Inspiration
Everyone at UCC has been there: first week, new building, already late, walking up and down corridors looking for WGB 106. GPS stops working the moment you step inside, wall maps don’t tell you which way you’re facing, and asking a stranger for the third time gets old. We wanted indoor directions that work as easily as Google Maps does outdoors, without a university having to install beacons or special hardware.
What it does
WayIn is indoor navigation that runs in your phone’s browser.
Find where you are from a room sign. Point your camera at any door sign. Claude reads it and works out where you’re standing in the building. Follow a line on the real floor. A red AR line is drawn on the floor in front of you, from where you are to your room. It goes up stairs and into lifts, with turn-by-turn voice prompts and vibration. Pick your kind of route. Choose Fastest, Quiet (avoids busy spots like the café and atrium) or Step-free (lifts only, for wheelchairs and buggies). Map a building by walking it. In Map mode you walk the corridors and tag rooms as you pass, so no floor plans or CAD files are needed. It gets better with use. The camera scans open floor and obstacles, so routes bend around things in the way, and paths people walk off the map are learned as shortcuts. Plan and side views. A 2D plan and a side view show the route across floors before you set off. Works on iPhone too. Where AR isn’t available, it falls back to a camera view with a compass arrow and step counting. How we built it One self-contained web page. There’s no app store install: open a link and go. WebXR and three.js. These draw the route on the real floor. AR hit-testing finds the floor and obstacles, which go into a grid of 25 cm squares. Claude’s vision models read the signs. Claude Sonnet handles a sign you scan on purpose, and Claude Haiku does the cheap background reads while you walk. Routing in two layers. Dijkstra’s algorithm works over a graph of corridors, stairs and lifts, with extra cost for busy spots and stairs removed for step-free routes. Then A* plans through the scanned floor grid to steer around obstacles. Lining the map up with the real world. Room signs fix your position, and a second sign fixes your direction. The map also learns which way is magnetic north while you map it, so the compass can catch a line pointing the wrong way. Your walking direction snaps the map onto the corridor you’re actually in. Challenges we ran into The line vanishing. Re-planning a route cleared temporary points that the line on screen still depended on, and the whole line disappeared. We rebuilt how routes are swapped, so the old line stays until a new one is ready. The line pointing behind you. One sign tells you where you are but not which way you’re facing. On a straight corridor, walking backwards fits the map just as well as walking forwards. We solved it by having the map learn magnetic north while it’s being mapped. Position drift. When a sign was read from a few metres away, everything was shifted by that much. We added re-centring onto the corridor as you walk, plus tap-to-correct on the plan. Bad scans getting saved. A misaligned session saved fake walls and fake shortcuts into the map. Now the map only learns once the line-up is confirmed. Depth sensing froze some phones. We made it optional, added a watchdog that recovers from a frozen AR session, and moved obstacle detection onto standard hit-testing. Accomplishments that we’re proud of It runs entirely in the browser, with no app to install and no hardware in the building. Anyone can map a building just by walking through it. Accessibility is built in: step-free and quiet routes are a first-class choice, not an afterthought. We built a simulation that maps a virtual building and navigates it in random orientations with a noisy compass. It reproduced our “line behind you” bug and confirmed the fix across 150+ random cases. What we learned Indoor positioning is mostly about combining imperfect clues. Signs, compass, walking direction and the camera are each wrong in different ways, and they only become reliable together. AR on phones varies a lot between devices, so every feature needs a fallback. Building a simulator early would have saved hours of walking up and down corridors to test. What’s next for WayIn Map all of UCC and share maps between users, so one person mapping a building helps everyone. Live busyness, so Quiet routes react to crowds as they happen. Accessibility extras: high-contrast mode, haptic-only guidance and screen-reader support. Integration with timetables, so it already knows where your next lecture is. Pilots in hospitals and shopping centres, where getting lost costs more than a late arrival.
