Import footways and shared-use paths! The analogue to

[dabreegster]

a-b-street/osm2streets#90.

Also remove the pedestrian plaza area type for now. It adds visual clutter and has no editing/simulation/semantic impact yet.

The pretty satisfying quick demo:

screencast.mp4

Rendering

There are footways and shared-use paths (mostly pedestrian oriented, but slow biking permitted) introduced. How do they look?


The dashed outline style and color is adapted from strassenraumkarte.osm-berlin.org, but the yellow is made darker to stand out from the map background better. I'd like to visually distinguish the two types (footway and shared-use) but don't have good ideas yet.

For the moment, curbs aren't drawn for intersections involving these new lanes:

Night mode is on the garish side:

Behavioral changes

You can't edit the lanes for these next cases, similar to service roads, light rail, etc. I haven't deeply tested how they work in the simulation yet, aside from verifying pedestrians will use them and that routing does seem to work for pedestrians. The shared-use paths are meant to let bikes use them too, but for now, this doesn't work at either the routing or simulation layer. The shared use paths in A/B Street are effectively also just footways, for the moment. To solve that, we need to think about how to simulate interactions between agents (#139) and also probably first deal with bidirectionality on one lane properly (a-b-street/osm2streets#91)

Bugs / data quality problems

The import still ignores separately mapped sidewalks and crossings. In places like Seattle where those are used, the footpaths can wind up dangling at the end. Acceptable to make progress, I reckon.

File size increases, because we're bringing in lots of new stuff. Central Seattle (a pretty large area) goes from 47MB to 78MB! Bristol goes from 26 to 31MB. This is unfortunate, but of course skipping a huge part of the street network previously was bad. I have ideas to reduce size eventually (not storing turn geometry, not storing redundant lane and road geometry).

TODO

I'm still regenerating everything and hit a few problems, will keep iterating. This change also unfortunately breaks most LTN proposals, because it changes block geometry in some cases.

[dabreegster]

Validation and regenerating data mostly done. The hassle is now fixing built-in proposal save files that refer to road segments that've now changed. This is not a new problem, but one's that really been causing me pain lately, so #995 to start thinking about it seriously

[Zireael07]

analogue to what?

[dabreegster]

analogue to what?

Ah, the commit message cut off. To a-b-street/osm2streets#90. The map importing logic has been split out to another repo, so when there's a big change, it happens in two independent commits / PRs

[michaelkirk]

This is great! Finally we can simulate those Foster Island commuters!

[dabreegster]

This is great! Finally we can simulate those Foster Island commuters!

We'll need to start tracking what percentage of a trip is partly underwater, pull in weather/tide data, and sell the results to REI to fund the project ;)

[dabreegster]

Another "regression": when sidewalks are tagged as separate ways in OSM but lack footway = sidewalk, then things will get messy, like around https://www.openstreetmap.org/way/561633983:

I'll call this acceptable. If somebody is working in an area with this problem, it's very easy to spot cases where footway=sidewalk are missing

[dabreegster]

OK, there's a subtle and big change resulting from this PR that caused a few bugs. Previously, the sidewalk graph was totally independent of the rest, so things like floodfilling didn't have to check turn types. But now we have shared use lanes that bridge the sidewalk graph with the driving/biking graph. I fixed a few places to filter during graph searches, but I wouldn't be surprised if more bugs lurk

[dabreegster]

Good enough to merge! The only thing I validated with deep paranoia is the LTN tool in Bristol (where there's now an active public consultation using it!) but everything else seems OK.

Biggest thing broken by this change is a bunch of built-in proposals, because OSM IDs changed. I don't want to go through the hassle of manually fixing them again until #995 means that effort has longer-lasting effects :)