[OSRM-talk] "nearest suitable road segment"

Jürgen Barthel osrm at checkin.com
Mon Aug 22 23:41:56 UTC 2016

On Aug 17, 2016, at 2:00 PM, Florian Lohoff wrote:
> [...]

(Florian Lohoff, I agree)

I can confirm the same problem, though I think this is a mix between 
OSM, Nominatim and OSRM. We route from municipalities to airports. Where 
that was available, we used the "Admin Center node" set in OSM, but even 
that is often NOT a road address. We updated all airports to a street 
address to avoid such conflicts on that side. We now have the task at 
hand to define a "starting point" for any given municipality on file. A 
major problem happened i.e. in the Alps, where the "centroid" is on the 
mountain top, OSRM taking mountain ways at extreme slow speed adding 
some 20-40 unnecessary minutes, comparing to the neighboring town having 
an admin center defined.

And we can't blame it only to Nominatim: We use geodata as origin and 
destination for the route as we found Naming to be extremely misleading. 
I cannot recommend that at all. We had major errors, where the way you 
entered the data had major impact on the accuracy of the identified 
geopoint. We had i.e. Münster (Lech), naming according to Wikipedia and 
the city, OSM relation/962251, with a population of 2000 mistaken with 
Münster (Westfalen). A major city there. And a minor difference in drive 
time by 570 km or 5:30h and a nice "hole" in our map in Bavaria until we 
solved that... Other examples where islands with "exclaves" where the 
centroid simply was in the water and the drive time started on the wrong 
(uninhabited) island...

So yes, I think it would be wise to improve OSRM on this. It might be 
needed to improve the OSM data, but moving streets or ways cannot be the 
answer. If I understood those comments right we talk about "adjusting" 
reality here for the sake of a route request...?

I'm also really concerned about that "indefinite distance" like on my 
Helgoland problem. Found the same true on other islands. That way, OSRM 
returns something where an error would make more sense! We can't solve 
this in a day or with a quick fix, but this is a quality problem that 
may have repercussions all over the planet(.osm). A general improvement 
(a.k.a. evolution). Issue is (in my opinion) that there should be a 
"best practice". And that might need the possibility to have an optional 
"routing reference point" (whatever you call it) on places, ways or 
points... If it's not there yet anyway?

How to proceed? I'm too new to say.

Cheers - Juergen

P.S.: Have been out sick. Sorry for the delayed response.

More information about the OSRM-talk mailing list