[Talk-GB] routing on the road network

[Frederik Ramm]

Hi,

On 05/16/2012 11:56 AM, Tim Pigden wrote:
> that there are no one-way streets leading to dead ends,

This is not common in OSM but I am not aware of anyone doing a network 
analysis that would fix such a problem.

> that you can't turn left off a flyover onto the road underneath

In OSM, turning is only possible at nodes, not where two roads happen do 
cross.

Some editors and existing QA tools will whine when you have a non-noded 
intersection between two roads and neither of them is marked a bridge or 
tunnel.

> we also assume that there is only one link connecting two points
> if there is only one physical road link,

Opinions as to what makes up a physical road link may differ, especially 
when pavements or cycle lanes are concerned. In some countries, people 
map pavements and cycle lanes as separate geometries (eg 
http://www.openstreetmap.org/?lat=64.138954&lon=-21.922888&zoom=18&layers=M).

> and that all intersections are
> proper intersections.

In the GIS world, a "proper" intersection is often one that is not 
traversed by any geometry, i.e. you will have roads beginning/ending 
there but not going through it. In OSM this is different; having a way 
traverse an intersection is normal and not considered an error.

> How suitable is OSM GB for routing, right now, with this level of
> detail? Do corresponding network analysis tools exist?

Tools exist but not in one central location. Some potential problems are 
caught by editors. Then there's OSM Inspector (tools.geofabrik.de/osmi) 
which has the Skobbler-sponsored "routing layer" that finds un-connected 
roads; keepright.at has something similar and also checks for suspect 
things like service roads branching off of motorways. Powerful 
project-osrm.org routing engine has the potential to be used as an 
analysis tool but doesn't produce any automated reports. The restriction 
analyzer at http://osm.virtuelle-loipe.de/restrictions/ specializes in 
debugging turn restrictions, and I'm sure there will be more tools I 
haven't mentioned!

> I did try about 3
> months ago to follow details for building a routable network using
> pgrouting and one of the programs in the chain  seemed to do thousands
> of fixups to the data.

Most likely the program broke up the roads into segments starting/ending 
at intersections, something that might be a required preprocessing step 
but nobody would suggest doing that in OSM. OSM is pretty suitable for 
routing in my opinion.

> If we did such an analysis (which would be quite a big investment) and
> came to the conclusion there were 10s of thousands of errors - which
> seems to be entirely possible - would there be any appetite in the
> community for fixing them? Certainly such an endeavour would be way
> beyond our budget.

Appetite, yes, but you can also easily chase people away if your system 
detects too many things where people don't think it's a bug at all, so 
some tuning might be necessary. One of the weaknesses of most of the 
existing systems (with the exception of checks in the editor) is that 
once you fixed the bug, it might take hours or even days for the bug to 
be properly closed. A system that allows users to hit some kind of 
"re-check this, I've just fixed it" button would surely create more 
enthusiasm for "cleaning up your area" than most of what we currently have.

Bye
Frederik

-- 
Frederik Ramm  ##  eMail frederik at remote.org  ##  N49°00'09" E008°23'33"