Collecting data for
humanoids.

The glove measures joint angles and grip force in newtons while your hand works. It writes episodes your training stack already reads.

Reserve the first run →
Washing up, filmed from the head. The hand sits under water or behind a plate for most of the clip.

01 · The gap

Robots can see.
They cannot

Policies break at the grasp. Once the object covers the fingers, the camera has nothing left to measure, and pressure was never in the frame to begin with.

02 · What one glove records

Force
100N

pressure at the fingertip, calibrated from 0.1 N, normal and shear

Position
26DoF

read at the joint itself, on the hand

Rate
120Hz

force and position sampled on the same frame

Sync
<10ms

one clock for both streams, checked on every episode

DG-1 concept render, palm view showing the fingertip force pads.

DG-1 · concept render

03 · The hardware

Built to disappear on the hand.

  • 01
    Three force pads.
    They sit on the thumb, index and middle finger and read normal and shear force in newtons.
  • 02
    Sixteen motion nodes.
    Each one sits flush in the knit and tracks its joint at 120 Hz.
  • 03
    One clock.
    It timestamps force and position together, so what comes out is an episode your stack can open.
DG-1 concept render: macro of the fingertips with a magnified callout on one force pad.

DG-1 · concept render

DG-1 · concept animation · specs are design targets

04 · The evidence

Video alone cannot capture the rich contact signals critical for mastering manipulation.

Meta FAIR, arXiv:2512.08920

−58%

accuracy once the object covers the fingers

visible joints
52 to 61
occluded joints
23 to 27
force in video
0

PCK on standard hand-pose benchmarks

See the full evidence →

05 · Side by side

How robots learn manipulation today.

The force column is where the table thins out.

Data source Finger position Grip force During contact
(fingers hidden)		 Cost to scale
Internet & egocentric videoGuessed from pixelsNoneAccuracy halvesCheap, scrapable
Vision hand-trackingEstimatedNoneFails at the graspCheap
Robot teleoperationRobot jointsRare, uncalibratedYesA robot per operator
SimulationPerfect but syntheticNo validated contact physicsYesNear-free
Digitus gloveMeasured · 26 DoF @ 120 HzCalibrated · 0.1 to 100 NUnaffectedHuman-speed, no robot

06 · Work with us

Tell us the hand you're training.

Send us the robot hand you train and the format you need. We reply with a draft LOI and a quote inside two business days.

Industry LOI · hold gloves from the first run at locked pricing.
Academic · research pricing, the pipeline software, co-authorship.
Consumer · creators, VR and telepresence, further out.
Your email draft is open. Press send in your mail app and you are on the list.
Your email draft is ready. Send it and we'll reply within two business days.