Original LinkedIn post
Live control from muscle signals alone
Forearm EMG decoded into input as it happens — the sleeve reads what the hand is doing while it does it.
Watch on LinkedInThe dataset
robotics is
missing.
We help manufacturers automate dexterous tasks by recording how their best workers perform them — and using that data to train robots.
Our EMG sleeve captures physical information that cameras miss.
Try a gesture ↗
Recorded forearm EMG · 32 ch · 6 repeats · 2,048 Hz
01EMG + video + hand tracking02Hand completely unobstructed03Robot needs no EMG at inference
Follow the signal Backed by
- CombinatorY Combinator
Boost VC- Sterling Road
- Oxford Seed FundOxford Seed Fund
Cameras show
what happened.
EMG reveals how.
Video tells you where the hand moved. EMG gives another signal for what the human was physically doing to make that movement happen — muscle activation, inferred effort, the preparation before contact. We measure further upstream in the control signal than a glove, and the hand stays completely free.
Illustrative signal. Real data streams from the wearable sleeve on the worker's forearm.
Record the expert. Deploy the robot.
Your line keeps running. We record how the work is actually done — then train a robot to do it.
- 01
Raw EMG · video · pose
Record
Your best workers do the task wearing our EMG sleeve, on camera, with hand tracking. The sleeve leaves the hand completely unobstructed.
- 02
Aligned dataset
Build the dataset
EMG, video, and hand pose are aligned into a task-specific dataset — including the muscle activation and inferred effort cameras miss.
- 03
Retargeted policy
Train
Demonstrations aren't copied one-to-one. They're retargeted into the robot's action space to train a policy for your task.
- 04
Robot's own sensors
Deploy
The robot runs on its own sensors — it never wears the sleeve. The EMG guides training, then drops away.
Every deployment produces physical-interaction data that exists nowhere else. That data makes the next deployment better.
In factories now.
A European manufacturer identified delicate, force-sensitive sewing as one of its hardest automation problems — and signed an LOI for us to support robotic implementation.
We're also on construction sites, recording electricians, plumbers, and demolition crews — completely different work, captured with the same sleeve.
We started with sewing because it's brutally hard for a robot: the fabric stretches, the grip changes constantly, and one clumsy move ruins the piece. Crack that, and easier tasks follow.
Built for tasks that
resist automation.
If a robot integrator has already told you no, that's our starting point.
Fabric, cables, gaskets — material that moves when you touch it and punishes a clumsy grip.
Rhythm / adaptationInsertions, fits, and surface work where inferred effort matters more than position.
Precision / restraintTasks built around how a skilled operator handles the tool, not just where it goes.
Grip / engagementSkilled manual work outside the cage — recorded where it actually happens.
Force / anticipationEvery skill leaves a different signal · illustrative traces
For robotics and AI labs — generalised datasets of human physical interaction, recorded during real production work: EMG, video, and hand pose, aligned.
Get in touchWatch it work.
The sleeve decoding real muscle activity, as it happens.
Original LinkedIn post
It knows what the hand is holding
Decoding the held object from EMG alone — the kind of grip and contact information a camera can't see.
Watch on LinkedInLive demos as posted on LinkedIn.
Meet the founders.
Axonnic / Oxford originsCEO & Co-Founder01
Maxim Williams
Oxford MBiol research — signal processing · EMG calibration paper in preparation (1st author)
Built with a hardware company from the ground up · Seen a unicorn scale from the early days
CTO & Co-Founder02
Arsh Patankar
Oxford medic — research in machine learning and AI in cancer and stroke imaging
Published multiple papers · Repeat founder — medical LLM used at scale · Worked on Oxford's first quadruped robot
Advisors
Voxblock & Skyscanner Founders
Oxford · Harvard · UC Berkeley Researchers
Top Neurosurgeons
Common
questions.
A robot's sensors tell it what it is doing. They can't tell it how a skilled human does the task — how much effort goes into a grip, which muscles engage before contact. EMG captures that from the worker, and we use it to teach the robot the parts of the skill a camera can't explain. Once trained, the robot runs on its own sensors.
No. EMG is used only during training. The deployed robot runs on its own sensors — the demonstrations are retargeted into its action space.
We measure further upstream in the control signal than a glove, and the sleeve leaves the hand completely unobstructed — your workers do the real task, with real dexterity, at real speed.
You identify a dexterous task you can't automate. We record your best workers doing it on-site — sleeve, video, hand tracking — build the task dataset, and work toward robotic implementation. Production doesn't stop.
Terms are set per engagement. Recordings from your line are used to automate your task; how data contributes to broader models is agreed contractually up front.
Join the
waitlist.
Manufacturers, robotics teams, researchers — or anyone who wants in early. Tell us what you're working on and we'll be in touch.
Manufacturer pilots · Lab datasets · Early access
No spam. Early access and updates only.