Google's Green Light project retimes traffic lights for 30% fewer stops | Hacker News Reader

Google's Green Light project retimes traffic lights for 30% fewer stops

newatlas.com

74 pointsrbanffy80 comments

I am astonished that green light engineering has not gotten more attention/application over the last, say, 20 years.

So much data is available and 99 percent of all traffic lights seem to be following simple timer logic at worst and some vehicle detection at best, to prevent giving green to non existing traffic.

It almost feels like a no-brainer to start optimizing light signaling by leveraging the massive amounts of data and simulation techniques we have since not far into this century.

That said, I'm not sure / concerned about, how well protected our (location) data is when sharing vast amounts of it between governments and huge for-profit companies.

The city I live in is just about perfect with the light-controlled intersections. Seriously, I love this place. A few reasons:

1. At night, many of the intersections here go red for all directions, and will immediately turn green for a car as it approaches. If you're going the speed limit, you don't even need to slow down. (To be clear, it gives you a green before you would ordinarily need to start braking)

2. Protected left turn signals are only installed where they're appropriate. The last place I lived just threw them up at every single intersection in some sort of ass-covering maneuver. Here, if the intersection is low traffic and has good sight lines, then you make your left after yielding to oncoming traffic.

3. The timing is super tight. I regularly notice cross-traffic signals going to yellow just as the final car passes through the intersection. I'm not sitting there wondering, "Green for who??"

4. If I enter a protected left-turn lane while the cross traffic yellow is lit up, it'll still give me the arrow. Everywhere else I've lived, I'm "too late" and have to wait for an entire new cycle before I can go. But here, as I'm passing over the induction loop, the controller instantly queues me up before the oncoming thru traffic.

I've been half-joking that I'm going to write a love letter to the municpal traffic engineering department, but I really think I should. I just keep noticing their work, especially when I see how bad it is in surrounding cities.

I don't need exact location but I'd love to know what country that is.

I'm in Southern California.

Almost all infrastructure spending is a no-brainer that puts more into the economy than it takes out, but the USA is completely allergic to it anyways. This part of infrastructure is not unique in that regard.

The US spends very high amounts of money on infrastructure ... but like in many other areas gets a very poor return on that investment. We underinvest relative the both the current need and potential benefits, but it isn't because there isn't a willingness to spend.

The US spends high amounts on everything from infrastructure to healthcare to education to anti-poverty.

The gaslighting that we don’t spend enough comes from the corrupt groups looting the public to deliver stupendously bad results.

Most people opposed to government spending oppose that corruption and theft, not the underlying project. Eg, rail projects that are ridiculously expensive by international standards.

Yep, in my local area, light rail gets proposed all the time. And I'm not against it in principle. But the specific project proposals, even with their optimistic ridership numbers, pan out to something like we could buy every rider their own mid-sized sedan and pay for a personal driver for the same overall sticker-price. Figuring out the cost-effective way to deliver those kinds of project, improving state capacity, seems supremely important for the US.

It’s a great investment for the landlords nearby!

https://en.wikipedia.org/wiki/Henry_George_theorem?wprov=sft...

and it gets a very poor return on investment because it is done inefficiently - think of the 60 foot wide roads and everything else you have seen.

Light timing created during the summer may be far less ideal or even dangerous for the same area during winter. Likewise, pedestrian interactions are not identical throughout the year and may be impacted by these timings. Drivers build expectations of how intersections, particularly ones near their home, tend to operate, so season or situation specific timing may actually create accidents around the change boundaries.

This is a problem that you can easily over optimize for and one that is difficult to measure the "full system" impacts of.

This. When they changed the timing of my usual intersection it caught me by surprise (and just entering the intersection on when it turned red). Thankfully no one on the cross street was trigger happy, but I can anecdotally attest that I was accustomed to a much longer yellow and behaved accordingly.

I do believe that's why most accidents occur close to home, we develop expectations and operate on them, blindsided when conditions change without warning.

But... to play devil's advocate, it's hard to develop expectations from a dynamic system, as it's operation is constantly changing/adapting to conditions.

Anything that changes needs adaption by its users, especially for frequently used (hence nearby in this case) mechanisms. There will be accidents in the short term but I'm pretty convinced in the medium to long run, things will get safer because you don't tend to shortcut/bypass a well optimized system as much as a dumb system. For example when it is red at night and you see noone is coming from other directions, you might get tempted to break the rules.

Similarly, in my home town quite a few people got killed in their cars by a train because of the long 'no train passes while red' time. In the end they solved the problem the opportunistic/expensive way: build 3 very expensive bridges. It would have made more sense to make the signal control electronics a bit smarter instead.

Was there a new red light camera set up at the intersection recently? The city-contracted companies in charge of the cameras alter the yellow light duration to increase ticket revenue supposedly

Nope. There are cameras, but that's presumably for license plate tracking. It's been nearly a month, so I can comfortably say it's not a red light camera.

> no one on the cross street was trigger happy

A green light is by no means a permission to blindly accelerate. This is something that needs to be hammered in during driving lessons.

From what I've seen, roundabouts/traffic circles work just as well. They naturally prioritize traffic in the heaviest direction without needing any power or technology at all.

> From what I've seen, roundabouts/traffic circles work just as well.

They do not.

Traffic circles work best for diffuse, medium-level traffic that appears continuously without burstiness and is roughly even to all exits.

When you have "rush hour overload on one direction" (which describes 90% of the US intersections with problems) traffic circles range from worthless to actively harmful. Traffic lights work much better in those situations.

And, at other intersections, anything will work including removing traffic controls.

Traffic engineering is all about tradeoffs.

The mayor of Carmel IN is my hero. He replaced over a hundred traffic lights with roundabouts.

https://www.cnbc.com/2023/08/16/why-roundabouts-are-so-uncom...

A buddy runs the traffic lights for a midsized city. One issue they’re running into is sensors/cameras failing after X years and not having the budget to replace them when needed. All the fancy algorithms are only as good as the data, when there is data.

They’ve had the technology for a 20 years. I remember in the early 2000s my traffic engineer uncle describing a system he installed covering the major artery in their township. All of the lights were networked it would batch up people turning onto the artery then virtually guaranteeing your batch would have all greens on the artery if you stayed with in 5mph of the speed limit.

I'm also amazed that no one has analyzed how much the additional accelration needed after unnecessary stops is contributing to our greenhouse gas emmisions.

Air polution from vehicles is directly related to acceleration, this is when most fuel is burned.

So I would think traffic opptimization would be a climate change prevention activity...

Well, if the negative externalities of gasoline / diesel were priced in correctly, then individuals would see the tradeoff between large / small and hybrid / ICE / EV cars, and say "yeah it'd be nice to save some of those expensive petrol if my city could unfuck its traffic." Or improve public transit

I worked in a field where signals were a big component. Nationwide - the variety is quite real. It could be skilled labor. Some are mechanical switches still.

Back when we (Zürich, Switzerland ) had a more car friendly city I recall the green waves that would trun lights green if you drove the speed limit. Back then 1980 and 1990s it was just timing and traffic detection rings under the pavement.

Nowadays public transport gets to go first and the "green wave" is optimized for bus lanes etc. (in the 80s to but not as extreme. ).

I don't recall having hitting a green wave in a long while because traffic is so bad during the day (this is by design to discourage driving through the city) something like that you only get to enjoy at night but by then many lights get set to orange anyway.

This really only works on one-way streets.

what are you supposed to do at an orange light?

Swiss traffic lights sometimes change over to a blinking orange schedule at night, where the junction is uncontrolled and you treat it as yield or sometimes a stop instead.

The middle light in a traffic light is called yellow in the US, and orange in (some parts of?) Europe.

In Boston you speed up so you don't get stuck at the imminent red light.

If we're responding to the question with answers that have nothing to do with Zürich, like parent comment asked for, then in New Delhi you just ignore the traffic lights.

Interesting use of data. But some places near me have actually decided to actively make their lights timing worse, to encourage fewer cars to travel through the city.

There's a pedestrian crossing in my city on a street with a decent amount of traffic that somewhat recently got reprogrammed to almost instantly stop cars when the pedestrian sensor is activated, and it doesn't have an internal cooldown timer. This ends up in situations where you're the tenth waiting for pedestrians at the crosswalk and you end up rolling three car spaces forward before the light turns red again because another pedestrian activated the sensor.

As a frequent pedestrian, this sounds great. A slow car is a safe car. I've often wondered why lights don't seem to be timed in cities for how long it takes to walk a block.

> A slow car is a safe car.

A slower car is less likely to severely injure you than a faster moving car. Your safety factor has possibly improved to an unknown degree given the particular intersection, roadway configuration and angle of incident, the car has in no way "become safe."

> I've often wondered why lights don't seem to be timed in cities for how long it takes to walk a block

The energy stored in a 3000lb+ object moving 30mph is far greater than the energy in an 150lb object moving 3mph. The overhead of having to start and stop movement as a pedestrian is not comparable and won't rise to the level of design criteria the way it will for vehicles.

This is less great when the result is:

* Massive lines forming that congest multiple blocks (increasing danger to primarily car occupants, decreases throughput of public transit even with dedicated lanes) * Drivers, becoming increasingly less patient, gun forward any opportunity they have to progress (increasing danger to pedestrians) * Idling vehicles for longer periods create increased pollution (debatable, but theoretically would affect everyone)

But I'd say the biggest losers are when this impacts buses.

I've definitely experienced convenient light timing walking through San Francisco.

How long it takes who to walk a block?

As opposed to most situations where the pedestrians have to stand and wait for permission to cross, normally whilst breathing in exhaust fumes.

Omfg I love this

(I don't own a car)

This is dumb. It should only changw to benefit others, not to just make it worse for you as a driver.

I live on a street that is two-way for 99% of its length and one way for the last bit ("do not enter" at that end). The purpose is to prevent drivers from using a neighborhood street as a cut-through to avoid traffic on the parkway.

It sounds like the lights timing as described is similar: make these streets less advantageous in hopes that drivers will take other streets (presumably a ring road in the upthread case).

Change*

weirdly enough i apparently live not too far from the "longest red light in the USA." it's at the intersection of the "haunted" Clinton Rd. & Rt. 23 in NJ.

Why not use artificial intelligence and cameras to detect the cars, and switch signals based on that? It's 2023, and we still have the dumbest traffic lights that are time based, even in the middle of the night. We should be able to use artificial intelligence and cameras to be able to say "there's no cars come in the opposite direction, I'll turn on the left turn green light and let this car go."

> artificial intelligence and cameras to be able to say "there's no cars come in the opposite direction, I'll turn on the left turn green light and let this car go."

You don’t need artificial intelligence or cameras to do that. Inductive loops can detect cars and bicycles and their velocity just fine and control nearby traffic lights, and have been doing that for decades.

I would think such local control is less prone to failures, too.

Interview with a traffic engineer (IMO one that’s a bit too focused on throughput): https://bicycledutch.wordpress.com/2016/06/21/traffic-lights...

Their systems aren’t completely local, given “every bus is tagged and the computer knows all the time-tables of every line. The system knows exactly when which bus needs to arrive at a certain intersection. The installation checks whether a bus is too early, on time, or running late and sets the lights accordingly”

Why use AI at all? An algorithm informed by traffic sensors seems perfectly sufficient and more reliable.

There are companies like Miovision (https://miovision.com) trying this.

The problem is you have to do this city by city and they are so many local regulations not to mention how cheap they are with budgets.

There's a long-running construction zone near my house where the road narrows to a single lane (it's a two-way street). At first they had flaggers at both ends controlling traffic, but at some point they rolled in some portable traffic light thingies that have cameras on them and do appear to work this way. I am a little suspicous of it though since the other day I noticed the light turn green for me while the last oncoming car was still clearing the street.

> Why not use artificial intelligence and cameras to detect the cars

Hell no. You want something that's explainable and predictable, not something that will allow a truck to run over a bunch of kids, just because it's painted in a pattern engineered to make it be identified as a bunny.

No we shouldn't, unfortunately, cameras fail trivially in uncommon-but-definitely-yearly weather scenarios in ~all environments.

I rage when I'm sitting at a red light and there's zero traffic utilizing the green for 30+ seconds.

Worst is when it turns red on your approach and there was never any cross traffic. I notice this happens frequently at a certain normally-busy road I often cross in the middle of the night and I think it works this way by design to give priority to the maybe-traffic on the main road.

This looks a bit over-engineered?

Roundabouts work mostly fine, reduce emissions, and are safer for pedestrians.

Seems easier to do this than to rebuild every intersection in the world.

And many roundabouts still need lights because lights prevent gridlock which roundabouts do not do.

> Seems easier to do this than to rebuild every intersection in the world.

Just the ones that need fixing, I guess?

Also, this solution introduces a lot of new moving parts, doesn't address pedestrian safety, and savings may not be significant enough in the long term, compared to a roundabout.

> And many roundabouts still need lights because lights prevent gridlock which roundabouts do not do.

Most roundabouts are small enough, so they don't really need traffic lights. Regardless, gridlock happens usually only during peak hours, which would account for, say, 3-4 hours a day. Intuitively, roundabouts would be more efficient than these smart traffic lights ~80% of the time, and outside that, regular traffic lights would do.

> > Seems easier to do this than to rebuild every intersection in the world.

> Just the ones that need fixing, I guess?

Which is the vast majority, at least in my part of the world. So, yeah, seems easier to do this than to rebuild (almost) every intersection in (this part of) the world.

I don't think I understand your point.

The solution proposed by Google doesn't seem to be too efficient, for several reasons: * Traffic-aware lights already exist. This could already reduces stops in the vast majority of the cases. * Google's system requires the deployment of an intricate, city wide network of smart traffic lights, driven by AI. I couldn't begin to imagine the kind of maintenance that would require. * Regarding that, this is a proprietary solution. Cities would likely have to pay Google to keep the network working.

To me, this draws some parallels to how Europe and the US handle house heating. Traditionally, houses in the US have oversized heating solutions, whereas European house construction is focused on insulation. Yes, swapping all the windows for energy efficient ones is more expensive in the short term, but it is cheaper in the longer term.

> Google's system requires the deployment of an intricate, city wide network of smart traffic lights, driven by AI.

Did you just make this up? From the article: "company says that rolling the changes out can be a five-minute job that uses the city's existing management systems".

Google's system isn't about some smart AI driven lights. It's basically just AI developed light schedules for dumb lights. It doesn't require any deployment.

> Cities would likely have to pay Google to keep the network working.

Oh, you really did just completely imagine some system of your own.

> Oh, you really did just completely imagine some system of your own.

Turns out, data acquisition and processing does cost money.

1. Everything you say could be true, and Google's solution could still be more efficient than tearing up 90% of the intersections in the US.

2. Retiming the traffic lights is an answer to a problem. Google's approach is only one possible way of getting to that answer. There may be more efficient ways.

It feels like you're misunderstanding something. This is not a system controlling lights in real-time. It's probably just generating a schedule as a CSV file that a city employee uploads to their existing traffic light control system.

Many cities already have such centralized traffic light control systems in place. This tool is just optimizing what already exists.

Besides roundabouts often being bigger and more unpleasant to cross for pedestrians than an equivalent traffic-light junction, roundabouts have a sweet spot - low to moderate traffic, and sometimes heavy traffic when there's a balance of exit usage. When not all exits are used frequently, heavy traffic can starve entrance from roads whose exits are not used. Exiting traffic frees up a gap for entering traffic to use. If there's nobody exiting before your entrance, you don't get to join. Traffic lights are sometimes added to roundabouts to patch this.

sounds easier to upgrade signals than redesign roads

Two choices: adjust traffic signal timings or rebuild intersections into roundabouts. Supposedly the former is an over-engineered solution and the latter is the elegant one. Beggars belief.

Couldnt agree more, theyre better for pedestrians than any intersection.

Honest question, how are roundabouts better for pedestrians than intersections? At a light-controlled intersection, the traffic is forced stop for a period of time, allowing for safe passage.

They're not. Roundabouts with any significant volume of traffic are, for the most part, terrible for pedestrians. Some rare but progressive roundabouts have zebra crossings on every arm which make them much better, but the continuous flow nature of roundabouts make them pretty hazardous.

> They're not. Roundabouts with any significant volume of traffic are, for the most part, terrible for pedestrians.

I don't think that's true in general [0] [1]

> Some rare but progressive roundabouts have zebra crossings on every arm which make them much better

Those are just... regular roundabouts? What would be the point on having roundabouts with no way for pedestrians to safely cross the street?

[0] https://www.fdot.gov/agencyresources/roundabouts/benefits.sh...

[1] https://www.iihs.org/topics/roundabouts

Mostly because of speed. Roundabouts physically force drivers to decrease speed. One may run a red light, but cement is hard to ignore.

I like this stuff, but we got to stop letting google do everything.

Why you might ask?

Well, it conflicts with their other goals. If we blindly following their directions we can be manipulated to drive by a store that paid to ensure so many people drive by. Or drive by specific advertisements.

The good news though is, if this is a product, then it's likely going to be canceled in 3 months.

> Well, it conflicts with their other goals.

Less flippantly, I do wonder whether this optimization does conflict with society's other goal of reducing average urban traffic speed to reduce danger to pedestrians.

Also discussed 5 days ago:

Google's AI stoplight program is now calming traffic in a dozen cities worldwide (engadget.com)

https://news.ycombinator.com/item?id=37846273 - Oct 2023 (9 comments)

I'm 100% convinced the traffic lights in my city are times so that you catch every red light. I can see the antenna that communicates with the next light. And when I get to the next light it turns red just before I get to it. Every. Single. Time.

I assume they're trying to collect more taxes from gasoline. Either that or they are about as stupid as you would expect from people in this part of the world.

I think it's pretty selfish of you to assume that it's in a city's best interest for you to hit green lights.

This project can be an excellent tool in a civil engineer's toolkit, but that doesn't mean that every intersection should be perfectly optimized for traffic to always flow unobstructed, especially around hot pedestrian areas and busy hiway entrances and exits.

And it's extremely presumptuous of _you_ to assume that I don't know insurance companies are the reason for this change. Stop reading into motives.

When I was a kid, we went through some town in the midwest when we were on vacation. They had no speed limits - just signs that said that the traffic lights were synchronized for 25 MPH. Once you figured out that they were telling the truth, you drove 25.

So you can get more efficient traffic flow and eliminate speeding (if you've got a perfect grid system).

Many French villages have adopted this slightly differently: there's one traffic light on the entrance to town, and it turns red if you're going more than 50kph after the village name sign (equivalent to a 50kph speed limit sign in France).

In my city if you're near the front of the pack and drive 15 mph _over_ the speed limit, you can get to the next one before it stops the entire pack.

Reply on news.ycombinator.com