[Tagging] Tagging average speed [Was: Re: Residential roads]

[Colin Smale]

  On 03/10/2010 17:36, Peter Wendorff wrote:
>  On 03.10.2010 17:16, Ralf Kleineisel wrote:
>> On 10/03/2010 05:04 PM, Anthony wrote:
>>
>>> Maybe it's just because of where I live, but I don't see how it 
>>> would be.
>> Well, where I live (Germany) we have a legal limit of 100 kph on roads
>> outside of cities, motorways excluded. This legally applies even to
>> small roads if there is no sign indicating a lower limit. On many roads
>> you can achieve this speed, too. But on the other hand we have lots of
>> narrow, twisty country roads where a normal driver does not go faster
>> than 60 kph. In the Alps it is even more drastic. For estimating the
>> time someone will probably need to drive along a road this information
>> would be very helpful.
> Sophisticated routing systems include the shape of roads (curves, 
> width, surface quality) and their importance for routing networks, but 
> I don't think, that should be added as mostly static attributes to the 
> osm database.
>
> This would be a real huge effort, and I fear, it wouldn't really help.
> To achieve a hopefully useful system, we would have to model:
> - speed between monday and friday
> - speed at weekends (in Germany different additional due to heavy 
> goods being forbidden to drive at weekends partly),
> - speed at different daytimes,
> - speed in different seasons (here where I live, there often is 
> agricultural traffic with really slow speed you have to wait behind; 
> at least in fall)
> - speed at different weather conditions (rain, icy road, leaves at 
> autumn)
> - speed at holiday times
>
> To be useful, this has to be
> - nearly complete
> - up to date with changes at road network around
> - measured in a slightly comparable, and objective fashion.
>
> I think, that's neither possible nor useful inside the osm database.
I agree with your comment that it would be a huge effort, and I'm afraid 
that storing objective data for all roads under all conditions would 
also cause an explosion in the size of the database. Routing programs, 
like many real-world programs, work with heuristics, deriving a value 
that's *mostly good enough for practical purposes* from the available 
data, possibly with a mechanism for an explicit override value for 
pathological cases where the heuristics give completely wrong results. 
These static values could then be adjusted according to dynamic data 
such as weather or traffic conditions.

It would be good to have some input from an expert in navigation systems 
to hear what kind of attributes would be useful for predicting journey 
time for a. choosing the best route and b. informing the user of the 
ETA. We can guess at road width, incline, curviness, priority at 
junctions, number of lanes (?), legal maximum speed *for the road*... 
Having got to some kind of data model for routing heuristic parameters 
we can then talk about having deterministic defaults per territory - an 
optimised algorithm for "what territory is point (X,Y) in" plus a 
"table" of defaults for each territory. (Normally territory==country in 
Europe but it doesn't need to be).