v1.2.3 · Android · offline by default
Navigation that routes around automated licence-plate readers — entirely on your phone. An ordinary route request — an address, a destination — with one extra constraint: prefer roads that no ALPR camera can actually see. Routing runs offline by default, so where you are and where you're going never leave the device.
0avoidance levels
120,838cameras mapped
0states, offline
v1.2.3latest APK
Trames (Latin): byways, side roads. · Debug-signed APK — sideload; no Play Store. Read the README.
Built on
🗺
how it works
A reader pointed at northbound traffic says nothing about the southbound carriageway. Treat it as a circle and the router detours around roads nobody is being read on. TRAMES models every camera as a cone pointing where it actually looks, and asks for the fastest path that crosses none of them.
1
Directional cones, not dots
Each camera becomes a wedge aimed the way it faces. A route that passes a reader on the far carriageway isn't penalised — it was never seen.
2
Baked into the graph
120,838 camera cones are compiled into GraphHopper's spatial index
at import as one area, alpr. A request just references it
— {"if": "in_alpr", "multiply_by": 0.05} — no geometry
attached, so the payload stays tiny.
3
Strength is a live knob
Because the penalty rides on each request, avoidance strength stays a per-route slider instead of being frozen into the graph. No forked routing engine required.
4
See what you're avoiding
The map draws the same cameras the router steers around — from the server's own snapshot, so the picture matches the route. Overpass is a fallback, not the only source.
out of the box
Avoidance is the whole point of the app, so it is no longer opt-in homework. A fresh install sets itself up on first launch — no engine to create, no URL to paste, no setting to find.
✅
First launch sets the driving profile to the offline engine at Strong avoidance. Download your state, enter where you're going, and the route already bends around camera fields of view — computed on the phone, with nothing sent anywhere. The only thing you touch is the destination.
🎚
Turn avoidance down, off, or up — the app seeds itself exactly once, so your choice is never overwritten on the next launch.
🚲
A camera is a camera whether you drive, ride or walk past it. All three profiles get the same five levels, and a level means the same thing in each.
🧩
Package org.soulstone.trames — a distinct app, not a
replacement. Keep your OsmAnd exactly as it is.
🛰
Point the engine at your own GraphHopper if you'd rather. Upgraders who configured one by hand keep it, custom strength and all.
🧭
TRAMES is a fork of OsmAnd, so turn-by-turn voice guidance, offline maps, lane hints and the rest are all there — the avoidance is one honest constraint layered onto a mature navigation app, not a demo.
avoidance strength
The levels aren't evenly spaced, and that's deliberate: measured on the continental graph, everything from off down to about a third leaves most routes unchanged — a penalty can't outweigh a highway. So the useful range is packed where it actually bites, and the effective strength shifts with how dense the cameras are around you.
OFF
Fastest route. No avoidance — TRAMES navigates like any other app.
LIGHT
Avoid readers only where the detour is nearly free.
MODERATE
A balanced trade — dodge most cameras for a modest cost.
STRONGdefault
Avoid readers wherever a reasonable alternative exists. Where TRAMES ships.
MAXIMUM
Avoid every camera it can find a way around, whatever the extra time.
the measurement study
We routed 17,580 real home–work commutes — drawn from Census LEHD LODES in proportion to how many workers actually make each trip, across 15 states — twice each: once normally, once avoiding the fields of view of 120,838 mapped ALPR installations.
17,580commutes routed, twice each
74.8%pass at least one reader
84.6%can route to zero exposure
+1.5 minmedian cost to do it (5.96%)
📈
A metropolitan grid always offers a parallel street a block over; a rural route often has one road and no alternative. So the burden of refusal falls hardest on the drivers with the fewest cameras to refuse. The full method, limitations, and the equity findings are in the paper. Read the paper (PDF) →
screenshots
Real routes over real ALPR data — the avoided cluster, the directional markers, and turn-by-turn guidance that already went the quiet way.
the honest part
Earlier versions routed through a server, which meant handing over every start and destination — the exact movement record this app exists to avoid. As of v1.2.0 that's inverted: the private path is the default and the network is opt-in. Here's the whole posture.
📴
Routes are computed on the device against maps you've downloaded, so your start, destination and timing never travel. Airplane mode changes nothing — avoidance keeps working. Avoiding cameras while streaming your itinerary to a host isn't privacy, so we stopped doing it.
🛑
Camera positions come from a 1.1 MB pack on the device. A camera query would carry a box around your current view — your location — so those requests are denied by default and only permitted if you deliberately switch a profile to online routing.
🌐
Prefer fresher camera data, or need an area you have no map for? Pick the online engine in navigation settings and it works as before — open, no account, no API key. Or point it at your own GraphHopper.
🧾
The client, the server, and the study are all public. Camera data is © OpenStreetMap contributors under ODbL, largely mapped by the DeFlock community.
🔑
The APK is signed with the public Android debug key. Anyone can build one that looks identical, so the signature proves nothing about who built it. Verify the SHA-256 on the release, or build from source.
get it
1 · download
2 · install
3 · get your maps
4 · just drive
Build from source, read the paper, or file issues: github.com/KaraZajac/TRAMES · the paper