Lots of traffic lights already have cameras on them. Hook 'em up to an AI and a goal of maximizing throughput.
https://www.researchgate.net/publication/281356849_On_Traffi... where you have nice (or almost) nice regular blocks with a constant speed. When you start deviating from this, the math all breaks down.
The ideal case would be to pulse the traffic such that given a certain amount of traffic flow and everyone following the proper speed limit that the time it takes for the traffic pulse to get to the next traffic light is green (and the pulse can continue).
This can be seen when you switch from one traffic light synchronization district to another - and hitting a traffic light.
There's a stretch on my commute where when I follow the speed limit (its 25 in this segment), I can go from one end of the traffic district to the next without ever hitting a red light (about 15 blocks).
The thing is that if the traffic isn't pulsed, then the spot where it changes from one traffic district to the next backs up significantly and causes other backups further down traffic.
First, the vast majority of signals aren't networked, they're on simple timers. And you can't just put them on the internet for hopefully obvious reasons.
Second, there's a bit of a butterfly effect. Change one thing and the whole system reacts in sometimes unpredictable ways.
Third, you need to account for pedestrians too. There's ADA requirements around how long signals are green for based on the width of the road.
Fourth, you need to account for pedestrian load. There's no cameras trained on the areas where pedestrians gather. If you're going to say they just need to press the button, no:
https://cal.streetsblog.org/2020/04/01/stop-touching-pedestr...
Basically, it's really not that simple.
> First, the vast majority of signals aren't networked, they're on simple timers.
Doesn't matter. Even doing the ones that are networked would be a big improvement.
> And you can't just put them on the internet for hopefully obvious reasons.
But you can put every neighboring building on the internet with their security cameras? It's not like you need hires or high frame rates to detect large moving blobs, count them, and estimate their speed & distance.
The problem is, the city has no incentive to do this, as the cost falls on other people.
To the first point, the networked lights are already in this mode (it's the reason why they added it in the first place). I can't find the report right now but upgrading all of the signals on 2nd from Denny to Yesler cost something like 30 million bucks.
The upgrade in the Mercer project area isn't exactly a glowing success:
https://www.theurbanist.org/2018/09/06/eleven-ways-adaptive-...
I didn't consider it because the solution is obvious. An interlock that makes it electrically impossible to have 2 greens, like an airlock has a piece of hardware that prevents both doors opening at the same time. I suppose I should also mention the obvious that if the internet connection goes down, or the camera lens gets iced up, etc., it should automatically revert to egg-timer mode.
See https://en.wikipedia.org/wiki/Therac-25
> I can't find the report right now but upgrading all of the signals on 2nd from Denny to Yesler cost something like 30 million bucks.
Somehow the money is always found for internet connected cameras to spy on us.
> The upgrade in the Mercer project area isn't exactly a glowing success:
These are solvable problems. Except point 11, which is ironically complaining about it working successfully at what it was supposed to do.
https://en.wikipedia.org/wiki/Sydney_Coordinated_Adaptive_Tr...
We have some trams, mostly fairly new routes, but hardly any share the roads with traffic. There used to be a lot that did, until the 1960s I believe, but they got rid of them in favour of the supremacy of cars. We have good buses and trains (and ferries) here though.
> These buttons have long been decried and criticized by advocates for walking, anyway. The buttons’ purpose is less to keep people safe than to reinforce the primacy of cars on the street by forcing people who want to cross a street to “beg” for a walk signal.
This makes it sound like buttons are some kind of implement of the bourgeoise, used to "reinforce the primacy of cars" and keep the pedestrians powerless. In reality, whether pedestrians like it or not, cars are primary. The roads literally exist for cars. As a pedestrian, I would do the very difficult task of pressing a button, if it meant that as a driver I could have less traffic and hit less red lights.
> Typically, a person expects the signal to show walk, then flashing don’t walk, then solid don’t walk. Signals follow a familiar pattern and people embed that pattern in their muscle memory. Adaptive signals, however, can flip from walk to don’t walk to walk again
That simply doesn't happen, adaptive or not.
Keep an eye out on the hn home page I'll be posting my poc in the next month
Off hand, I'd say lights are often timed such that once you get one green you get a lot of greens, then you're asking for more stop and go traffic simply because someone showed up and threw off the sequence.
Take away that and you gain some benefits, but you also lose in other areas.
Not an easy problem.
That's what AI is good at. Besides, isn't a little incongruous to argue that the optimal solution for a complex optimization problem is a 1950's egg timer?
Besides, if you have an intersection that deals with 500 cars and 2 bikes, what would make the most sense to optimize for?