Autonomous Cars Hard Time Coping With Anti-Gridlock Laws, It’s Dicey Behavior

Dr. Lance Eliot, AI Insider

Image for post
Image for post
Gridlock is fought by anti-gridlock laws, but can driverless cars handle this

Don’t block the box.

If you aren’t familiar with that phrase, you likely have not yet been to a high-traffic locale. The rest of us know that it means staying out of an intersection in a manner to prevent others from being blocked by your presence, almost like playing musical chairs, but with cars, in an intersection, and it is a kind of driving task that seems simple to describe though manages to stymie a lot of drivers.

Gridlock, it’s a pain in the neck, or worse.

When I use the word “gridlock” it is meant to suggest that cars will at times enter into the intersection while the light is green and fail to make it fully across the intersection before the light turns red, ending up stranded in the intersection and serving to block traffic. You’ve undoubtedly experienced being blocked by cars that were momentarily stranded in an intersection. And if so, you were probably irked (pissed off!) that those drivers misjudged the situation and are blocking your fully legal efforts to get across the intersection.

Here in California, we are known for having been one of the first states to enact an anti-gridlock law that specifically prohibits the blocking of an intersection, doing so in 1987, and it reads as follows:

“A driver of a vehicle shall not enter an intersection or marked crosswalk unless there is sufficient space on the other side of the intersection or marked crosswalk to accommodate the vehicle driven without obstructing the through passage of vehicles from either side.” California Vehicle Code Section 22526.

The law itself seems rather self-evident and easy to understand. Well, of course you should not enter into an intersection unless you know that you can make it to the other side. Seems pretty simple. Basic driving 101, as they say. To make things even clearer, many locales refer to this as the “Don’t Block The Box” rule, wherein the intersection is considered to be a box as shaped by the four sides of the respective crosswalks. I suppose that the phrase “anti-gridlock” sounds more formalized and intricate, while the less pedantic notion of just don’t block the intersection box is much more forthright.

Blocking Intersections Is A Temptress

Why did the chicken attempt to cross the road? Because it was hoping to get to the other side, and if it couldn’t do so it figured that being part-way there was better than not any of the way there at all. I think that’s how the old joke goes, though maybe I’ve augmented it a bit. Each morning, I see chickens, uh, make that drivers, whom are seeking to get across the intersection. They can plainly see that there is traffic on the other side of the intersection. They can plainly see that the traffic on the other side is completely choked full and backed-up all the way to the crosswalk. There is no chance of squeezing into that morass. And yet, there are drivers that start into the intersection anyway.

Why do they do so?

One form of logic is that the impatient driver hopes that by the time the light goes red, the traffic on the other side will have moved forward, and therefore they will indeed safely and fully make it across the intersection. They are betting that the traffic up ahead will move on a timely basis to allow them to make it across. Often times these hopeful drivers make a lousy bet and it turns out the traffic up ahead stays put. This means that the driver becomes stranded in the intersection and will be blocking traffic that is trying to next move through the intersection from the perpendicular side.

The cars that are blocked by the interloper are likely to get rather riled-up about the situation. Even if you’ve gotten stuck in the middle of intersection previously, you likely have little sympathy for others that do so. You perhaps were even berated by other drivers when you were in the middle of the intersection, and so you might be inclined to berate other drivers for their similar transgression.

Sometimes the stranded car inspires other drivers to honk their horns at the driver. This used to be a common occurrence in many downtown cities that experience much gridlock. You would hear a continual stream of honking horns due to the continual pattern of cars stuck in the middle of intersections.

There are some drivers that will butt up against the tail-end of the traffic that is on the other side of the intersection. They often are completely covering the crosswalk. They are also likely somewhat protruding into the intersection via the rear portion of their car. This seems like a better spot to be than to be utterly in the middle of the intersection.

The Dance Of The Hidden Veils

What happens next is a dance of the partially gridlocked intersection.

Pedestrians using the crosswalk will try to walk either behind the stranded car or go in front of the stranded car.

For those pedestrians that opt to go in front of the stranded car, they are essentially blocking the car from being able to move forward if perchance the traffic moves during the time that those pedestrians are crossing the crosswalk. The driver will get frustrated at those pedestrians because the driver wants to extricate themselves from their stranded position and now the pesky pedestrians are in the way.

From the perspective of the cars that are trying to get through the intersection, and for which there is now a car sitting in the crosswalk and protruding into the intersection, you’ve got a kind of dance that arises there. The cars that are closest to the stranded car will need to judge whether they can make it safely past the posterior and remain in the lane of traffic. At times, those cars will not be able to make it in their own lane and thus need to decide whether to come to a stop, or whether to try and swing into the lane next to them.

If the driver opts to swing into the lane next to them, this can cause the drivers in that lane to then be confounded. Some of those cars might be moving very fast and figure that the blockage of the protruding car doesn’t directly impact them and so they are just going to zoom through the intersection. Meanwhile, the cars in the lane that is partially blocked are fishing around and trying to see how much they can swerve into that other lane without actually causing a crash.

It is a dangerously cascading and unsafe situation, all because of that driver that thought they could try to squeeze into the other side of the intersection but misjudged the matter.

There is the other variant of a car that gets stranded fully in the intersection and cannot make it to even the edge of the other side. In that case, other cars that now have a green light will try to flow around the stranded car. It is like having a big rock placed into the middle of a stream. The water tries to flow around the rock, either behind the rock or in front of the rock. Likewise, cars try to go to the right or left of the stranded car.

This is another highly dangerous situation. Any of those cars could hit the stranded car. Any of those cars could hit each other.

The driver in the stranded car might also try to move ahead, in spite of the cars flowing around it. Imagine s rock in a stream wherein the rock decides to move. This can be confusing to the other drivers.

So far, I’ve focused on just one car that might get stranded either in the intersection or at the edge of an intersection. There could be more than one car that gets into such a pickle. If you have a multitude of cars that all decide to try and get through the intersection and are unsuccessful, it is as though a dam have been built by an eager beaver in that intersection. The dam will hold-back traffic that is now wanting to use the intersection and legally has a green light to do so.

That’s when you get true gridlock.

Gridlock Is Bad And People Tend To Need Guidance To Avoid It

The basis for calling it gridlock is that most downtown areas are structured into a series of streets that resemble a grid. In your mind, think of a spreadsheet with rows and columns. You’ve got streets that intersect with each other on this grid. New York City is probably the most notable grid structured downtown area.

The grid of streets can get boxed-up by the fundamental aspect of blocking an intersection. If you block one intersection, it can cause the traffic coming into that intersection to get stopped. This can cascade to other intersections. If everyone opts to flood into the intersections and block them too, you effectively lock-up the entire grid and no one can move.

Before the advent of anti-gridlock laws, the gridlock situation would often go wildly out-of-control. It might start rather small, such as one intersection gets blocked, but the effect would spread like a rampant virus and quickly the entire grid would get overwhelmed. It is a nightmare to be stuck in it.

In fact, the traffic gridlock situation is a fascinating indicator of human behavior. If all the human drivers were completely cooperative, the mutual cooperation of politely keeping the intersection free would mean that the traffic would flow nicely. Mutual cooperation in this case begets mutual benefit. The trouble though is that we humans are endowed with a sense of “greed” and therefore what might be good overall is not necessarily as good for you in particular. You are seeking to maximize your own benefits, which might or might not coincide with maximizing it for everyone all-told.

Enforcement Saps Resources

One reason that these lawbreakers might not care about the long-arm of the law is that it is a difficult law to enforce.

Think about it. Let’s assume that a police officer has to be present to witness the act of your blocking the intersection. What are the odds of a police officer happening to be there when you make your untoward move? Probably low odds that a police officer will be nearby (and, if you are being intentional and devious, presumably you were on the look for the cops and if you didn’t see a police officer you then took your chance at blocking the intersection).

In general, and especially for a bold scofflaw, they will likely decide it is “better” to try and get stuck in the middle of the intersection or make it to an edge, presumably allowing them to reduce their travel time and get to their destination sooner, for them alone, and that since the likelihood of getting caught seems relatively low, and even if they do get caught the nature of the penalty will likely be low, it all adds up to a go-for-it mentality.

Some locales have mounted intersection cameras to catch these scofflaws. This helps to deal with the aspect of not readily having a police officer around to see the act and also it reduces the time and disruption to traffic of an officer issuing a ticket. On the downside, the cost of setting up the cameras and its actual role in being able to discourage the scofflaws all tends to make it not particularly viable as a solution to preventing the gridlock instigators (they also have at times tried to fight the law, when caught, by claiming issues associated with the cameras).

AI Self-Driving Cars And Dealing With Anti-Gridlock

What does this have to do with AI self-driving cars?

At the Cybernetic AI Self-Driving Car Institute, we are developing AI software for self-driving driverless autonomous cars. One core aspect involves the AI being able to contend with the potential of gridlock situations.

Allow me to elaborate.

I’d like to first clarify and introduce the notion that there are varying levels of AI self-driving cars. The topmost level is considered Level 5. A Level 5 self-driving car is one that is being driven by the AI and there is no human driver involved.

For self-driving cars less than a Level 5, there must be a human driver present in the car. The human driver is currently considered the responsible party for the acts of the car. The AI and the human driver are co-sharing the driving task.

Another key aspect of AI self-driving cars is that they will be driving on our roadways in the midst of human driven cars too.

Returning to the topic of gridlocks, here’s what we need to consider for the AI of an AI self-driving car:

  • Detect that a gridlock exists when it is there,
  • Determine how to best maneuver within a partially gridlocked intersection,
  • Deal with stranded cars that are sitting in a crosswalk,
  • Be on the watch for cars behind you that might get panicked due to a brewing gridlock,
  • Avoid getting stuck in the middle of an intersection as part of a gridlock,
  • Don’t instigate the starting of a gridlock,
  • Aid other cars as to dealing with the gridlock if feasible to do so
  • Other

There are some AI developers that would assert that there is nothing special for the AI to do about a gridlock situation. For them, the normal everyday operation of the self-driving car should be sufficient for dealing with gridlocks. If the AI can drive a self-driving car and navigate the roadways, it is perceived by the AI developers that there is nothing extraordinary about the gridlock circumstances and therefore no special attention is needed.

We disagree.

The gridlock circumstance is a special case and requires a specialized module or capability to be handled.

Dealing With The Edge Case

It might well be considered an “edge” problem by some, namely that it is not necessarily at the core of what the AI needs to do to drive a car minimally. There are though so many driving aspects that are entangled with the capability of dealing with gridlocks that we argue it is not readily classified as an edge problem and needs to be placed higher up in the priority of aspects needing to be dealt with by the AI.

Some AI developers use the piped piper approach of having the AI simply follow a car that is ahead of the self-driving car. This is not a wise move necessarily for the gridlock situation. Just because the car ahead of you opts to go into the intersection does not mean you should too.

You’ve likely seen many times that cars sit at the intersection waiting to rush across it, the moment that there is a spot open on the other side. Typically, one car at a time opts to make a dash across the intersection. It is as though a door opens momentarily on the other side and so one car can make it into that door. The door then closes for a moment. If the door opens again, another car makes the dash.

If you blindly follow a car that is making the dash to the other side, you are likely going to end-up starting the gridlock because the other car will make it but you won’t. That’s why many times you’ll see a gridlock start to form. One or more cars tried to follow one lead car, and the lead car made it across while the other cars did not. Those other cars are now stranded and panicked.

So, the pied piper approach won’t cut it. The AI needs to be looking further ahead and trying to predict whether or not there is a viable opening for the AI self-driving car to make it fully across the intersection.

I am sure there are some AI developers will say that the problem is the human driver. If we had only AI self-driving cars, they would all be civil toward each other and not cause any blocking of any intersections. Furthermore, they could use their V2V (vehicle-to-vehicle) electronic communications to ensure they coordinated their efforts. If somehow an AI self-driving car did get stranded in an intersection, it would merely use V2V to forewarn all other nearby self-driving cars.

As I mentioned earlier, we are not going to have the Utopian world of only AI self-driving cars, certainly not for a very long time from now. Thus, cross out the “there won’t ever be gridlock” notion for the foreseeable future.

In terms of identifying traffic situations that might include the emergence of a gridlock, it is possible via analyzing large-scale datasets of traffic data to be able to more readily spot such moments. Using Machine Learning (ML) and deep learning, usually consisting of Artificial Neural Networks (ANN), you can deeply analyze thousands upon thousands of traffic situations and train the neural network to be able to identify the potential gridlock aspects. When a driving situation begins to arise, the ML can potentially participate in making the detection of it.

Context Matters Significantly

The context of the detected driving situation will then shape what action the AI self-driving car should next take.

It could be that the AI will opt to proceed cautiously and not get stuck in the middle of the intersection. Or, it could be that the AI will opt to wait it out and try to step into the matter after some iterations of the traffic signal, hopefully without further exasperating the gridlock.

The AI needs to consider the possibility too of escaping the situation altogether, taking a different route to avoid the possible gridlock location.

Furthermore, you need to include the factor of the other cars nearby to you and what actions they might take.

Generally, we assert that the detection and contending with gridlocks should be a core part of the AI’s capabilities for driving a self-driving car. It involves sophisticated driving tactics and cannot be ignored, nor can it be handled by formulaic expressions.

Gridlock detection and maneuvering are vital to an AI self-driving car being able to drive safely in real-world driving situations.


Traffic gridlocks. We all abhor them.

Yet, our very own actions can spur them.

Even if your locale does not have laws that specifically ban gridlocks, it is likely that if gridlock begins to grow, there will be a backlash against those drivers that spark them. An AI self-driving car needs to be an active participant in aiding anti-gridlock actions. Just like the old question about chickens, we need to ask, why did the AI self-driving car try to cross the road?

The answer ought to be because it ascertained that it was safe to do so and accomplished the feat like a champ.

For free podcast of this story, visit:

The podcasts are also available on Spotify, iTunes, iHeartRadio, etc.

More info about AI self-driving cars, see:

To follow Lance Eliot on Twitter: @LanceEliot

For my blog, see:

Copyright © 2019 Dr. Lance B. Eliot

Written by

Dr. Lance B. Eliot is a renowned global expert on AI, Stanford Fellow at Stanford University, was a professor at USC, headed an AI Lab, top exec at a major VC.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store