By Dr. Lance B. Eliot, the AI Trends Insider
Toll roads, love them or hate them.
For many years, the west coast of the United States was proud to say that toll roads were verboten. Sure, there might be one here or there, but otherwise the west coasters believed that the prevalence of toll roads on the east coast was abysmal. Over my dead body would we have toll roads, exclaimed many drivers in the Los Angeles and surrounding areas. Barbaric, was the apt description for toll roads. This was California, the land that made the word “freeway” come to popularity and did so because it was indeed a “free way.”
With complaints about clogged freeways and rising commute times, the tollroad concept began to become a tolerated topic. At first, tolls were reserved for special circumstances like a bridge crossing. Even there, it was initially a toll in both directions, but the cries of foul were so large that things were changed to charge a toll in just one direction. Then, some brand new freeways were built and for those the notion of having a toll was introduced. This seemed to make sense. Rather than upset the status quo and put a toll on an existing freeway, one that was considered free, instead plop a toll onto a new freeway. There weren’t any traditions on the use of the new freeway. Starting it with a toll established a new tradition for that particular freeway. It also seemed sensible to the general public since they assumed that the cost to build the new freeway had to be covered somehow and a toll was the reasonable way to do so.
Whichever way you cut it, toll roads are now everywhere, east coast by tradition and west coast by more recent adoption. In this modern age, I know that you are thinking that handling a toll is pretty easy. Just get the cars to have some kind of transponder and when the car drives up to a toll position, the toll system can register the transponder and charge the person for their use of the toll road. No need to slow down or stop the car. No need to dig in the car for loose change to pay for the toll. No need to carry on conversation with a toll taker that has nothing to do other than take tolls, give change, and talk about the weather.
Well, we aren’t yet living in a world with transponders on every car. As a result, most of the toll roads still need to have a means to let toll payment occur via cash. On some toll roads, they charge you as you exit the toll road. At the exit position are buckets that you place cash into. The bucket collects the cash, counts it, and then either opens a gate or flashes a signal to say you’ve paid. If you try to skip or cheat it, the system takes a photo of you and your license plate. The next thing you know, there’s a ticket being mailed to you, or at least to the registered owner of the car.
For entry into a toll road, most still have a human toll taker. The human stands in a guard shack. A car drives up to the guard shack. The driver hands cash to the toll taker. The toll taker receives the cash, counts it, and provides back any needed change. The toll taker then either opens a gate, or sets a signal that lights up to say that you can proceed. The toll taker is supposed to work fast. Cars are supposed to be ready to make payment. It should all happen in an orderly manner. Efficient payment and green lighting of cars to move ahead onto the toll road.
The reality is that often there is utter chaos at the toll gates. Especially you see turmoil during rush hour. You’ve got cars driven by drivers that aren’t familiar with the terrain and the toll gate, and so they don’t know which lane to get into. There are some lanes reserved for transponder payers, other lanes for cash only, and some lanes for both cash and transponder. When I went across the Oakland Bay Bridge the other day, I noticed that another part of the confusion was how much the toll was. There were numerous signs and each was like one of those contracts that has small print and you need to be a lawyer to figure out what to do.
In the case of the Oakland Bay Bridge, the signs said the price was $6 if it was a weekday between certain daytime hours, or $5 if it was other hours during the daytime, and $4 if it was weekends but only with some added provisions. There were other signs about the number of axles on the vehicle and how much you needed to pay per axle. I could see some of the drivers being quite perplexed and unable to figure out beforehand how much the cost would be. They tended to just hand over a ten-dollar bill to the tolltaker and hope that the toll taker would be honest and return the right amount of change for the actual toll required.
Why all this discussion about toll roads?
It’s because a self-driving car has to contend with toll roads. For a true self-driving car, which I define as a Level 5 (see my column on the Richter scale for self-driving cars), the self-driving car should be able to figure out what to do. The main aspect to figure out is usually the entry point and the exit point. Once a self-driving car is actually on the toll road, the road is about the same as any other road. There’s nothing particularly special about the toll road itself. Just more asphalt and it cost you a few bucks to have the privilege to drive on it.
At the Cybernetic Self-Driving Car Institute, we are developing AI software that aids the self-driving car in terms of traversal during the entry into the toll road and the exit from the toll road. This is a kind of “edge problem” for self-driving cars, meaning it is not a core aspect, it only comes up on occasion, but that it is something that ultimately a self-driving car needs to be able solve and carry out (see my column on edge problems for self-driving cars).
Let’s walk through the fundamentals of how the AI system handles the tollroad situation.
Using maps, the AI would already be likely aware that a toll entry is coming up. That being the case, it could also let the occupant of the self-driving car know that a toll is nearing. The occupant could decide to try and have the self-driving car navigate a different path to avoid dealing with the toll and the toll road. You’ve likely seen this option on your existing GPS system. The GPS will let you know that the path requires a toll, and you can select an alternative path if you wish. This interaction with the occupants of a self-driving car would be undertaken via in-car commands (see my column about in-car commands for self-driving cars).
Suppose the AI system doesn’t have a map, or has a map that is faulty or outdated and does not indicate that a toll road is ahead. In which case, the AI system has to detect that a toll road is upcoming, doing so without the use of a map. One way to detect a toll road is upcoming would be by reading the road signs that usually precede a toll road. (See my column about reading of road signs by self-driving cars.)
When the self-driving car gets near to the entrance of the toll road, this is usually where the heavy lifting takes place. There is going to be a lot of stimuli that needs to be collected, analyzed, and used to make crucial and real-time decisions.
In the model of a toll entrance with toll guard shacks and multiple lanes, which is a prevalent enough model to warrant its own AI capability, the self-driving car needs to collect the radar, LIDAR, camera data and figure out that there are guard shacks, there are lanes, there are signs about what to do. The AI system examines the images from the camera to ascertain which lane has which purpose. Changing lanes in this situation can be dicey.
I say this because you’ve got to keep in mind that the self-driving car is going to be mixing with human driven cars. Human drivers are notorious for rapid lane changes in these situations. They also tend to get really close to other cars. They tend to be rude to other cars.They tend to play chicken with other cars. It’s a cruel world there at the entrance to the gates of the tollroad.
There are some self-driving car makers that are saying that the toll road problem is not a problem, simply because they will hand the driving over to a human driver in the car, and let that human driver deal with the few minutes involved in traversal at the toll gate entrance. This might be OK for the levels less than a 5, but it is not allowed for a Level 5. A level 5 car is supposed to be able to be driven by the AI in whatever manner a human could drive the car. In this case, a human could traverse the entrance area, and thus so should a Level 5 self-driving car.
Another reason that some self-driving car makers claim this is a non-problem is by saying that any self-respecting self-driving car will already have a transponder or at least some method of doing the automatic payment for the toll. First, I disagree that it is axiomatic that all self-driving cars will be equipped with a transponder for the paying of tolls. Even if it came with a transponder, there are numerous different kinds of transponders and one that works for toll roads in say Los Angeles does not necessarily work for paying tolls in San Francisco. Until we have an across-the-board standard transponder for all toll roads everywhere, you are out of luck about thinking that the self-driving car will be able to simply have a universal transponder in it.
Of course, instead of a transponder, it could be that a self-driving car might have other means to deal with paying the toll. For example, if the self-driving car has some kind of electronic communication capability such as using the Internet, it could presumably pay the toll via electronic means (see for example my column on the use of blockchain for self-driving cars). Once again though, the electronic communication needs to be pre-established and if you are driving across the country, and maybe you set up some means to pay electronically in California, but you are now in Arizona, the electronic communication might not do you any good there.
Even if you magically had either a transponder or some other electronic communications, the self-driving car still needs to figure out which lane to use. I used to think that all toll gates had the transponders lanes to the left and the cash lanes to the right. I’ve since then discovered that it seems much more randomly arranged. I’ve seen it all. There have been transponder lanes to the right, and cash lanes to the left. There are truck-only lanes and car-only lanes. There are mixed lanes that allow both transponders and cash, and sometimes are truck-only or car-only, or are for both trucks and cars. It’s a free-for-all.
So, the self-driving car needs to scan for roadway signs about how to handle the toll road entry. Interpreting these signs can be tricky. As I said earlier, it can be tricky for humans, and so it is likely then also tricky for the AI. There is also not an obvious indication by simply inspecting the line-up of the vehicles. In other words, you could detect that trucks are all in one lane, and deduce that it probably is a truck-only lane. Thus, at least you could eliminate that lane for a self-driving car. Trying to ascertain whether the cars ahead of you have a transponder or not, there’s no clear cut visual clue. It used to be that a transponder was bulky and at times visible to the naked eye. Nowadays they are much subtler and often hidden on a car.
Another approach to figuring out the right lane would be to do V2V (vehicle-to-vehicle communication). Your self-driving car could communicate with a fellow self-driving car that is nearby. It might ask the other self-driving car as to which lane should be used. The other self-driving car might know, perhaps it has used this toll road before and “learned” how to traverse it. The V2V would allow the learned self-driving car to bring the other self-driving car up-to-speed about what to do at the toll entrance. We are still pretty far from having V2V, and certainly not on any widespread basis, so this is a low odds approach currently.
You might say that another approach would be to “learn” from crowdsourcing. By this, I mean that the self-driving car might be connected via electronic communication with a server being maintained by the self-driving car maker. The server might be a collective wisdom from all of the self-driving cars that the car maker has put onto the road. The self-driving car approaching the toll road could inquire of the collective data on the server as to what to do when it gets to the toll road. This though requires that an electronic communication be available, and that the server would actually have something useful to say about traversing the toll road. Another low odds approach currently.
Looks like we’ll need to use brute force. The AI will have to figure out by the signage and by the behavior of other vehicles as to which lane to get into. It needs to then moderate the speed of the self-driving car as appropriate for the toll gate entrance configuration. Usually, you tone down your speed as you get close to the toll shack. You then come to a complete stop at the tollshack.
At this point, the self-driving car has presumably alerted the human occupant that a toll needs to be paid. The self-driving car has positioned the car at a place that would allow for payment. The human in the toll shack could have already been imaged scanned and the self-driving car can do a pretty good job of lining up the car window to where the human is, allowing the human occupant to readily reach out and provide the needed cash.
I know that it might seem silly that we’ve had all this AI figure out where to place the self-driving car so that then a human can pay the other human some cash. Those of you that abhor the use of human labor when automation can suffice, will certainly be disturbed that this last step involves a human to human transaction. All I can say is that I agree that eventually this will no longer be the case. We should ultimately have it all be electronic.
A smarmy person might say that we should develop a robotic arm that could reach out of the car and pay the cash toll. An even smarmier person could say that we should develop a robot arm for the toll taker. You’d then have one robot arm that hands over cash to another robot arm. This, you have to admit, is even more beguiling than the scenario of one human arm reaching out to pay to another human arm. One would think that if we went to the trouble to use robotics arms, it would by then be the case that no cash needed to be physically exchanged and it would be all electronically handled.
Assuming that the cash payment is made, the AI then would detect that the gate is opened or that a green light or something has signaled that the self-driving car can proceed. At this juncture, the self-driving car can switch into a more traditional driving mode, since we are now getting into the toll road and as I said before, this part is pretty routine.
The only twist though is that often when a car first passes past the toll gate, other cars are doing likewise. This can create a crazy situation of cars trying to accelerate rapidly and also do wild lane changes. You’ve probably seen this before. It is as though human drivers think they are at a Nascar race and that someone has shot off the starter gun to get going. The self-driving car needs to realize that this is a common aspect and be wary of other cars that might tend to threaten the self-driving car. By threatening, I mean that the other cars might be very aggressively driven and cut off the self-driving car.
The exit situation is generally similar. The AI needs to be looking for signage about whether a payment is needed at the exit point. The self-driving car needs to detect the exit point and scan for the payment location, if there is one. The AI needs to alert the human occupant about an upcoming payment. The self-driving car then needs to come to a stop at whatever placement is appropriate. The human occupant pays and the self-driving car needs to detect that payment was made and that it now okay for the self-driving car to proceed.
I have one suggestion for everyone. Get rid of tolls, and we’d not need to have self-driving cars that knew how to traverse the entrance and exits of tollroads.
Now, I realize that some will complain at me and say that tolls aren’t the problem. It’s what we do with the money. It’s that the money is needed to fund our transportation infrastructure. Yes, I get it. Didn’t mean to start a whole sociopolitical debate.
Meanwhile, this is an interesting edge problem for self-driving cars. It not only solves an actual driving problem, it also showcases how subtle the human driving task can be, and yet so complex. Namely, the act of traversing a toll gate for entering and exiting a toll road takes a lot more “intelligence” than it might seem at first glance. Watch any novice teenage driver try to do this for the first time, and you’ll witness first-hand how difficult it can be. Anyway, can I say down with tolls, or will that make some readers get upset at me?
This content is original to AI Trends.