Stop Signs for Profiles

[daniel-j-h]

Profiles are currently unaware of stop signs.
We want them for better ETAs.

Tagging: highway=stop on nodes.

http://wiki.openstreetmap.org/wiki/Tag:highway%3Dstop
http://taginfo.openstreetmap.org/tags/highway=stop

[1ec5]

The wiki page makes its tagging recommendation sound authoritative, but how prevalent are tags such as priority_road in places like North America, where the concept of a priority road has no formal/legal status and would not be signposted as such? In my armchair-mapping travels in the U.S., I've seen three tagging styles that closely parallel tagging for traffic signals, and usage for each is much less common than for traffic signals:

1. highway=stop on the intersection node, even for non-all-way stops
2. Less commonly, highway=stop on each approach
3. Just as commonly as 2, a stop relation

Hopefully my observations are an outlier and approach 2 is the most common generally.

[MoKob]

From discussion on traffic signals (#2758 (comment)), stop-signs cannot be modelled just like traffic signals since stop-signs give a precedence to one way while penalising another. The same can be said to some extend for traffic signals -- the phases are not necessarily of the same duration -- but nothing we have currently in our scope.

The issue here is: we need to assign the stop-sign only to a specific way direction.

[danpat]

In the last couple of days, I've been doing a bunch of test driving with a high-resolution GPS and looking to improve the Lua turn_function.

One thing is fairly clear from the data - whether you need to stop or not makes a big difference in the turn cost.

Right-hand turn that required a stop (this intersection has a stop sign for traffic heading west, north-south is a 'priority road' with no stopping required):

Right-hand turn that required no stop:

Across all the data, there was a roughtly 2-3 second additional penalty for a turn that requires a stop. Turns without stops fall into the 2-4 second range, turns with stops fall into the 4-8 second range. This seems significant enough that I think we want to support this where tags are available.

My current thinking is that we need to pass more data to the turn_function, for example:

function turn_function(angle, approach_speed, exit_speed, turn_radius, stop_required, has_signals)

This would allow the turn function to include all the factors (that I can see) are relevant to estimating the average turn penalty.

We don't have all of this data available where the turn_function is called yet, so there's work to be done. #2822 adds approach_speed and exit_speed, but based on my testing, there are more factors than that.