Study: Traffic signals need a fourth color

[wanderer2575]

As more self-driving cars enter roadways around the world, many aspects of driving could change forever.  More electric vehicles (EVs) and autonomous vehicles (AV) are already leading engineers to implement electric roadways and city planners to design dedicated lanes for AVs -- and now, the humble traffic light is next for a makeover.

For the dawning age of the self-driving car, transportation engineers from North Carolina State University are proposing the addition of a fourth "white light" whose function would be to alert humans to simply "follow the car in front of them."

That's because to leverage the full potential of autonomous cars means to also embrace their ability to act as a kind of hive mind.  Because they're essentially computers on wheels, these cars can network with both the traffic light and other nearby self-driving cars to help ease traffic delays and increase fuel (or electricity) efficiency.

Popular Mechanics, 10/28/2025

https://www.popularmechanics.com/technology/infrastructure/a69178503/fourth-color-traffic-light-white-study/

[NE2]

April Fools is that way ---->

[kurumi]

Stomatopods have been advocating for this for years (where "advocating" means punching a hole in the aquarium glass)

[kalvado]

We discussed it a few years ago.
https://www.aaroads.com/forum/index.php?topic=32879

[Max Rockatansky]

So the takeaway is that Popular Mechanics has now also reached the point at which they are recycling three year old stories. 

[vdeane]

What happens if the human driver needs to go a different place than the car in front of them?  If they're going straight and you need to turn left, are you then stuck not getting to your destination the way you want to?  And what happens if a human does something unexpected?

[Dirt Roads]

What happens if the human driver needs to go a different place than the car in front of them?  If they're going straight and you need to turn left, are you then stuck not getting to your destination the way you want to?  And what happens if a human does something unexpected?

My understanding was that all left turns were signalized as segregated onto separate turn lanes at any four-color traffic signal.  Therefore, if a non-autonomous vehicle (ergo, a plain ole' car or truck) is following an autonomous vehicle and the rattle-trap gets a "white ball" then they are getting a protected movement to proceed right behind the autonomous vehicle.  In almost all cases, right turns are already protected when the through lane gets a "green ball". 

But the "white ball" eliminated all of timing to transition between phasing when autonomous vehicles having conflicting movements are being interleaved in the approach to the signal. 



One of the issues we have in the Personal Rapid Transit (PRT) industry is when the interlocking (ergo, an mechanically-controlled intersection) is quite long, a PRT car with a malfunctioning propulsion system might take too long to the clear the intersection when the wayside control system has cleared the next PRT car to proceed on a conflicting route.  The safety measures we take to protect what railroaders call the "end-of-train" effectively adds back all of that transition timing between conflicting routes.  You can liken this to a fast wayside-autonomous Tesla trying to make a left turn behind the path of a slow wayside-autonomous semi-truck plodding through an uphill intersection at 10 MPH.  My analysis indicates that a four-color traffic signal controller should only "check-in" autonomous vehicles that are the roughly the same length and approaching the intersection above a minimum speed.  But I am known to be overly cautious in these calculations.

[wanderer2575]

So the takeaway is that Popular Mechanics has now also reached the point at which they are recycling three year old stories. 

A thousand pardons.  I didn't see the topic and didn't know the idea has been around awhile.

At least I didn't bump a thread that's been dormant for three years, right?

[vdeane]

In almost all cases, right turns are already protected when the through lane gets a "green ball". 

Not when pedestrians are involved they aren't.  And if that's using a white signal, then there's the potential for left turns to be an issue, too.

[Max Rockatansky]

So the takeaway is that Popular Mechanics has now also reached the point at which they are recycling three year old stories. 

A thousand pardons.  I didn't see the topic and didn't know the idea has been around awhile.

At least I didn't bump a thread that's been dormant for three years, right?

Further in your defense Popular Mechanics did change the publish date of the article.  The only way we would know is that an older thread to the same article link existed. 

Besides, it isn't like searching for special threads is easy nowadays. 

[Dirt Roads]

analysis indicates that a four-color traffic signal controller should only "check-in" autonomous vehicles that are the roughly the same length and approaching the intersection above a minimum speed. 

BTW, this presents a logic quandary for the four-color traffic signal engineer.  The railroad methodology for a train that doesn't "check in" properly (or with the proper credentials) is to prevent that train from entering the interlocking.  However, the air traffic control approach assumes that you can't stop the "offending" airplane and you protect the other airplanes from entering the trailing airspace.  In the slow-truck example above, since the "offending" might be running fully autonomous it probably makes sense to let it keep running through the intersection and slap the conflicting movement with a "red-ball" and make it wait until that portion of the intersection is cleared.  But like the railroad methodology, you could slap the slow-truck with a "red ball" and let the next vehicle through (which may not be the conflicting movement that we've been talking about).  It's not obvious which methodology results in improved throughput.

[Dirt Roads]

In almost all cases, right turns are already protected when the through lane gets a "green ball".

Not when pedestrians are involved they aren't.  And if that's using a white signal, then there's the potential for left turns to be an issue, too.

Here's where the driverless train industry deviates from the autonomous vehicle industry.  In most systems, it is imperative to eliminate hazards such as people, animals, large tree branches, and other potential obstructions from the driverless train environment.  Most systems require a full-shutdown just to let somebody work in the same tunnel with a driverless train.  Just sayin'.

[Road Hog]

Why does it have to be white? Why not blue, or magenta, or black light like my deadhead cousin?

[kalvado]

Why does it have to be white? Why not blue, or magenta, or black light like my deadhead cousin?

Because if it was blue, you would be asking the same question except for "white" and "blue" switched around

[Rothman]

Shouldn't be white due to common red-green colorblindness, where current green lights can be confused with white.

[Scott5114]

Shouldn't be white due to common red-green colorblindness, where current green lights can be confused with white.

But that would require a tech bro to think about the ramifications of his own technology on people other than himself.

[Road Hog]

The only colors you need are Stop, Go and Go Very Fast.

[kphoger]

The only colors you need are Stop, Go and Go Very Fast.

I asked Google what the color "go" looks like, and the A.I. suggested it's probably yellow.

[PColumbus73]

Arrows are still a thing. Why do we need a white ball when we can replicate what Montreal does with R-Y-G-GA(Up/Left/Right)? A human driver would have an easier time something familiar than try to understand a new signal aspect (the white ball) which essentially carries the same meaning as an arrow.

Secondly, I would assume it would be the autonomous vehicles that need the white ball over the human drivers. Since the autonomous vehicles are more likely to be a hazard in the event of a malfunction due in part that the selling point is the lack of human interaction and intervention.

I would also think, assuming the white ball is meant for the autonomous vehicles, that the white ball would have some signal or frequency embedded in it to guide the vehicle through the intersection. Essentially overriding the car's onboard system in the event that the vehicle might get 'confused' by the intersection, issues with signs or lane markings, or some other malfunction.

[Dirt Roads]

Why does it have to be white? Why not blue, or magenta, or black light like my deadhead cousin?

Shouldn't be white due to common red-green colorblindness, where current green lights can be confused with white.

So-called "lunar white" (which is actually a bluish fresnel lens) has been used as the fourth color on many railroads for more than a century.  On the other hand, railroads tend to not permit someone that is color-blind from serving as locomotive engineers or signal maintainers (fortunately, I don't have sufficient color-blindness to have cancelled my career, but I was always worried that it would get worse).

Arrows are still a thing. Why do we need a white ball when we can replicate what Montreal does with R-Y-G-GA(Up/Left/Right)? A human driver would have an easier time something familiar than try to understand a new signal aspect (the white ball) which essentially carries the same meaning as an arrow.

Because the white aspect ball doesn't give the driver permission to enter the intersection; it only gives the driver permission to follow the automated vehicle ahead.  If the automated vehicle decides to stop in approach to the intersection, the driver following must also stop.  Worse, if the automated vehicle enters the intersection and turns right, the driver following that wants to go straight through on the white signal must stop.  But your point is well-taken; the purpose of the white signal is going to be quite confusing.

[1995hoo]

Considering how many people seem to find flashing yellow arrows confusing (I'm convinced my wife still doesn't understand them), I have to think introducing something totally new would be far more difficult for most people to understand.

[Rothman]

Why does it have to be white? Why not blue, or magenta, or black light like my deadhead cousin?

Shouldn't be white due to common red-green colorblindness, where current green lights can be confused with white.

So-called "lunar white" (which is actually a bluish fresnel lens) has been used as the fourth color on many railroads for more than a century.  On the other hand, railroads tend to not permit someone that is color-blind from serving as locomotive engineers or signal maintainers (fortunately, I don't have sufficient color-blindness to have cancelled my career, but I was always worried that it would get worse).

A lot of states still don't check car drivers for colorblindness...

[PColumbus73]

Arrows are still a thing. Why do we need a white ball when we can replicate what Montreal does with R-Y-G-GA(Up/Left/Right)? A human driver would have an easier time something familiar than try to understand a new signal aspect (the white ball) which essentially carries the same meaning as an arrow.

Because the white aspect ball doesn't give the driver permission to enter the intersection; it only gives the driver permission to follow the automated vehicle ahead.  If the automated vehicle decides to stop in approach to the intersection, the driver following must also stop.  Worse, if the automated vehicle enters the intersection and turns right, the driver following that wants to go straight through on the white signal must stop.  But your point is well-taken; the purpose of the white signal is going to be quite confusing.

I feel like the counterpoint is similar to when we (generically, not you specifically) were arguing about the 150 MPH AI Highways. It seems like we're trying to combine railroads with vehicular roads. Problem with vehicular roads being the inherent randomness of car traffic, roads, and cities.

I am having a difficult time conceptualizing the difference between a green ball (Go, permission to enter the intersection) and lunar white (follow the leader). Generally speaking, I feel like traffic management relying on drivers/vehicles to 'follow the leader' is not an ideal strategy. I also don't see how intersection control with autonomous vehicles cannot be achieved with the standard R-Y-G signal.

Assuming a human driver is following an autonomous vehicle that stopped in an apparently clear intersection, how is the human driver to assume that the autonomous vehicle is operating correctly? Following the autonomous vehicle would necessitate an understanding between the two beyond that of 'the autonomous vehicle has technology and is therefore smarter and more correct'.

[Dirt Roads]

Assuming a human driver is following an autonomous vehicle that stopped in an apparently clear intersection, how is the human driver to assume that the autonomous vehicle is operating correctly? Following the autonomous vehicle would necessitate an understanding between the two beyond that of 'the autonomous vehicle has technology and is therefore smarter and more correct'.

Pardon me, but I'm going to reverse your questions.  I'm a natural skeptic to these types of proposals, but always willing to give them a fair shake (and have been employed on occasion to dig deep into the technology and software). 

First, the premise of the N.C. State proposal (which was not a study) is that number of cars with some level of autonomous control features is approaching a percentage whereby intersections could be optimized further by taking advantage of the onboard technology.  The Popular Mechanics article made this commentary:

...to leverage the full potential of autonomous cars means to also embrace their ability to act as a kind of hive mind.

Thus, if everyone "follows the leader" through the intersection the opposing traffic presumably cannot "follow the leader" into the path of a conflicting route.  In my original analysis of this arrangement, I assumed that every single route has a queue area and its own separate R/Y/G/W signal heads with "red balls" for inherently conflicting movements, but I don't think that the N.C. State grad students had this in mind.  Thus, there is a possibility that all signals could be displaying a "white ball" all of the time.  Notwithstanding, here is my original concern posted in the other thread:

I have a concern about the use of a white aspect on traffic signals for this purpose.  Human drivers will soon learn that whenever one driver gets a "Follow-the-Leader" aspect, then everyone else can do the same (even if the traffic signal goes into a different phase and turns yellow, then red).  It is one thing that in New Jersey, six to eight cars will run the light after it turns red.  It is another thing when 15 to 20 cars decide to "Follow-the-Leader" to hop over to the next junction.  This is a sure-fire way to get into a gridlock situation somewhere down the line.

So my concern was that too many people would "trust" the technology.  Your concern that hardly anyone would trust the technology is also quite valid, but most modern-day traffic signals are already designed to detect vehicles waiting in the queue (ergo, you didn't go through the intersection even though you were permitted to "follow the leader").

I am having a difficult time conceptualizing the difference between a green ball (Go, permission to enter the intersection) and lunar white (follow the leader). Generally speaking, I feel like traffic management relying on drivers/vehicles to 'follow the leader' is not an ideal strategy.

In the worst case of playing "Follow the Leader", rather than having a green ball/green arrow for a particular movement (opposed by red ball/red arrows for all opposing movements), the traffic signal simply goes out of phase and everybody gets a "white ball" that gives everyone permission to follow the leader along the same route.  Once the traffic signal controller detects that some of the queues are starting to get loaded to capacity, it needs to go back to phased operation to clear out the queues.  Unfortunately, this requires that the traffic signal controller will also need to be capable of detecting whether there is a sufficient number of autonomous vehicles approaching the intersection in order to cancel out the R/Y/G phase and flip back to "follow the leader".  (That's not how I envisioned this all working, but it simplifies the picture).

Here's where you start to invoke something similar to railroad technology.  This either means that autonomous vehicles need to have the capability of "checking into" each traffic controller to report their operating status; or more expensive, the traffic controller needs to communicate back-and-forth with most of the autonomous vehicles approaching the intersection in order to set the system back into "follow the leader" mode.  Railroaders would call this "vehicle-to-wayside communications".

I also don't see how intersection control with autonomous vehicles cannot be achieved with the standard R-Y-G signal.

It certainly can, but the throughput cannot be significantly improved unless most of the autonomous vehicles are reprogrammed to ignore the red ball and the yellow ball and simply plow through the intersection when it thinks that there is a significant gap.  I can make that happen safely on a PRT system, but I have serious concerns about whether to let a Tesla drive through a red signal.

[PColumbus73]

We might have to put together an animation of how the system would work.

To me, it sounds like the designers proposing the 4-aspect signals and autonomous vehicles just assume they can platoon cars through an intersection in a nice, clean formation.

If we use New York City as an example, there's a car on Seventh Ave headed toward Lower Manhattan, behind it is a car going to Queens, and behind that is a car going to New Jersey. Assuming these cars are traveling in the same lane, eventually the cars going to Queens and New Jersey are going to have to turn somewhere. So, either by lane change or turn off, eventually one of those cars will create a gap in traffic. Now multiply that by several thousand for all the vehicles that travel through the New York metro. I'm not trying to dismiss the idea, but I think the idea sounds overly simplistic in relation to how traffic works.

This also assumes an international standardization of AI technology systems in vehicles. Can a system in a GM vehicle communicate with a Toyota, and so on?

Then, we mix in human drivers. Assuming the three cars on Seventh Ave are all autonomous and we add a fourth human driver wanting to go to Brooklyn. He gets a white 'follow the leader' signal, but in obeying the signal he misses his turn (or believes he's going to miss his turn) and the human driver gets stressed or confused. How can the system be designed not only to take advantage of AI systems in autonomous vehicles, but also accommodate human perception and reactions?