I like this as a single aggregate measurement to base scores on. Some of the difficulty comes from displaying this score as the simulation runs through the day. At first the average might look great (a change helps the bus in the early AM), but then rush hour hits, and the change actually makes things worse. Or maybe a re-timed traffic signal helps buses going northbound in one spot, but hurts the southbound buses, so also seeing the change between each pair of stops is important. https://imgur.com/Fk1GfKG is what I've got for this so far, but I think a different way to visualize this is necessary. :)
This is totally on my radar. I want to compile to wasm and target WebGL. I'm using https://github.com/glium/glium/ to target OpenGL on Windows, Mac, and Linux right now. I've been meaning to check out https://github.com/grovesNL/glow/, which also supports WebGL.
I'm pulling in trips (when does an agent leave, where do they depart from and go to, what mode of transport do they use) from https://www.psrc.org/activity-based-travel-model-soundcast, which is a very fancy model that uses land use, census, vehicle counts, etc. This is the most realistic data that's public, from what I've found.
Map data comes from OpenStreetMap. Aside from the problems I mentioned with inferring good geometry for some intersections, it's pretty solid.
Drivers in A/B Street have to find parking, and walk between buildings to parked cars. So the amount of on- and off-street parking in an area is pretty important; there might be lots of congestion in the sim if I don't know about parking. I'm inferring on-street from http://data-seattlecitygis.opendata.arcgis.com/datasets/bloc..., which is extremely inaccurate, and I have no data source yet for private parking attached to individual buildings.
Until I have some of this data, calibrating the sim against reality is pretty tough.
Creating a [traffic simulation](https://github.com/dabreegster/abstreet/#ab-street) that's both realistic enough to generate results meaningful in the real world, but easy enough to use that anybody living in a city could use it to experiment with some change to cycling or transit infrastructure. Some of the problems hiding in there:
- Inferring reasonable data when it doesn't exist. How are traffic signals currently timed? No public dataset exists, so I have heuristics that generate a plan. If I can make the [UI](https://raw.githubusercontent.com/dabreegster/abstreet/maste...) for editing signals fluid enough, would it be worth trying to collect this data through crowd-sourcing?
- Figuring out what to even measure to demonstrate some change "helps a bus run faster." Should people try to minimize the 99%ile time for all buses of a certain route to make one complete loop over the whole day? Or to reduce the worst-case trip time of any agent using that bus by some amount? Or to minimize the average delay only during peak times between any pair of adjacent stops?
- Less technical: How to evangelize this project, get the city of Seattle and advocacy groups here using it, and find contributors?
