tap localization via accelerometer

Train a classifier to recognize where on the phone you're tapping — screen, back, edges — purely from motion data. iOS requires explicit permission for the DeviceMotion API.

how it works

Pick a zone, tap that part of the phone repeatedly to collect samples, then switch to inference.

Screen uses your touchscreen taps as labels — tap anywhere on the capture surface.
Back & edges use accelerometer impulses — physically tap that part of the phone. Hold the phone firmly elsewhere.
5–10 samples per zone is enough to start. Inference unlocks at 3+ samples in 2+ zones.
In inference, confirm or correct each prediction — feedback gets added to training (RLHF).

x y z ω

0.00 m/s²

training zones / tap to start 0 samples

threshold 2.5 cooldown 350ms

x y z ω

0.00 m/s²

predicted zone

tap any surface

actual touch

predicted (accel)

error: —

trigger

—

confidence

—

peak g

—

nn dist

—

tap the actual zone — or confirm

0 right / 0 wrong —