The traffic simulation is now unmaintained

[dabreegster]

Since roughly summer 2021, work on A/B Street has been focused on Ungap the Map, the LTN tool, splitting out osm2streets, improving the OSM import, etc. That's partly because getting the traffic simulation to work reasonably feels endlessly hard, and because there's been clear demand that these more specialized tools are more useful for promoting changes.

I'm making this issue to clarify my current priorities. There are dozens of open issues and ideas for making the traffic simulation in A/B Street better. I have no intention of working on them in the foreseeable future. The work I find important right now might introduce new problems in the simulation, and I'll do the minimum required to make things not crash or get much worse, but I'm not going to put much effort into it. I'm not giving up on the simulation entirely, but improving OSM importing and merging parallel dual carriageways and such is complex work, and slowing down to try to keep the simulation working as it did previously feels like wasted effort. When the dust settles from improving importing, I suspect lots of hard problems in the simulation layer will get much easier to deal with anyway.

In the meantime, if anyone's particularly interested in taking over maintenance of the simulation or working on fixes / new ideas with it, I'm happy to mentor / be involved. Let me know!

[dabreegster]

A cursory list of issues that're backlogged accordingly: #868, #859, #829, #824, #767, #698, #688

[jamesneb]

Would be interested in giving this another shot.

On a personal note, I am now living in Seattle - might be cool to get together at some point :)

[dabreegster]

Would be interested in giving this another shot.

It's a big project, but some smaller ideas to start (more above):

• Smart traffic spawner #767, some easy way of specifying example traffic programatically or in the UI, with some kind of "spawn regularly enough to keep the roads 50% full" logic
• Cyclists should use empty parking lanes #698, though this may wind up being hard on the map_model side less so than the sim side
• Rethinking the entire protocol between agents and intersections and when they can start turns. Make leader agents call ahead before starting a turn, and stop cutting #900 was an attempt

On a personal note, I am now living in Seattle - might be cool to get together at some point :)

That's awesome! I moved to London a year ago, but there are still some A/B Street alumni around Seattle.