HH - C++: On limited set of maps routing leaves motorway where it should not

casta commented 4 weeks ago

Describe the routing engine used (required)

[x] OsmAnd's in-app offline routing
[ ] Any online routing provider (YOURS, openrouteservice, OSRM, etc.)

Describe your routing Profile (required)

Using car routing with default parameters, no restrictions enabled, with HH C++ routing engine (which is the default on 4.7.17).

Describe your start and end points (required)

Please find attached 3 GPX files generated with OsmAnd routing: osmand-routing-bug-gpx.zip

HH C++.gpx is the faulty routing computation, using default HH C++ engine
A phase 2.gpx is the same routing computation (same start and end, same routing parameters) but using A* phase 2 engine. This one is fine
HH C++ crop.gpx is a routing using HH C++ engine again, but I moved the starting point a bit to somewhere farther in the previous route. It does not exhibits the same fault as the first one.

Describe the actual route (required)

To make it obvious, see this screeshot:

The HH C++.gpxroute leaves the motorway at some point before entering it back later in the route. It should not, as it’s not optimal.

Describe the expected route (required)

The A phase 2.gpx is the expected route. “Cropping” a bit the route at the beginning make HH C++ (HH C++ crop.gpx) fix it’s computation too.

Describe what Maps you used (online or offline) (required)

[x] Offline maps offered within the OsmAnd app for download. 01/05/24 edition
[ ] Online (tile / raster) maps

Your Environment (required)

OsmAnd Version: 4.7.17 fdroid version
Android/iOS version: Android 14 AP1A.240505.005 (CalyxOS)
Device model: Pixel 7a

Anything else? (optional)

No response

yuriiurshuliak commented 4 weeks ago

The bug was not reproduced. I tested the path construction using all routing algorithms, and it consistently routed through the motorway. You can verify it using our Web Version here: link

Please try updating your map or resetting your profile and check again.

casta commented 4 weeks ago

Ok after your comment I tested again and I have some new information. Here are screenshots with large zoom enough to see the downloaded maps in green.

I have this issue only if have the strictly minimal needed maps that covers the morotway (“Creuse” and “Corrèze”). The whole road is covered by the downloaded maps:

But if I download the whole “Nouvelle Acquitaine” region, I cannot reproduce anymore:

Does this HH algorithm need more data around the targeted road to do the routing ?

yuriiurshuliak commented 3 weeks ago

The bug was not reproduced. I attempted to build the navigation excluding the "Nouvelle Acquitaine" region and experimented with routing types HH x Java and HH x C++. In my scenario, the navigation routed across the motorway.

OsmAnd~ 4.8.0#2440m, released: 2024-06-05

casta commented 3 weeks ago

Well, it seems that your starting and endpoints are not exactly the same as in my GPX test case. As you can see, to reproduce, I have to set specific starting and ending point (see HH C++.gpx in my opening post), as soon as I change the starting point, the route is “fixed” (HH C++ crop.gpx).

But anyway, if you think that this behaviour is normal, feel free to close.

vshcherb commented 3 weeks ago

It's tricky to reproduce and hard to test as it's very long route to trace each segment, I will leave it open to find other cases where we would fix it easier.

osmandapp / OsmAnd