ProductUnited States

Origami Robotics starts mass-producing S1 glove that records 20 finger joints and touch

Published 2 min read
Source

San Francisco startup Origami Robotics said in a blog post on Oct 21 that it has started mass production of S1, a glove that records 20 finger joint angles with encoders, pressure at the fingertips and palm, and wrist-camera video on one timeline, so human demonstrations map joint for joint onto its own robot hand.

The company reports millimetre-level fingertip accuracy against OptiTrack motion capture and says the glove kept measuring when cameras lost sight of the fingers; it gave no price or shipping date.

Origami Robotics is betting that the robot-data race will be won on what each recorded hour contains, not on how many hours there are. In a blog post on Oct 21 titled "A million hours of what?", the San Francisco startup introduced S1, a skill-capture glove, and said it has started mass production.

What S1 measures

According to the company, S1 reads 20 finger joint angles directly with joint encoders, measures pressure at the fingertips and palm, and records video from a wrist-mounted global-shutter camera on the same timeline. Origami says the glove kept measuring when a hand worked inside a container out of the camera's view, while the vision pipelines it tested as baselines (MediaPipe Hand Landmarker, and WiLoR detection followed by HaMeR reconstruction) lost the hand or jumped between finger poses. Checked against OptiTrack motion capture, fingertip positions computed from the glove were accurate to the millimetre level for the index, middle and ring fingers in the glove's wrist frame. Wearers performed 30 demanding tasks, from fast Rubik's Cube solves to card shuffling, and a musician played violin, piano and percussion in it.

The design is tied to Origami's own robot hand: both share the same joint arrangement and tactile sensors, so each glove angle becomes a position target for the matching robot joint without a retargeting solver. An earlier rigid-shell glove, built seven months ago, was used in a couple of factories but fitted too few hands, which led to the current soft shell. A model to correct for contact differences between human and robot fingertips is still in development, and a second-generation hand is coming. No price, customer or shipping date was given.

MEKAWIRE analysis

Origami is selling the opposite of mass video: fewer hours, but each one carries joint angles and contact force that cameras cannot see.

That puts it at one end of a spectrum MEKAWIRE has tracked this month. Figure's Index app has reportedly gathered 37.5 million phone videos, and Maxinsights raised a $30 million seed on more than 2 million hours of head-camera footage, while Sharpa pairs its AE01 haptic glove with its own hand, as Origami does.

How robot makers are capturing human manipulation data (MEKAWIRE coverage, Sept to Oct 2026)
  • Origami Robotics
    Capture method
    S1 glove: joint encoders, palm and fingertip pressure, wrist camera
    Scale or spec disclosed
    20 joints; mass production started
    Date
    Oct 21
  • Figure AI
    Capture method
    Index phone app, paid creators
    Scale or spec disclosed
    37.5M uploads, 112 countries (reported)
    Date
    Oct 22
  • Maxinsights
    Capture method
    Head-worn cameras, 5,000+ contractors
    Scale or spec disclosed
    2M+ hours; $30M seed
    Date
    Oct 22
  • Nature Will
    Capture method
    Egocentric human data for Natura-1 hand model
    Scale or spec disclosed
    1M+ hours
    Date
    Oct 13
  • Sharpa
    Capture method
    AE01 haptic glove plus W02 hand
    Scale or spec disclosed
    21 active DoF hand
    Date
    Sept 29

Figures as disclosed by each company in the stories linked; Figure's Oct 22 numbers are reported from its all-hands video. Units differ and are not comparable one to one.

As of Oct 25, 2026

The weakness is scale and lock-in: a glove matched to one robot hand produces data that transfers most cleanly to that hand, and the accuracy figures are the company's own test, not a general guarantee.

Why it matters

Hand makers and humanoid teams that struggle to turn video into force-aware manipulation now have another source of contact-rich data. Watch for Origami's second-generation hand, its contact-alignment model and the first named buyers of S1. See the Origami Robotics profile.

Good or bad news for the company?

 

Key facts
HQ
San Francisco, 450 Bryant Street
Status
Mass production started; price and ship date not disclosed
Product
S1 skill-capture glove for robot-learning data
Sensing
20 finger joint angles by encoder; pressure at fingertips and palm; wrist global-shutter camera
Accuracy
Millimetre-level fingertip position vs OptiTrack (index, middle, ring; company test)
Announced
Company blog post, Oct 21, 2026
Robot link
Same joint layout and tactile sensors as Origami's robot hand; no retargeting solver
Sources