BERLIN, 15 August 2026 — Input-A, the Berlin and Delaware start-up building the wand — a handheld remote that types what you write in the air — today published the Air-Writing Atlas: an analysis of 52,191 letters, digits and shapes written in the air by 349 paid adult volunteers, each recording carrying the writer’s age, sex and handedness. It is, to the company’s knowledge, the largest such dataset described in public.
The recordings were collected to teach the wand’s recognition model to read air-writing. Tested leave-one-writer-out — trained on everyone else, then read on a person it had never seen — the model reads left-handers as well as right-handers (68.5% vs 68.2% of letters, cold start), women as well as men (68.0% vs 67.4%), and the over-60s as well as anyone. Whole words come out exactly right 85% of the time for a first-time writer (317 writers), and 93% after a short personalisation pass (12 writers, all 12 improved).
The one group the model struggled with was the 18-to-24-year-olds: 55% of letters against about 70% for everyone else, a gap that repeated on a second, later model. One in five of the under-25s wrote “micro-letters” — under 20 cm of hand travel per letter, thumb-typing scale — where most people draw a letter the size of a dinner plate. The model, trained on dinner plates, reads micro-letters 8% of the time. With the micro-writers set aside, the remaining under-25s still read a little lower, wrote smaller, hesitated more and followed on-screen instructions less closely.
“I’m not going to claim a generation has forgotten how to write; twenty-three people is a lead, not a verdict,” said John Usher, co-founder of Input-A. “But it does look as though the hand that grew up on a glass screen makes small, quick, thumb-scale marks — and a machine trained on the rest of us has to learn that too.”
Two further findings from the same data:
- Men and women write the same in the air. Nine kinematic measures — time per letter, size, speed, slant, roundness, smoothness, initiation latency, direction of circles, and the wobble of the hand at rest — showed no meaningful difference (every effect size within ±0.2). A model given all nine guesses a writer’s sex correctly 58% of the time. The one measurable difference is the pitch of the tiny wobble in a hand holding the phone still — slightly lower in women, the same amplitude in both — most likely because the same phone is a larger load for a smaller hand.
- Older hands don’t shake more; they slow down. Time per letter is flat from 18 to about 50 (1.7 s), then rises about a fifth of a second per decade. The wobble of a resting hand — its amplitude and its pitch — does not change with age across 18–78, and neither do the size, slant or roundness of the letters.
Also in the atlas: 85% of right-handers’ free circles run anticlockwise — matching the figure Google’s Quick, Draw! found for the US on a screen — while a third of left-handers’ circles run clockwise (odds ×2.9); left-handers’ downstrokes lean back; the fastest glyph is the letter i (1.1 s) and the slowest a diamond (2.4 s).
How the study was done. Participants were adults recruited and paid on Prolific who consented to their phone’s motion data being recorded for research; only aggregates are reported. Each wrote a–z, 0–9, five shapes and six arrows three times each, holding a phone like a TV remote; motion was recorded by the phone’s accelerometer and gyroscope at 60 Hz. Hypotheses, glyph sets and tests were written down before the analysis ran; effects carry 95% confidence intervals and false-discovery control, and every reported effect replicates within iPhone and within Android and with each recruitment wave left out. Recognition figures are leave-one-subject-out. Limitations: an online-panel sample; 13 people over 65 and 23 under 25; two phone platforms of many models; hand size not measured. The study is kinematics — how the hand moves — not graphology, and makes no medical claim. Input-A has patents pending on the wand’s recognition and personalisation and sells no product yet.
Charts, the write-up and the method: https://the-wand.ai/blog/air-writing-atlas/ · Try the recogniser in a browser: https://phone-demo.the-wand.ai/write
About the wand / Input-A. Input-A (Inputa Inc., Delaware; engineering in Berlin) makes the wand, a small handheld remote that lets you write on a screen by drawing letters in the air. Handwriting is recognised from motion alone, on the device — no camera, no microphone — and the wand connects as a standard Bluetooth keyboard. The recognition model was trained on air-writing collected from paid, consenting participants; the Air-Writing Atlas is what that data says about the hands that produced it. Founder John Usher is an audio-DSP engineer (McGill PhD; an upmixer credit that shipped inside Dolby Atmos; TWS-earphone signal processing). https://the-wand.ai/about/
Media contact: John Usher · [email protected] · Berlin (CET)
Assets: six charts (1600×1000 PNG) and a six-page PDF at https://the-wand.ai/press/air-writing-atlas/cards/ · OG image https://the-wand.ai/og.png