Audio and haptics
The interface a driver already knows
Early versions of NaviGuider spoke everything. Field walks killed that fast. What replaced it is a parking sensor for walking — and it was built from the first line to move off the ear and onto the foot.
Three channels, and what each is allowed to carry
Pulse — continuous
Where the walkable space is, and how close the nearest hazard is
A pattern lands in about a tenth of a second. Speech takes seconds to say the same thing, and by then the distance has changed.
Speech — rationed
Names of things, route instructions like “turn left in fifty feet,” and urgent stops
Nouns and directions cannot be encoded as a pattern. Everything else can, so everything else is.
Silence — the default
Nothing, on purpose
The most important input on an unfamiliar sidewalk is the street. A hazard is announced once per approach and re-announced only after you close 0.6 m or ten seconds pass — four for urgent classes.
Side, rate, tone
Three independent channels, each carrying one thing. The right-hand column is the reason the whole scheme exists: every channel has a hardware counterpart waiting for it.
Paper, table 2
| Channel | Encodes | Audio today | Haptic shoe |
|---|---|---|---|
| Side (stereo) | Action | Left ear = move left; both = ahead | Which foot |
| Rate | Distance | 1.0 s interval at 2.5 m → 0.12 s at 0.5 m; urgent 0.10 s | Pulse rate |
| Tone | Category | Drop-off 220 Hz · obstacle 440 Hz · stairs or step 590 Hz · small step up 880 Hz, down 660 Hz | Motor site |
| Chime | All clear | A single 1318 Hz chime, once | Single soft tap |
The pulse interval is a clamped linear map of hazard distance, from a one-second interval at 2.5 m down to 0.12 s at half a metre, so perceived urgency grows continuously as the distance closes. Urgent classes sit at 0.10 s. Path clear requires two full seconds of hazard silence before the chime.
Why shoes
The honest problem with pulse audio is stated in the paper’s own limitations: it occupies a channel blind travelers rely on. Hearing is how you find the crosswalk, track traffic, and place yourself on a street. Spending it on a proximity signal is a real cost, not a detail.
Vibration does not compete for that channel. The published evidence is encouraging on both counts: a randomized clinical trial of a chest-mounted camera with vibrating wristbands found obstacle contacts fell 37% in home use by 31 blind and visually impaired participants, and separate work shows blind users discriminate vibration patterns at 65 to 90% recognition across ten patterns, with haptic coding preferred over speech in noisy environments and over sonification for privacy and audibility.
Feet are the specific choice because the mapping is already spatial. Left foot means move left. There is nothing to learn and nothing to translate — the cue arrives where the action happens.
Why the audio came first
Not as a placeholder to be thrown away. The pulse encoding is the hardware specification, written in software so it can be walked, tuned, and tested on real sidewalks before anyone builds a shoe. Side, rate, and tone map one-to-one onto which foot, pulse rate, and motor site. When the transducer changes, the code and the traveler’s mental model do not.
That also means the two can be evaluated against each other. The planned participant protocol allows audio-only, haptic-only, and combined conditions — but only after the shoe hardware has been independently verified.
What speech keeps
Audio is not being retired. It keeps exactly the jobs a pattern cannot do:
- Immediate danger. An urgent stop is spoken, because a stop has to be unambiguous and instant.
- Route instructions.“Turn left in fifty feet” is a direction with a distance and a noun in it. No pulse carries that.
- Names. When you ask what is around you, the answer is words.
- Refusals. When the app cannot resolve risk, it says so, in a sentence.
One voice arbitrates all of it. Candidate utterances are ranked and exactly one wins; the rest are dropped rather than queued, because a queue is how a guidance system starts talking over a street.
An open audit finding
Worth stating because it is unresolved: a near minor obstacle could end command evaluation before a farther, more severe drop was considered, because the arbitration returned nearest-first. The fix — evaluate imminent severity before convenience or route commands — is an open item in the paper’s failure catalog, not a shipped change.
E0 → E2The pulse encoding is implemented and test-backed. The shoe mapping is a design, unbuilt and unevaluated. The 37% and 65–90% figures are other researchers’ results with other hardware, cited as the reason to try — not as NaviGuider’s own outcomes.
The full encoding table and the announcement governor are in the paper.