Smarter programming of stoplights could improve efficiency of urban traffic | Hacker News Reader

Smarter programming of stoplights could improve efficiency of urban traffic

phys.org

62 pointsmeneses45 comments

With all the image processing cameras out there, and the cameras already mounted on the traffic lights, it boggles the mind that nobody is making traffic light controllers that optimize traffic based on current conditions and a cost function.

For example, if 5 cars are coming from one direction, turn the light red for the 1 car on the cross street.

I'm pretty fed up with the light perversely stopping the 5 cars at the last moment to let one car, that is already stopped, enter the intersection from the cross street.

I wonder how much gas and congestion can be saved this way.

While this sounds great on paper you have to remember that traffic signals are absolutely mission critical components to the traffic system and what you are describing pushes them to the cutting edge of what many might consider "experimental" technology. There can be no trial and error.

Also remember that traffic systems are not developed by start ups that can iterate quickly with seemingly endless funding and/or no overhead. These systems are usually under the control of municipalities that are already strapped for cash and certainly do not have elite teams of developers at their disposal.

Sounds like a startup needs to start start developing and licensing traffic light software to municipalities. Seems reasonable enough to me.

That's not a terrible idea but once again you run into the issues of reliability and budget. At the end of the day you need to convince city officials to spend what little money they have remaining in their budget on your new system because it offers a greater value than all of their other proposed upgrades to the city AND that what you have is as reliable or better than the current system they have in place. This means asking them to trust a small company they've never heard of to manage the transit systems for their city and potentially drop contracts with companies that they have been working with for a very long time (remember this is politics). Also keep in mind that software or hardware failure in these systems could result in multiple fatalities. I'm not trying to be a naysayer, it's just important to recognize that getting something like this implemented is not as easy as convincing consumers to try/buy your product.

The best bet would probably be to get a forward-thinking city to experiment with this in a small area and then go from there.

The city already has no problem funding cameras mounted on every traffic light, and all kinds of other surveillance equipment. A generic, self-adapting light controller would work for probably 90% of intersections. I'm hard pressed to see how it would be that expensive.

It would not be hard to put in a physical interlock so both lights could not be simultaneously green, regardless of what the software commands. I don't believe this is a significant fatality risk.

I would assume mounting cameras on traffic lights is an easier sell to a city, since they can be marketed as a revenue generation device.

This sort of local development work sounds like a project for Bill Gates.

While traffic lights are mission critical, we can absolutely make it fool proof, the first time. It's actually not very difficult.

Don't forget, many lights right now are using weight sensors in the ground (SEE EDIT-2), A. that is very costly, B. They are prone to damage...a lot of the time, and C. with current technology in cameras, I'd say they will be more reliable and give better information about on-coming cars from all directions to the traffic light.

It is quite simple to say, if the conditions are bad for visibility, or the camera is inoperable, or even a bird is sitting in front of it, whatever it is, the simple fall back can be 'dumb'. A timer. As they have been for many many years.

I've thought about this a lot and imagine in the future that each intersection will actually be able to communicate with the on board systems of autonomous vehicles and create a very efficient system. At that point, many intersections would not even require vehicles to stop. It would regulate the speed of each vehicle so each one could go through with minimal energy impact. ...Sure...it might be a 'close your eyes' moment...but regardless, we can make it safer than how it is today...with humans making the stop/go decisions.

Whatever upgraded systems in traffic lights we move to, we have to start thinking about autonomous driving. It will be a HUGE leap forward in not only getting around efficiently, but getting around quicker! Human driving tendencies are inherently inefficient. The slinky effect on roads and highways is quite annoying.

EDIT: Another 'efficient' thing we can do is place short range charges into the ground near the stopping point at lights. This is down the road when a majority of cars on the road are fully electric. But as the vehicles sits at a light, it would charge. The power would be obtained through using photocells on the surface, so when a car is not currently waiting on top of it, they are building a charge. This of course would require the price of batteries to go way down to be feasible, but essentially, all forms of consumer transportation will be powered by solar.

EDIT-2: I'm wrong about weight sensors, I actually did not know about inductive loops at lights. TIL. Either way, they have a limited range of detection.

I'm pretty certain there no lights that have "weight" sensors. The sensors for triggering signal changes use large induction loops to detect when ferrous metal (a car frame) is near by.

Yep as I found out with my Carbon road bike, they don't see me (or a bunch of us) which can be fun.

A strong neodymium magnet attached to your shoe or one of your components should be enough to trigger the sensor for the lights, I've seen them work on scooters and small motorcycles.

This is an excellent idea, thanks!

Currently off my bike due to ill health but will suggest it to the cycling club.

No worries! Some people suggest using a stainless metal case around the magnet since they tend to oxidize a bit, but there are a few good guides around with ideas.

I agree with you that the cameras could be made fool proof, or close enough with a fall back at least that it would still be beneficial. I just wanted to point out that the vast majority of car sensors use an inductive loop which is far less prone to damage and relatively inexpensive compared to weight sensors.

Certifying and proving such a system was actually fool proof would be quite costly though, as in house certification would probably not satisfy safety regulations.

Side note: most lights use inductive loops not weight sensors for vehicle detection.

Oh I agree. Building a prototype and proof of concept alone would go along way to getting more on board.

I think the biggest challenge though is getting municipalities to replace current systems. They have already invested millions and millions to get the current traffic system up. It would almost have to be the next cycle of traffic lights, when the old ones get replaced (depreciate fully). And of course, cities would probably want to do their own independent testing/study...

It would be sweet to start seeing them appear on new construction though.

Think I remember reading that the failsafes in traffic signals are hard-wired. e.g. the green light in one direction is grounded through the green light in the crossing direction, so they can't both be illuminated at the same time. The idea is that no matter what goes wrong with the control software or sensors, you can't get conflicting green lights.

Just to add to this:

Current traffic light controllers and/or vehicle detection system (VDS) units are also fairly ruggedised: they're designed to live in outdoor cabinets with minimal active cooling, survive up to +50C (or more) while still operating, and have long replacement lifecycles.

Modern equipment capable of processing imagery "in the field" - some kind of x86 machine - would have a hard time meeting these requirements for the same cost. Ruggedised machines do exist, but at a significant price premium.

The other option would be to ship that data back over fibre (common), but that also creates a dependency issue on the network and may require network hardware upgrades to deal with the big increase in h.264 encoded video traversing the network.

I'd love to see more of this happening, but it's an extremely conservative industry/market.

This exists in plenty of places. See the comment below yours.

I would like to see this as a temporary fix for when a signal goes down... toss a bolas with a camera + light that wraps around the cross-bar and can temporarily handle things. (Good luck getting it down!)

Had a friend who worked in Energy Efficiency for a U.S. state. It was a real struggle for them to get people to switch traffic lights over to LEDs from incandescents, even though the LEDs were superior - lot of risk-aversion.

It's a neat thought, and everyone who's watched a light turn red ahead of them just as they cleared a green has thought it before, but there is a serious downside. Urban stoplights should be optimized for the safety of non-motorists (pedestrians and cyclists) above all else. Giving cars more uninterrupted travel leads to higher speeds and more collisions. I'm happy that where I live there's more attention paid to traffic calming and bicycle infrastructure than motorist travel-time optimization.

>Giving cars more uninterrupted travel leads to higher speeds and more collisions

Not necessarily. In a well-timed system, cars that travel the posted speed limit are indirectly rewarded by not needing to stop. Going too fast or too slow would mean you would be out of sync with the timed system.

You can time traffic lights to encourage calmer traffic, by having "green waves" paced for bicycles, e.g. ~15 mph. Drivers should eventually clue in and slow down between lights since it just makes them wait longer.

Virginia currently uses smart traffic signals:

> Traffic-responsive signals vary the timing of the lights according to the amount of traffic. They use sensors to detect the number of vehicles on an approach. The time the light stays green adjusts to let as many drivers as possible through before the signal changes to respond to traffic coming from another direction.

Back when I lived in NoVa, the lights near me had this and it was awesome

Lots of lights have inductive loops to detect cars, but they are too close to the lights. The cars have already stopped by the time the loop detects them.

How does the Virginia system detect the cars?

The VA system detects how many cars are waiting to ensure that as many as possible make it through in a cycle (up to a limit), not detecting them as they are inbound to keep the light green.

That's a familiar system, but it requires a car to come to nearly a full stop before the sensor picks it up. This leads often to the lights alternating so out of phase it produces the maximum delay for both streets.

I sometimes get the feeling that traffic signals in my home town (Chicago) are designed to maximize ticket and speed camera revenue, rather than improve the flow.

I don't drive that much these days. When I did, I would occasionally get on the roads very early, when it seemed to me that the Washington, DC, traffic lights were well timed. Once the rush hour started, though, it was not possible to maintain the speeds for which the lights were timed. I suspect that this is the case many other places.

I wonder if, after a certain density of traffic, these advanced/well timed schemes are actually detrimental?

> I sometimes get the feeling that traffic signals in my home town (Chicago) are designed to maximize ticket and speed camera revenue, rather than improve the flow.

You get that feeling because it's true:

http://www.chicagotribune.com/news/watchdog/redlight/ct-yell...

The hidden cost of living in a bankrupt city (and state).

Startup opportunity: micro payments to change the light green in your direction ASAP. Voice-activated bidding with a floating fee that everyone sitting at the light can contribute towards.

There is an interesting side effect here. Autonomous learning.

Basically when the stoplights in my city switched to a 'three program' light (they have 'normal workday', 'normal non-work day' and 'commute' patterns available it means that people who drive "on autopilot" start having the wrong thing anticipated when they are sitting at the light, (for example expecting a left turn signal to go green before oncoming traffic gets the green, or expecting turn signals to activate before through traffic) and that leads to some interest effects. Most notably people "jumping" into traffic when their anticipation is incorrect.

The other thing that this does with non-native/less well trained drivers, is put them in lose-lose situations. I got out and talked to a guy who was sitting in a non-left lane (it was one lane over), waiting for the left turn light to activate. I explained to him (he barely spoke english) that from the lane he was sitting, the left turn light would NEVER activate and by sitting there while the light in front of him was green he was at risk of being hit by a car. He had no clue. It was very sad.

This seems like a nice idea but I actually think it would be terrible. The big discovery recent-ish is "Traffic Expands to Fill Available Road Space" [1]. A situation where traffic lights become smarter to keep traffic balanced between main streets and side streets and between different intersections of the city would just mean result in every single intersection of a given city being in grid-lock simultaneously rather than just some of them.

Already, I've notice that rush-hour lights seem to tuned to take longer than mid-day light - assume that's because any light-change is going to be inefficient (logically enough). But since rush is fated to total congestion anyway, the result just congestion plus tortuously long lights.

[1] http://www.culturechange.org/issue8/traffic%20expands.htm

Or, we could use roundabouts[0]

[0]: http://www.wsdot.wa.gov/Safety/roundabouts/benefits.htm

Roundabouts are OK but they don't work so well on multi-lane roads. The multi-lane roundabouts I've seen are confusing to most motorists and tend to slow to a crawl. The problem is there is limited space to change lanes once you are on the roundabout, so you have to enter in the proper lane based on where you want to exit. Inevitably there's someone who didn't use the proper lane and everything clogs up while they stop and wait for a gap in the next lane (which happens rarely because roundabouts promote continuous traffic flow).

They work fine here for 3 lane roads. The issue is high traffic from more than one direction simultaneously.

My country thinks otherwise. Roundabouts are great for low to medium traffic conditions but at high traffic they are a nightmare for any driver. The wait times can be excruciatingly long.

Highway commutes will be significantly streamlined when merges are done in a perfectly interlaced fashion with autonomous drivers. Anyone who has tried to get on the highway in the US during rush hour probably knows exactly what I am talking about.

Imagine driving on roads where all the vehicles followed the rules and you didn't have to pay attention and the road system worked with the vehicles to improve efficiency.

I would love to see this implemented in SF. Every day I drive I see one light turn green just as the next one turns red. You really can't do much worse.

Note that some cities optimize for driver inconvenience, to discourage driving through the city center.

If you want to minimize traffic, instead maximize the number of people transported per vehicle.

It would be even better to minimize the number of people transported.

Reply on news.ycombinator.com