The two-adjacent signals causing extra traffic is something I encounter daily in Berkeley -- Ashby / Claremont & Ashby / Domingo. I probably blow an average of 5 minutes a day cause of that double intersection.
To highlight one other point from the video: You want a light to be able to clear the line built up each time. If not, flux in > flux out and that's a recipe for a growing line, which causes a traffic jam.
> You want a light to be able to clear the line built up each time
Ideally, yes, practically, no. In reality you also have to consider what comes downstream of a traffic light. It doesn't help at all if one traffic light has perfect throughput only to put everything into gridlock in the next node.
That means in most real traffic systems each intersection lane will have to have less then ideal throughput because you don't want to overwhelm the next section, as this could be even worse for the system overall.
If you have some engineering background you can also think about this as impedance matching. If you have a 4 lane road narrowing to 2 lanes you have a sudden impedance increase, leading to flow issues. If you put an intersection inbetween you now have a way to manage that impedance mismatch with signaling times, but you have to both look at the impedance before and after the intersection.
I'm sure AI can figure this out better than a human can. Obvious and empirical, we certainly haven't figured it out or optimized it, with all the engineers that CalTrans has. How often have we been in a four-way controlled intersection and the green is on for the empty side. There's no way that's impedance matching, it's more like Good to go boss, call it a day.
There's no incentive to modernize traffic light control because it's public works and there's no capital profit motive for it in the same extent as private. I would not advocate privatizing it either because that will bring different consequences. The hero will be that new young Caltrans engineer who's grown up in the AI generation and will probably be fed up with the legacy technology.
> How often have we been in a four-way controlled intersection and the green is on for the empty side. There's no way that's impedance matching
Who claimed it was tho? We were talking about throughput. If all your intersection knows is time, it can't decide based on traffic. Which is why inductive loops, camera based systems and so on exist, but (1) they are not deployed everywhere and (2) if they are deployed there may still be a time based limit for the case of defective sensors.
Our traffic systems are already pretty well optimized where we are willing to spend money on them. Note however that if you 100% optimize any system for perfect throughput, it has zero slack to deal with unexpected fluctuations in traffic patterns. That is as true for traffic as it is for networks. Throughput is important, but so is comfort, resilience, predictability and cost. All different metrics you have to keep in mind¹ and all of them can suffer when you over-optimize for one.
The view from inside the traffic that we are part of may make that seem inefficient, but seen system wide the main problems usually are not traffic lights, but harder to change structural issues.
¹: or rather you don't have to keep them in mind, since there are people who specialize on just that and are pretty good at it
efavdb · · focus · HN ↗
To highlight one other point from the video: You want a light to be able to clear the line built up each time. If not, flux in > flux out and that's a recipe for a growing line, which causes a traffic jam.
atoav · · focus · HN ↗
Ideally, yes, practically, no. In reality you also have to consider what comes downstream of a traffic light. It doesn't help at all if one traffic light has perfect throughput only to put everything into gridlock in the next node.
That means in most real traffic systems each intersection lane will have to have less then ideal throughput because you don't want to overwhelm the next section, as this could be even worse for the system overall.
If you have some engineering background you can also think about this as impedance matching. If you have a 4 lane road narrowing to 2 lanes you have a sudden impedance increase, leading to flow issues. If you put an intersection inbetween you now have a way to manage that impedance mismatch with signaling times, but you have to both look at the impedance before and after the intersection.
dogscatstrees · · focus · HN ↗
atoav · · focus · HN ↗
Who claimed it was tho? We were talking about throughput. If all your intersection knows is time, it can't decide based on traffic. Which is why inductive loops, camera based systems and so on exist, but (1) they are not deployed everywhere and (2) if they are deployed there may still be a time based limit for the case of defective sensors.
Our traffic systems are already pretty well optimized where we are willing to spend money on them. Note however that if you 100% optimize any system for perfect throughput, it has zero slack to deal with unexpected fluctuations in traffic patterns. That is as true for traffic as it is for networks. Throughput is important, but so is comfort, resilience, predictability and cost. All different metrics you have to keep in mind¹ and all of them can suffer when you over-optimize for one.
The view from inside the traffic that we are part of may make that seem inefficient, but seen system wide the main problems usually are not traffic lights, but harder to change structural issues.
¹: or rather you don't have to keep them in mind, since there are people who specialize on just that and are pretty good at it