ffff. Can confirm, people setting up traffic lights are idiots. Sometimes they're set up based on 15 minutes of observation. I don't care how many weeks you run your crappy simulations, they're so inaccurate that its laughable.
Source: I built a startup in this industry to prove it.
Edit: I kid you not, I once went to a conference and a traffic engineer for a sizable city got on stage to present the best idea they had and it went like this:
> So once traffic gets to 80% on the main road, we're just going to change all the lights to green and everybody on the side streets can just suck eggs for 5-7 minutes.
That's it. That's the best their simulations could come up with. These people use interns with clipboards to get their data.
I worked at a local department of roads and motor vehicles where their traffic light management software was written in the late 80s to early 90s and looked like it.
Mind you, they had modernised it! It worked on 32-bit!
If anyone thinks that this Windows 95-era application had any kind of smarts in it at the same level as modern machine learning, AI, or even basic queuing theory, they would be sorely mistaken.
I believe all it had were some basic weekday-weekend and peak-afterhours scheduling capabilities. It also had sensor integration, but only at a few hundred key intersections around the city.
