Vision models trained for walking
Object detection fine-tuned on our own street recordings for holes, bollards, poles, bins, trees and
crosswalks, plus a walkable-ground model that separates paving from grass and soil.
Distance, not just names
Every warning comes with metres. Distance is estimated from known object sizes, the ground plane and
a monocular depth network – from a single phone camera.
Speaks only when it matters
Early and urgent warnings, priorities that let an urgent warning interrupt, and silence when the way
is clear. A natural human voice, not a robotic one.
Built for VoiceOver
Two large touch zones, one tap, one hold. Everything is labelled for screen readers and spoken aloud.
Private by design
Camera frames are analysed and discarded. No video is stored, no account is needed to try it.
Runs today
Working prototype: iPhone camera streamed to an AI server over Wi-Fi.
Next: a fully on-device app for iOS and Android.