Project history

As of June 2020.

tldr: A/B Street has been in active development since June 2018, but the idea has been festering since I was about 16.

Project history

Backstory

I originally wanted to tell a much longer story here of how I came to work on A/B Street, but I'm not sure this is the right time yet. So consider this the quick version.

I grew up in Baton Rouge, where driving is effectively the only mode of transport. (I've gone back and made a point of taking long walks to confirm how antagonistically the city is designed towards other modes.) Very early on, I fell in love with a Nintendo 64 game called Banjo Kazooie, which led me to the online fan communities of the early 2000's. I wanted to create games too, so I started learning programming via library books and lots of questions on IRC. Because I never had any confidence in art, I wound up working on roguelikes, which led to a fervent interest in pathfinding algorithms and collaborative diffusion. When I started driving in high school, I quickly realized how bad people were at it. I remember being stuck at the intersection of Florida Blvd and Cloud and first wondering if the pathfinding algorithms could help with traffic. Can you see where this is going?

I moved to Austin for college. One of the first days of class, I shuffled down the stairs of Gearing Hall past a crackly old speaker apocalyptically announcing the weather forecast (details add color, right?) into a seminar demanding a totally open-ended first assignment to do something interesting. After I left, somebody stopped to ask me for directions, but I didn't know campus well yet. I thought about how Google Maps gave really silly walking directions. So I decided I'd hand-draw a map of campus, showing all of the construction, how to cut through the labryinth that is Welch Hall on hot days, and where to find the 24/7 robot coffee machines, and hack together a routing engine to help people find the shortest path between their classes. The feedback I got on this assignment included something along the lines of, "I was really pretty impressed first that you would be so stupid as to actually try to do this..."

But I did, and that led me to discovering OpenStreetMap, which it turns out was pretty pivotal. (The first version of my campus map was seeded vaguely off an official paper map, but mostly I walked around and invented half-assed surveying methods on the spot.) Next semester, I joined a freshman research stream with somebody who had worked on AIM, UT's demonstration that autonomous vehicles wouldn't need traffic lights. Everything came together, and I started a 3 year journey of building AORTA, a traffic simulator for AVs. Guided by the research lab, I explored the really bizarre idea of letting AVs bid to turn lights green sooner and micro-tolling all roads to disincentivize congestion. Both of these mechanisms would be incredibly unfair to people without the spare cash to back up their high value-of-time, but I brushed this off by saying the currency could be based on carpooling, EVs, etc.

It was great to try research in college; I learned I really dislike munging data and compressing my work into 6 pages of conference paper LaTeX. So I moved to Seattle to work in industry instead, on something completely unrelated to transportation. Lots of things began unravelling for me in Seattle, but one of them was biking. In Austin, I had picked up mountain biking, and all but stopped driving; it was an amazing place to explore and commute by bike. Seattle was different. There were many more cyclists around, but the experience felt more stressful, the drivers more aggressive. I had plenty of near-misses. I kept commuting by bike, but the joy of it was gone. I started noticing how many cars were parked on narrow arterials and wondering why that was a fair use of space. I started paying attention to the public discourse around bike infrastructure in Seattle and feeling like the conversation was... chaotic.

Fast forward to late 2017. This is where I'll omit chunks of the story. I visited London, my first experience with a city that took public transit seriously. When I returned, lots of latent ideas stopped fermenting and started exploding. I threw together a prototype of A/B Street and started the arduous process at work of open-sourcing it and applying to a program to let me work it on for a few quarters. A few months later, I wound up quitting instead, and began to work on A/B Street in earnest.

Year 1 (June 2018-2019)

I skimmed through git and summarized roughly what I was working on each month, calling out milestones. "UI churn" is pretty much constantly happening.

June: polyline geometry and lanes, building paths, protobuf -> serde
July: pedestrians, bikes, parked cars, lane edits
August: porting AORTA's discrete-time driving model
September: multi-leg trips, buses, the first ezgui wizard, randomized scenarios
October: A/B test mode (and so per-map plugins), forking RNG for edit-invariance, intersection geometry
November: clipping / borders, using blockface for parking, time travel mode, test runner framework
December: bezier curves for turns, traffic signal editor, a first attempt at merging intersections, right-click menus, a top menu, modal menus
- the grand colorscheme refactor: a python script scraped cs.get_def calls at build-time
January: careful f64 resolution, ezgui screencapping, synthetic map editor
- grand refactor: piston to glium
February: attempting to use time-space intervals for a new driving model, new discrete-event model instead
- Feb 19-27: conceiving and cutting over to the new discrete event model
March: fleshing out DES model (laggy heads), first attempt to build on windows, gridlock detection
April: first public releases, splash screen and rearranging game modes
May: fancier agent rendering, attempting to use census tracts, finding real demand data
- milestone: discovered PSRC Soundcast data, much more realistic trips

Year 2 (June 2019-2020)

June: contraction hierarchies for pathfinding, stackable game states
July: OSM turn restrictions, misc (I think I was in Europe?)
August: pedestrian crowds, agent color schemes, parking blackholes, a big raw_data refactor to store Pt2D, attended first hackathon
September: offstreet parking, associating parked cars with buildings using Soundcast (before that, anybody could use any car!), implemented texture support for some reason, doing manual MapFixes at scale to fix OSM bugs
- milestone: got the smallest montlake map to run without gridlock
October: parking sim fixes, opportunistic lane-changing, starting challenge modes
November: prebaked sim results, time-series plots, undo for edit mode, traffic signal editor grouping turns
- milestone: Yuwen joins project
December: the UI reform begins (flexbox, minimap, trip timelines, cutting over to SVGs, info panels, scrolling), started naming releases sensibly
- Project leaked to HN, woops
January: UI reform continues, the modern tutorial mode appears
Feburary: UI and tutorial, all text now pure vectors, port to glow+WASM
March: lockdowns start in US, start grouping trips as a person, population heatmap, left-hand driving, info panel and typography overhauls. started engaging with Greenways, started effort to map traffic signals
April: Orestis joins and starts the pandemic model, trip tables, the optimize commute challenge, refactor for people's schedules and owned vehicles, trip time dat viz, MAJOR progress fixing gridlock at the sim layer
May: gridlock progress, upstreaming fixes in OSM, differential throughput and first real write-up, long-lasting player edits, dedicated parking mapper, maybe vanquished the HiDPI bugs, multi-step turn restrictions, random bios for people, and docs like this to prep for launch ;)
- milestone: relying on pure OSM, no more MapFixes

Retrospective

What poor judgments have cost me the most time?

UI churn: I should've studied some UX on my own and started with a clear idea of how to organize everything
OSM data quality: I should've gained the confidence to upstream fixes earlier
Intersection geometry: I should've realized sooner that simulation robustness is more important than nice appearance.
Geometry primitives: I sunk too much time into the polyline problem and f64 precision.

Trivia

The name was almost "Unstreet" or "Superban" (superb urban)

9.2 KiB Raw Blame History