Allow to specify level or layer for snapping

[Nakaner]

If routes start or end in multi-level stations, GraphHopper selects the closest edge. This can lead to wrong results.

Karlsruhe Durlacher Tor–Karlsruhe Europaplatz (routed via Gerwigstraße) vs Karlsruhe Durlacher Tor (U)–Karlsruhe Europaplatz (routed via direct tunnel)

Expected route:

Route with slightly different start location (about 3 m offset):

This issue exists for map matching as well if the request contains one point per stop only.

It cannot solved because the locations for stops provided by transport authorities are often of low precision, do not distinguish between multiple tracks/platform or OSM can be inaccurate (especially underground stations).

Possible solutions:

• Allow to specify a value of layer=* or level=* for snapping for each individual point of a route request.
• Currently snapping onto tunnels or bridges can be forbidden by snap_preventions. Snap preventions could be extended to specify preventions per point and by the "road_environment" default (no-tunnel-no-bridge-no-ferry edges).

[Edefritz]

Hi and many thanks for open sourcing this project, it is really helpful :)

I've stumbled upon a similar issue which seems to be related to multi-level stations. Whenever, a point is set on a multi-level station, the resulting route seems odd.

It seems you already know what the issue is, but in case you need more examples I figured I can attach my case here as well:

https://routing.openrailrouting.org/maps/?point=52.534766%2C13.195621&point=52.525165%2C13.369396&point=52.510696%2C13.435159&locale=en-US&elevation=false&profile=non_tgv&use_miles=false&layer=OSM%20Carto

The second waypoint sits on Berlin Hauptbahnhof. The routing engine always tries to approach the stop from the northern direction it seems, even though the optimal route would be west-east-bound.

Also, the route stops just at the border of the stop, probably that is what you mean with "Graphhopper selects the closest edge". If that point is removed, a "proper" route is returned which still goes through the problematic stop.

Anyways, if you can find a solution, I'd be interested as well :)

Thanks!