Navaid relation?

classic Classic list List threaded Threaded
33 messages Options
12
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Florian Lohoff-2
On Wed, May 22, 2019 at 12:37:21PM +0200, Mateusz Konieczny wrote:
> > Again a footway between the house and a road will NOT help for
> > car navigation because for cars a footway is NOT a routable
> > part of the graph.
> >
> Car navigation may use footway data to select best dropoff point.
>
> In exactly the same way as you proposed with navaid relation,
> but without adding subjective data.

Look at my school example - would still be broken. Yes - there is a
footway - but not the original parking lot for the school.

Different street. ~1km detour. No parking at the selected spot - Private
property.

You are trying to fix an algorithm with new assumptions which break in
other aspects. You need to have a way to EXPLICITLY define a location
where to navigate to.

Flo
--
Florian Lohoff                                                 [hidden email]
        UTF-8 Test: The ­čÉł ran after a ­čÉü, but the ­čÉü ran away

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging

signature.asc (849 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Mateusz Konieczny-3
In reply to this post by Florian Lohoff-2



22 May 2019, 09:43 by [hidden email]:

Hola Mateusz,

On Wed, May 22, 2019 at 12:26:01AM +0200, Mateusz Konieczny wrote:
21 May 2019, 23:46 by [hidden email]:

> - Houses which are routeable by road a but are near road b or vice
> versa.
>
> Adding more roads aka service/driveway does not necessary make it more
> deterministic.
>
Can you give example of residential building with fully mapped roads, footways
and obstacles where well written router will fail?

- Baumstra├če 43a, G├╝tersloh, Germany
It does not have a connection to Baumstra├če but to
Hermann-Vogelsang-Stra├če.

It still will be routed through Baumstra├če and the driveway to
Baumstra├če 45a

IMHO unfixable without bending geometry
footway access, driveway. Fences are also unmapped.

I am not sure whatever we have smart routers, but here even human would fail
due to lack of data.

I added driveway based on aerial image but it is unlikely to be enough.

- Dalbker Stra├če 40a, Oerlinghausen, Germany
Dalbker Stra├če 44a, Oerlinghausen, Germany
Fences/hedges/whatever barrier is there is missing though
really smart router (that is using footway at start and the end)
would work correctly here.

Have you checked whatever this improvement is suggested for routers with public
bug trackers?
moving it further to
Dalkbker Stra├če within the outline of the Building.
within the outline? In case it is acceptable


_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Florian Lohoff-2
In reply to this post by Mateusz Konieczny-3

Hi

On Wed, May 22, 2019 at 12:42:45PM +0200, Mateusz Konieczny wrote:

> 22 May 2019, 09:53 by [hidden email]:
> > Hi Marc,
> > On Tue, May 21, 2019 at 10:38:23PM +0000, marc marc wrote:
> >> > What is the expectation to get navigated to when selecting a park?
> >> there is no such thing as "a single point that makes everyone agree"
> > Yes there is - there has to be an explicit location you will ne navigated
> > to for a certain feature.
> >
> This is blatantly untrue. Depending on location you will prefer to be routed
> to different entrances and it is not considering different modes of transport.
How can i select the entrance on OSRM/openstreetmap website, MAPS.ME,
OsmAND the entrance? You cant.

And where is an entrance to a Golf Course? And when navigating by car to
a mall i dont want the entrance - i want the Car park. Can i select
that from a drop down when selecting a Mall to navigate to?

Mapping OSM Objects to navigational locations is implicit right now and
depends on the algorithm used to find the nearest location on the
routable network for the specified mode of transportation. So i bet
i can generate a situation where OSMAnd, OSRM and MAPS.ME will bring
you to different spots more than 5km apart. Its just a matter of
constructing the right geometries with the tags one or the other
navigational preprocessing takes into account. You cant specify
the right location explicitly - so you need to rely on implicit geometry
processing by algorithms. And then you still have 2-3% of broken
destinations.

> > Nope - there isnt enough information. Its all just implicit and works
> > for 95% of the cases. It breaks horrible in others and we fake
> > geometries to fix it, blame the application, invent tags to guide the
> > nav/routing which only fit half of the object and only half of the
> > apps support. In all cases the user is in trouble.
> >
> Please give a specific example.

See my School example. Its a pretty nice fuck up. You get "near" the
School but in a oneway maze 1km away from the parking lot. And yes -
there is a footway. Everything is mapped as it is. Still - breaks.

https://lists.openstreetmap.org/pipermail/tagging/2019-May/045416.html
https://lists.openstreetmap.org/pipermail/tagging/2019-May/045423.html

And dont troll me with "add a footway" - A footway is not in the
routable graph for cars and will not even be in the database when
looking for the nearest point.

> If navigation is simply doing nearest road point on matching then it requires change to both
> - properly use footway data
> - use your proposed relation

footway != car

And it wont solve the issue. See the school example. There is a footway
and it will prefer the location it does not most likely. Still broken.

nearest road point will only be on roads for THAT mode of
transportation.

> I see no reason for preferring second solution.

Flo
--
Florian Lohoff                                                 [hidden email]
        UTF-8 Test: The ­čÉł ran after a ­čÉü, but the ­čÉü ran away

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging

signature.asc (849 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Mateusz Konieczny-3
In reply to this post by Florian Lohoff-2



22 May 2019, 12:49 by [hidden email]:
On Wed, May 22, 2019 at 12:37:21PM +0200, Mateusz Konieczny wrote:
> Again a footway between the house and a road will NOT help for
> car navigation because for cars a footway is NOT a routable
> part of the graph.
>
Car navigation may use footway data to select best dropoff point.

In exactly the same way as you proposed with navaid relation,
but without adding subjective data.

Look at my school example - would still be broken. Yes - there is a
footway - but not the original parking lot for the school.

Different street. ~1km detour. No parking at the selected spot - Private
property.

You are trying to fix an algorithm with new assumptions which break in
other aspects. You need to have a way to EXPLICITLY define a location
where to navigate to.
I noticed no link to that case, but here routers should direct car to mapped
parking within school area (it is a public parking, right?).

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Florian Lohoff-2
In reply to this post by Mateusz Konieczny-3
On Wed, May 22, 2019 at 12:55:44PM +0200, Mateusz Konieczny wrote:

> > - Baumstra├če 43a, G├╝tersloh, Germany
> >  It does not have a connection to Baumstra├če but to
> >  Hermann-Vogelsang-Stra├če.
> >
> >  It still will be routed through Baumstra├če and the driveway to
> >  Baumstra├če 45a
> >
> >  IMHO unfixable without bending geometry
> >
> https://www.openstreetmap.org/way/273023376 <https://www.openstreetmap.org/way/273023376> has no mapped entrance,
> footway access, driveway. Fences are also unmapped.
Footway -> not for car
fences -> not in graph for navigation - not connected to graph-> purely
cosmetic object.

43a has no footway/driveway. The carport is directly at the side of the
road.

> I am not sure whatever we have smart routers, but here even human would fail
> due to lack of data.
>
> I added driveway based on aerial image but it is unlikely to be enough.

Where is it ? 43a does not have a driveway.

> > - Dalbker Stra├če 40a, Oerlinghausen, Germany
> >  Dalbker Stra├če 44a, Oerlinghausen, Germany
> >
> Fences/hedges/whatever barrier is there is missing though
> really smart router (that is using footway at start and the end)
> would work correctly here.
>
> Have you checked whatever this improvement is suggested for routers with public
> bug trackers?

Okay - first you tell me - its solvable - then you tell me data is
missing, now you tell me the software is broken.

And there is a footway - its directly in front of the door. Still - you
will get routed to a differen street. The footway is NOT in the graph
for cars.

All objects you put into this argument do not have any influence on any
routing app/software mentioned in this thread before.

Flo
--
Florian Lohoff                                                 [hidden email]
        UTF-8 Test: The ­čÉł ran after a ­čÉü, but the ­čÉü ran away

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging

signature.asc (849 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Florian Lohoff-2
In reply to this post by Mateusz Konieczny-3
On Wed, May 22, 2019 at 01:01:20PM +0200, Mateusz Konieczny wrote:

>
> 22 May 2019, 12:49 by [hidden email]:
>
> > On Wed, May 22, 2019 at 12:37:21PM +0200, Mateusz Konieczny wrote:
> >
> >> > Again a footway between the house and a road will NOT help for
> >> > car navigation because for cars a footway is NOT a routable
> >> > part of the graph.
> >> >
> >> Car navigation may use footway data to select best dropoff point.
> >>
> >> In exactly the same way as you proposed with navaid relation,
> >> but without adding subjective data.
> >>
> >
> > Look at my school example - would still be broken. Yes - there is a
> > footway - but not the original parking lot for the school.
> >
> > Different street. ~1km detour. No parking at the selected spot - Private
> > property.
> >
> > You are trying to fix an algorithm with new assumptions which break in
> > other aspects. You need to have a way to EXPLICITLY define a location
> > where to navigate to.
> >
> I noticed no link to that case, but here routers should direct car to mapped
> parking within school area (it is a public parking, right?).
Search for it - When i add a link i add a specific location - I did this
intentionally - Because with the search you map object -> specific
location and then you can query OSRM again for routing.

It is the official parking lot for the school. Its vis a vis to the
entrance.

Flo
--
Florian Lohoff                                                 [hidden email]
        UTF-8 Test: The ­čÉł ran after a ­čÉü, but the ­čÉü ran away

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging

signature.asc (849 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Mateusz Konieczny-3
In reply to this post by Florian Lohoff-2



22 May 2019, 13:00 by [hidden email]:

Hi

On Wed, May 22, 2019 at 12:42:45PM +0200, Mateusz Konieczny wrote:
22 May 2019, 09:53 by [hidden email]:
> Hi Marc,
> On Tue, May 21, 2019 at 10:38:23PM +0000, marc marc wrote:
>> > What is the expectation to get navigated to when selecting a park?
>> there is no such thing as "a single point that makes everyone agree"
> Yes there is - there has to be an explicit location you will ne navigated
> to for a certain feature.
>
This is blatantly untrue. Depending on location you will prefer to be routed
to different entrances and it is not considering different modes of transport.

How can i select the entrance on OSRM/openstreetmap website, MAPS.ME,
OsmAND the entrance? You cant.
Why would you want to manually select entrance?

See for example router selecting appropriate entrance based on location:


http://brouter.de/brouter-web/#map=16/50.0645/19.9179/standard&lonlats=19.918041,50.059245;19.917483,50.062799&profile=shortest
And when navigating by car to
a mall i dont want the entrance - i want the Car park. Can i select
that from a drop down when selecting a Mall to navigate to?
It should be automatically preferred for car routing. I admit that I have no router to
recommend, as I am not using car. But have you tried requesting improvements
for your navigation (assuming that it has a a public bug tracker and it was not
requested already)?
Mapping OSM Objects to navigational locations is implicit right now and
depends on the algorithm used to find the nearest location on the
routable network for the specified mode of transportation. So i bet
i can generate a situation where OSMAnd, OSRM and MAPS.ME will bring
you to different spots more than 5km apart.
What is wrong with that?
And dont troll me with "add a footway" - A footway is not in the
routable graph for cars and will not even be in the database when
looking for the nearest point.
Router can use footway data to select good dropoff point.
And other info like parkings.


_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Mateusz Konieczny-3
In reply to this post by Florian Lohoff-2



22 May 2019, 13:03 by [hidden email]:
On Wed, May 22, 2019 at 12:55:44PM +0200, Mateusz Konieczny wrote:
> - Baumstra├če 43a, G├╝tersloh, Germany
> It does not have a connection to Baumstra├če but to
> Hermann-Vogelsang-Stra├če.
>
> It still will be routed through Baumstra├če and the driveway to
> Baumstra├če 45a
>
> IMHO unfixable without bending geometry
>
https://www.openstreetmap.org/way/273023376 <https://www.openstreetmap.org/way/273023376> has no mapped entrance,
footway access, driveway. Fences are also unmapped.

Footway -> not for car
fences -> not in graph for navigation - not connected to graph-> purely
cosmetic object.
It is possible for routers to use fence/footway data to select good dropoff location.

I am not claiming that there are routers doing this now, but it is possible to do
and better solution that manual routing with relations.

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

Colin Smale
In reply to this post by Florian Lohoff-2

Navigation software needs to move on. Instead of mapping a destination POI to a single point in every case, it needs to handle a list of points. Each point may have filters or qualifiers, such as opening hours or mode of transport; this can lead to some of the points being disqualified. The routing needs to calculate a route to each remaining point and choose the best one using the routing criteria (shortest, fastest etc). The multiple route calculations don't always need to restart from the departure point - I'm sure a heuristic to limit the recalculation to the last X km would be fine in most cases.

Then in the OSM data we need a way of indicating that multiple points are initially equivalent, serving the same POI. For example, multiple car parks serving the same public park. This can be purely geometric, by virtue of the car parks being enclosed by the parks outer perimeter, or by some kind of association relation. Reducing a car park polygon to a candidate point for the above multiple-routing case can be based on a node with entrance=* or the intersection with an accessible highway. Where a car park has multiple entrances, add them all into the candidate point list as above.

Regarding the use of footways for vehicle navigation, there are plenty of multimodal public transport planners out there, which know about trains, buses, walking etc. They work much more from point to point, not limited to just the vehicle segment of the journey. A car router/navigator can do the same, can't it? It will of course rely on the data it is given....

C.


_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

marc marc
In reply to this post by Florian Lohoff-2
Le 22.05.19 ├á 12:06, Florian Lohoff a ├ęcrit┬á:
>> you mean https://www.openstreetmap.org/way/273023376 ?
>> it's a good example of missing datas.
>> no entrance, no way between the entrance and the public network.
>> I feel that the relation type=navaids should be called type=missingway
>
> Again a footway between the house and a road will NOT help for
> car navigation because for cars a footway is NOT a routable
> part of the graph.

again multimodal app DO it (for e.g. switch footway with
access=destination in a BRouter car profile, you may also add a weight
despite i didn't known how to add a big weight for missing way)

when datas are available, an application can choose to prefer the most
complete routing to reatch a entrance even if a part is done by foot
compared to a closer routing by car but for witch the routing till the
destination is unknown and can therefore cross a barrier or anything
that invalidates the routing.
it's a matter of weight between the footway and the "no-way" and/or a
test to avoid a routing between the last routed-by-car node and the
destination through a barrier

so first :
- add missing datas that have already an osm tag
- have an app that use those datas
MAYBE AFTER that some additional datas are needed,
but without using currently usable infos, our exemple
get the reply "use eisting datas"
_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

dieterdreist
In reply to this post by Mateusz Konieczny-3


sent from a phone

On 22. May 2019, at 12:42, Mateusz Konieczny <[hidden email]> wrote:

If navigation is simply doing nearest road point on matching then it requires change to both
- properly use footway data
- use your proposed relation

I see no reason for preferring second solution.


check the Rome Fiumicino airport, Nominatim uses a cycleway to describe the address and routes kilometers away from the entrances:

maps.me for example routes to the entrances (not sure how they do it, whether they use external data as well or use a different mechanism to reduce the area to a point for navigation)

Cheers, Martin 

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

dieterdreist
In reply to this post by Florian Lohoff-2


sent from a phone

> On 22. May 2019, at 12:49, Florian Lohoff <[hidden email]> wrote:
>
> You need to have a way to EXPLICITLY define a location
> where to navigate to.


this is ÔÇťentranceÔÇŁ/barrier=gate, or not?

Cheers, Martin
_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
Reply | Threaded
Open this post in threaded view
|

Re: Navaid relation?

dieterdreist
In reply to this post by Mateusz Konieczny-3
Am Mi., 22. Mai 2019 um 13:12 Uhr schrieb Mateusz Konieczny <[hidden email]>:
How can i select the entrance on OSRM/openstreetmap website, MAPS.ME,
OsmAND the entrance? You cant.
Why would you want to manually select entrance?


because you might want to in order to state where you want to go? Usually people do this via addresses. Imagine a huge shopping mall, or a different huge structure with several entrances (train station, theme park, nature reserve, ...). For a shopping mall, you may want to go just to one store, so the parking should be close (for pedestrians, not as the crow flies) to this store, for example.

And when navigating by car to
a mall i dont want the entrance - i want the Car park. Can i select
that from a drop down when selecting a Mall to navigate to?
It should be automatically preferred for car routing.


this is more complex than just the closest parking, your intentions may have an impact on the kind of car park, e.g. if you go to the airport to drop your friend off, you will typically want to drive in front of the entrance, but if you are flying yourself and come by car, you will park on a parking farer away from the entrances (long term parking), and if you are going to pick someone up, you will go to the car park for short term parking. (I wrote "airport", but it is the same situation with different terminals on bigger airports).


 
And dont troll me with "add a footway" - A footway is not in the
routable graph for cars and will not even be in the database when
looking for the nearest point.
Router can use footway data to select good dropoff point.
And other info like parkings.


+1, just because something is not "naturally" in the routing graph does not mean, the information could not be used during the preprocessing to create navaid points. We must not necessarily provide the information explicitly, if it could be computed from inherent information.

Cheers,
Martin

_______________________________________________
Tagging mailing list
[hidden email]
https://lists.openstreetmap.org/listinfo/tagging
12