Waymark

Inspiration

My grandpa lost most of his sight, and watching him get around showed me how much of a walk depends on things a white cane can't tell you. The missing curb ramp at the next corner. The sign post halfway down the block. The lifted slab you only find with your foot. He isn't alone. About 7 million Americans live with vision loss that glasses can't fix, and more than 1 million are blind (JAMA Ophthalmology, 2021). A cane knows what's under its tip. We wanted it to know what's 30 meters up the block.

What it does

Waymark gives a smart cane a memory of the sidewalk. An iPhone clipped to a white cane carries 3,405 sidewalk hazards from Seattle's U District, mapped by volunteers through UW's Project Sidewalk. They're stored right on the phone, so it works without signal. As you walk, Waymark says things like "Reported obstacle ahead" or "Reported missing curb ramp ahead" long before the cane or the phone's LiDAR could reach them.

It never talks over the cane's own safety warnings, and it always says "reported", never "detected". Then it learns. When the LiDAR or a sighted helper confirms or rejects a report, Waymark updates how confident it is and remembers it for next time. A map shows every hazard on the route, colored and shaped by how likely it is to still be there.

How we built it

Waymark is a native iOS app written in Swift and SwiftUI, built on existing LiDAR hardware (IOS devices). At DubHacks we built the sidewalk memory on top of it.

• Data: a Python pipeline pulls Project Sidewalk labels, cleans them up, and packs them into one offline file.

• Confidence: each report gets a score for how likely it is to still be there. It starts from how many volunteers agreed or disagreed with it, then goes up every time we confirm it and down every time we don't find it. Our own checks fade over time, so one from months ago counts less than one from today.

• When to speak: missing a real obstacle is a lot worse than a false alarm, so we set the bar based on that. For obstacles, Waymark speaks once it's at least about 17% sure the obstacle is there (1 in 6), because a miss costs five times more than a false alarm.

• Checking reports: LiDAR passes, curb drops, and a helper's "Confirm" or "Not there" taps all update the confidence. If the phone can't tell, nothing changes.

• Safety: Waymark speaks at the lowest priority, only when nothing else is talking, never right after a LiDAR warning, and at most once every 20 seconds.

• Testing: over 1,100 unit tests, a UI test suite, and a full replay of our route through the real app. In the replay, every planted report was spoken at the right time and never over a safety warning.

Challenges we ran into

The hardest part was finding reliable data about the real street. Project Sidewalk is a great start, but most of its labels come from street imagery taken in 2019 or earlier, so a lot of them might be out of date. The street data follows the middle of the road, not the sidewalk. And our demo stretch of The Ave only had one mapped obstacle. Finding obstacles that are actually there today, and proving it, meant building a way for the app to check reports while you walk. It also meant being honest about what a phone can't see yet. Curb ramps, for example, are too gentle for the LiDAR to confirm.

Accomplishments that we're proud of

• A cane that warns you about hazards before it reaches them, with no internet needed.
• It never interrupts the cane's own safety warnings, and we tested for exactly that.
• You can watch it learn. A report goes from 50% to 43% when a helper says it's gone, and it remembers after the app restarts.
• A color blind friendly map where each pin's shape and color both show how confident we are.
• It runs on a real iPhone, and the full route replay passes start to finish.

What we learned

Public data is only as good as the last time someone checked it. We learned to treat every map label as a claim to verify, not a fact. Building for blind users also changed how we think: say less, never interrupt a safety warning, word everything as a report, and test the logic before trusting it. The hardest part of an accessibility tool isn't the feature itself. It's making sure the feature never gets in the way.

What's next for Waymark

• More cities. Project Sidewalk runs in dozens of them, and our pipeline works for any city.
• Shared memory, so one person's walk can update everyone's map (opt in and anonymous).
• A gentle tap on the Apple Watch for reported hazards, for people who would rather feel it than hear it.

Built With

• accessibility
• computer-vision
• crowdsourcing
• html
• lidar
• machine-learning
• openstreetmap
• python
• swift