[OSM-talk] Divided/Non-Divided Intersection

[andrzej zaborowski]

Hi,
sorry to resurrect an old thread, but I haven't seen it answered in the end.

On 11 July 2010 10:23, Maarten Deen <mdeen at xs4all.nl> wrote:
> John Smith wrote:
>>
>> On 11 July 2010 06:43, Chris Dombroski <cdombroski+osm at icanttype.org>
>> wrote:
>>>
>>> I ask because I think this is the cause of stupid GPS directions at times
>>> "make a left, followed by a slight right"
>>
>> Isn't that a problem with the routing software, not the data?
>
> Not IMHO.
>
> If you have a layout like this (use a fixed-width font):
>
>     | |
> A----+-+
>     | +----B
> C----+-+
>     | |
>
> And you want to go from B to A, why would routing software say "go straight
> on" and not "go right, then go left"?

My opinion is that it is a routing software issue after all.  When a
road segment is just about 5m long, the software should just skip it
in the driving directions and look at the angle between the A and the
B roads.  It can go as far as doing what Alan said, i.e. join all of
the nodes in a junction (a concentration of nodes where roads meet)
into a single node, possibly in the preprocessing phase.  I don't
agree that this should be done in OSM data, it will prevent more
clever routing direction being given when more clever routing software
is written (e.g. software telling your car where to stop to wait for
green light, which lane to take etc).

The lengths need to be chosen carefully because in e.g. pedestrian
routing, 5m may be significant (say you're looking for an exit from a
maze).  A car can't even make turns that tight, and you're interested
in the bigger picture when you reach a junction, the routing is not
clever enough to get you through a junction anyway.

>
> And option is to map it like this:
>     | |
> A----+ |
>     |\|
>     | +----B
>     |/|
> C----+ |
>     | |

Cheers