[Routing] Bad routing: U turns for highway links?

[Apollinaris Schoell]

On 17 Jul 2010, at 11:36 , Marcus Wolschon wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Am 17.07.2010 19:20, schrieb Apollinaris Schoell:
>> 
>> In this example it seems just too easy for a good router to do the
>> right thing without hard facts.
> 
> 
> How?
> Please post the algorithm you are thinking about.
> 

typical intersection like this one requires 2 turns

70-110 degree from main road to ramp and 5-20 degree from ramp to trunk

the cloudmade route choose 80degree + 160 degree turn when it was possible with nearly exact same length to do  110 degree + 5 degree 
I think something like a exponential cost increase for the angle of a turn will result in 90% of the unclear situations 


>> This can be done by the router or some intelligent preprocessing
>> if compute time is critical as an example mkgmap does these type of
>> preprocessing to reduce the number of unnecessary announcements and
>> creates the routing graph accordingly.
> 
> Details?

I haven't done any of this or looked into the code. Mark explained a bit in some postings in the mkgmap dev list.
just to be clear. this is not for the routing itself just for the announcements.  Don't announce if the angle is small and roads continue with similar tags. the routing itself is done by the Garmin device

> 
>> 
>> We might reach the state where all rouads are mapped with lanes,
>> yellow lines, all turn restrictions …
> 
> Actually, over here we have cities where all streets are there and
> people start mapping
> single trees and individual lamp-posts.
> 

you are so lucky! in US there is lot of imported data of questionable quality. But even with plain tiger data routing is possible. there is a program for that. but I have never tested how good it is. so it is possible to fix things on the router side