchmarks from independent labs are worth more to this field than anything we could claim ourselves.
How a programme runs
Step 1
Scope
One call to define the task, the modalities and what “better” looks like measurably. We'll tell you if we're the wrong tool.
Step 2
Pilot
A bounded recording - tight protocol, small subject count, synchronised to your streams. You train on it and we both read the result.
Step 3
Scale
If the pilot moves the metric, we scale subjects, sites and session hours, and hand over the tooling to run it yourselves.
Specs.
If your model needs the hand and your cameras keep losing it, the conversation is the same one. Share with us the task and the failure case.
Name
Company or lab
What you're building
The task, the model, and where the hand data breaks down.
Current sensor stack
e.g. egocentric RGB-D + mocap + LSL
What you want first
A scoping call
Send →