KIH: Keyboard In my Hand

Mechatronics · XR · team lead · Dec 2025 – Sep 2026

A glove-type wearable Hangul keyboard for eyes-free Korean text entry. 16 tactile switches are embedded across the phalanxes of a pair of gloves (8 per hand) and pressed by the thumbs. Input is sequential rather than chorded; repeated taps on the same button add strokes, mirroring the graphical derivation principle (가획) of Hangul. Enumerates as a plug-and-play USB HID keyboard — no drivers or companion app required.

Motivation

Three gaps drove the design:

VR/AR text entry, where physical keyboards break down;
affordable tactile input for Korean visually-impaired users;
posture-related VDT syndrome (7.05 M reported cases in Korea, 2024).

Mapping & Input Method

Each of the 16 buttons is mapped to a base jamo. Repeated taps on the same button add strokes — the same operation that graphically derives ㅋ from ㄱ, ㄲ from ㄱ, or ㅏ from ㅣ — so no separate shift key is needed.

16-button Hangul mapping layout — 16-button base-jamo mapping across both gloves.

Multi-tap timeline for stroke addition (가획) — Multi-tap timing: successive taps within a window add strokes (가획).

Results

A within-subject study with 20 participants compared KIH against a TV-style cursor on-screen keyboard. After 10 minutes of training per condition, each participant completed 15 one-minute trials typing random words without looking at the device.

2.3 → 10.6 WPM from trial 1 to 15 (4.6×), while the cursor keyboard plateaued at 5.6 → 7.5 WPM; KIH overtakes it at around trial 9.
Error rate 17.2% → 4.1% — speed and accuracy improved together.
KSPC 1.27 — a 26% improvement over frequency-blind layouts; 74.2% of input is a single tap.
Per-user tap-window calibration; total BOM cost ₩14,184.

Learning curves and error rates for KIH vs TV on-screen keyboard — Learning curves over 15 trials: KIH keeps improving; the cursor keyboard is saturated from the start.

Jamo frequency vs. required tap count — Jamo frequency distribution vs. required tap count — the basis of the KSPC gain.

System & Stack

Arduino Leonardo (ATmega32u4), native USB HID. Firmware in C++ (Arduino); calibration, study and analysis tools in Python. Hardware: 16 tactile switches, gloves, wiring harness. MIT license.

System overview: prototype with signal flow diagram — System overview: prototype with signal flow.

Beyond the Glove: KIH-XR

The input method does not depend on the glove. KIH-XR runs the same 16-spot layout on camera hand tracking alone: MediaPipe estimates 21 joints per hand, virtual buttons are placed on the phalanxes, and a thumb-tip touch counts as a tap. The firmware's multi-tap state machine is ported unchanged, and a calibration wizard fits each user's sweet spots, tap radius and tap window. Only the joint source is device-specific, so headsets that already track hands (Vision Pro, Quest) could use the same layout without extra hardware. Built as a live webcam demo for the Creative Design Festival.

KIH-XR source & docs ↗

Recognition

Bronze Prize, 15th SNU College of Engineering Creative Design Festival (Sep 2026).
Grand Prize, SNU Mechatronics Competition (Dec 2025).

Role

Team of 4, team lead. Led the design and pitch; earned faculty praise for a compelling presentation.

GitHub ↗ ← Back to profile