Some Phoenix traffic lights are being run by AI

Most of us have spent time sitting at a red light, wondering why that light was still red when there was no other traffic going through the intersection. Phoenix is trying to reduce the number of times that happens — in part, by using AI.

The city has been working with a platform called No Traffic for about five years; on its website, the company says among other benefits, its technology has reduced Phoenix’s afternoon commute times by nearly 30%. It also says it’s reduced red-light running in the city by 70%.

Simon Ramos, Phoenix’s traffic engineering supervisor, said while the technology is relatively new and needs vetting, it’s also showing a lot of promise.

Ramos spoke more about it with The Show, starting with how Phoenix uses AI in its traffic lights.

Full conversation

SIMON RAMOS: So in the sense AI, many vendors use this terminology and it's more so say adaptive technology. So city of Phoenix has about 1,300 traffic signals and we have the challenge of managing, operating and monitoring traffic level congestions with about a team of six folks.

So we leverage various types of technologies such as AI/adaptive detection systems. So these smart detection system cameras basically just count vehicles on the roadway, kind of monitor their speeds, monitoring, you know, platoons getting through signals or not.

So one of the functions of them is to count the vehicles during the AM and PM and midday rush hours. So a traffic engineer can basically monitor and react to the signal timing in real time. So as far as the AI element, we do have some systems out there that basically react on their own as far as reacting to traffic levels of congestion and on, say, non-reoccurring events, such as crashes or collisions or construction projects and such.

MARK BRODIE: So of the 1,300 signals that the city has, how many of them are tied to a camera or some sort of system that allows folks to maybe switch up the timing of when the light changes?

RAMOS: Roughly about 30% of our signals have this type of technology or some type of technology. So the smart cameras or the video detection systems out there, you'll see when you're driving at a signalized intersection, if you look up on the mast arm or the street light, you'll see somewhat of a camera or a dome type device pointed at the vehicles themselves.

That's basically the number one function is detecting if a vehicle is there or not. So that way the controller or the computer on the corner of the intersection basically provides that green time to the vehicles by demand.

BRODIE: So just to make sure I'm understanding, at one of these intersections that has a camera like this, it will look and see how heavy the traffic is, maybe how many cars are looking to be backed up or might be backed up at a red light.

And the AI doesn't change the lights, right? It tells you or your colleagues in the operation center, look, we've got to, maybe we want to make the red light a little shorter here, or we need to get, keep traffic moving through this intersection.

And then a human is still responsible for changing it. Is that right?

RAMOS: Somewhat. We do have some of these types of cameras, and this is a different group of cameras that we're piloting, evaluating with our partners here in the Valley. So some of these do actually adapt to traffic levels and adapt the signal timing or adjust the signal timing on their own.

City of Phoenix

/

Handout

BRODIE: Really? And how many of those do you have?

RAMOS: We just finished a pilot or a project of roughly about 30 intersections. We're doing research also with the University of Arizona on another study for red light running and monitoring and studying driver behavior. I believe that's another 20, 25 intersections as well. So they're scattered throughout the city.

BRODIE: What did you find in those 30 intersections or so, 30 lights, where the AI was basically changing the lights on its own?

RAMOS: We did find some benefit, obviously, with any new technology. It's going to take some, say, tweaking of the technology or their algorithms to kind of better improve.

So obviously it's not going to be as good as somebody monitoring traffic in real time. So that would equate to an individual monitoring five or 10 signals at once and having control of them and being able to make those timing changes on the fly.

BRODIE: But as you say, there are six of you, right? So I mean, that's 1,300 signals is a lot for six humans to be monitoring. Is this the kind of thing where automation makes sense in terms of trying to keep traffic moving and do jobs that in your case, humans just don't have the capacity to do right now?

RAMOS: Correct. For a city this large, yes.

BRODIE: You mentioned that there are some tweaks that you have to make. I'm curious what kinds of issues, even if there are sort of minor issues, you've run into?

RAMOS: So how this technology works is it's more using kind of machine learning algorithms. So you'd have to teach that technology exactly what it's looking at as far as identifying pedestrians crossing the road versus is it a car or is it a bus, things like that.

Some other types where it may think that it identified a collision, which in effect it didn't. So at the end of the day, somebody has to go in and validate what the camera is seeing or marking as a, say, a collision.

BRODIE: And what is the end goal here? I mean, do you envision a time when all of the city's traffic lights are controlled by AI?

RAMOS: So I guess in a perfect world, if we do have this technology deployed at all 1,300 of our signals, we would rely on this technology as far as the data that it would provide us. So we can make us a sound engineering decisions to make improved countermeasures at the intersections. So the kind of help us identify which signals are, you know, for example, have red light running events or which signals are prone to accidents, which signals have higher vehicle counts, which ones have, you know, more pedestrians.

And then that data is then, you know, kind of vetted and evaluated through traffic engineers to make better decisions on, say, infrastructure, striping, signing, things that such.

City of Phoenix

/

Handout

BRODIE: Have you found that these platforms provide that kind of data so far?

RAMOS: Yes.

BRODIE: And has it been useful?

RAMOS: It has. When we're referring to the no traffic and the project that we're working with the University of Arizona, 20 years ago, we would not be able to collect this type of data, but we can actually determine when the yellow light comes on, what exactly is the driver doing? Are they trying to accelerate and beat the red light? Or are they, in fact, trying to not enter the intersection or reduce their speeds?

That tells us, you know, if it is in fact, you know, vehicles making attempt to not run the red light, but they just don't get it in time, then basically we would start looking at our signal timing. Do we need to reevaluate how long that yellow should be so the vehicle can either safely stop or safely get through that intersection?

BRODIE: Are there any safety concerns that you have or you have noticed with this technology? And I'm thinking especially of the signals that are not being switched by humans.

RAMOS: So as far as safety issues, all of our signals are timed to national standards. So yellow time is calculated by a person and then vetted, and then that gets deployed out in the field. That time serves every single time. That goes the same for pedestrian time to walk across the street. So we haven't fully relied on a pedestrian walking up to the corner of the intersection. The intersection will still be equipped with a push button to guarantee the pedestrian gets enough time to cross the street and road.

Now, the addition to that, if the pedestrian does not have enough time to cross, the detection system will then kick in, determine someone's still in the crosswalk or something is still in the crosswalk, and provide that extra time. But there's fail-safes and timing parameters that are put in place to ensure that the intersections are still operating for safety standards.