The gap is between the block and the cane tip

What already works

A white cane is an excellent instrument. It gives you the surface under and just ahead of you with zero latency, no battery, and total reliability. A guide dog does more and does it faster. Neither is being replaced here, and any product that talks about replacing them should be treated with suspicion.

At the other end, turn-by-turn navigation is genuinely good at getting you to a block. GPS puts you within somewhere around five to ten metres and then stops having opinions.

What does not

Between the cane tip and the block there is a band — roughly one to twenty metres ahead — that neither tool addresses. In that band live most of the things that make a new route hard:

• Sidewalk that narrows, ends, or is closed for construction with the detour signed visually.
• Scaffolding poles, sandwich boards, e-scooters, and bins that were not there last week.
• A curb that is where the map says the corner is, but four metres further along than you expected.
• A crossing whose location, orientation, and signal you have to confirm from scratch.

None of this is dangerous to a trained traveler. All of it costs attention, and attention is finite. The cumulative effect is that unfamiliar routes get planned around, delayed, or not taken. That is the cost we are actually trying to reduce.

Why now

This band was not addressable on a phone until recently. Three things changed:

1. Depth on a handset. LiDAR on iPhone Pro models gives measured distance, not estimated distance. A curb height of 0.14 m is a measurement, and measurements are what let guidance be specific instead of hedged.
2. Persistent scene understanding. ARKit keeps plane geometry across short look-aways, so a wall does not cease to exist because you turned your head. Guidance can be continuous rather than frame-by-frame.
3. On-device inference that fits the power budget. Terrain and obstacle classification can run locally, continuously, during a real walk — which matters both for latency and for privacy.

Why the restraint

The same capabilities make it very easy to build something that demonstrates beautifully and fails badly. A system that names every object it sees looks impressive in a two-minute video and is unusable on a twenty-minute walk. A system that says “safe to cross” converts a probabilistic estimate into an instruction that a person will act on with their body.

So the hard design work here is not detection. It is deciding what not to say. Most of what NaviGuider perceives never reaches the traveler, and the parts that do are filtered by whether they change what you should do in the next few seconds.

What we owe the people who use it

An assistive product asks for trust before it has earned it, because the person evaluating it cannot independently verify most of what it claims. The only honest response is to be specific about limits, to validate with blind travelers and orientation-and-mobility specialists before claiming readiness, and to publish what did not work alongside what did.

That is what the research program is for, and why NaviGuider is not on the App Store yet.