Human sideDermis Glove
Sensing patches on every fingertip and the palm, on a thin stretch liner, synced with webcam or VR hand tracking.
Calibrated touch for human and robot hands
Dermis pods give a human glove and a robot hand the same calibrated sense of touch. Swap a skin, recalibrate in minutes, and keep reading pressure and shear in newtons.
6 pods · synced · in newtons
01The problem
Robots learn to grip by copying human demonstrations. But the glove on the person and the sensors on the robot measure touch in different ways and different units, so what the person felt never reaches the robot.
In recent studies, robot policies that use touch beat the same policies without it by 16 to 64 percentage points on contact-heavy tasks. The gains are largest where grip force and slip decide the outcome.
Key removal and plug insertion: NeoteAI and Fudan’s VTLA vs. vision-only, reported by BestHub (Jul 2026). Simulated contact tasks: N0-VTLA vs. vision-only ACT, arXiv 2607.23782. Wiping: OSMO glove vs. vision + proprioception, arXiv 2512.08920. Baselines differ by study.
02Idea
One modular pod system for both hands. The pod on your demonstrator’s glove is the same part as the pod on your robot — same skin, same magnet, same calibration — so touch recorded by people reads the same on the robot.
Sensing patches on every fingertip and the palm, on a thin stretch liner, synced with webcam or VR hand tracking.
Twin pair
Same contact. Same newtons.
Paired location by location and pressed with the same load on our rig. Every kit ships with its measured twin gap.
Identical pods on mounts for your robot hand — or sold alone, to match glove data you already collect.
What makes it different
Change the skin, keep the newtons — matched on glove and robot.
Change a skin, run a short check with known weights, and you’re back in newtons in minutes — no lab rig, no vendor visit. Each pod checks its own baseline at power-on and flags drift before it ruins a dataset.
Skins and pods snap on and off. Each carries its own ID and calibration, so a worn skin is a quick part swap, not a lost dataset.
One sensor core, any cap: fingertips, palms, gripper jaws or tool handles. Use tested caps for popular hands, or design your own to our open mount spec.
Glove, robot pods, hub, calibration and software in one kit, matched from one design. No stitching vendors together.
| Who | Human glove | Robot skin you can buy | Calibrated as a matched pair | Recalibrate yourself after a change |
|---|---|---|---|---|
| Tactile data companiese.g. 6thSense: sells data; glove stays in-house | In-house only | — | — | Doesn’t apply |
| Robot-side tactile sensorse.g. XELA uSkin; no human-side device | — | Yes | — | Per-sensor calibration is a paid add-on |
| Robot hands with built-in skine.g. Proception ProHand and ProGlove | Same skin as their hand; sale not announced | Only on their own hand | Same skin; no calibration published | Not published |
| Open-source magnetic skinse.g. AnySkin, eFlesh, OSMO | Build it yourself | Build it yourself | Left to each lab | Left to each lab |
| DermisThe matched touch layer | Yes | Yes, also sold alone | Yes, per kit | Yes, in minutes |
Why it matters
Every recording keeps each skin’s calibration ID, so data captured on a Dermis glove replays on any robot wearing Dermis pods.
The Dermis flywheel
more robots ⇄ more data
The touch half of human demonstrations finally transfers to the robot, not just the motion.
Calibrated units and open exports instead of one lab’s private format.
Models that infer touch from video still need measured touch to train on and be checked against.
03How it works
Record demonstrations with the Dermis glove, synced with video and hand pose.
Every pod reads in newtons, and each skin carries its own ID and calibration.
Robot pods match the glove location by location, with a measured twin gap.
Export to LeRobot or MCAP and run the policy on a robot that feels the same way.
Founding Kit targets, measured and published per kit.
04Product
Calibrated 3-axis touch for the human demonstrator and the robot hand that learns from them. Everything a lab needs to capture, calibrate and export matched touch data.
A thin stretch liner with five fingertip pods and a thenar palm pad. Light enough for long sessions, and compatible with the hand tracking you already own.
The same pod as the glove — same skin, magnet and calibration model — on printed mounts for your robot hand.
Lossless, timestamped, self-describing data from every pod, over USB-C.
Dermis Studio on macOS is the instrument; the Linux recorder, Python SDK and ROS 2 driver are what robot stacks run every day. Both write one session format.
Illustrative signals. Dermis Studio for macOS is in development; the Linux recorder, Python SDK and ROS 2 driver write the same session format.