Brad Templeton Home
Traffic Congestion & Capacity
Traffic congestion is another one of the banes of human driving, and in particular of uncoordinated driving. The NHTSA reports the average person in the U.S. wastes 41 hours a year (a full work week) in congestion and 8 billion gallons of fuel is burned because of it.
Robocars are still using regular roads, and will be subject to congestion and capacity constraints of the roads. Roads are expensive and in many cities there is not room to build more roads. Will robocars just end up waiting in traffic jams?
Some congestion is caused by accidents and safer robocars would do a lot to eliminate that. Much congestion, though, is just caused by both peculiar human driving patterns and the fact that too many people try to use the same road at the same time. This is both the "rush hour" phenomenon, and any rush hour has its own localized super-peaks in different regions. Many cities with "grid" streets could actually handle far more traffic than they do now if they spread it more evenly over their grid. Traffic lights and other traffic modifiers (like stop signs, transit, congestion charging, one way streets and the rules of the road) have a lot to do with congestion.
There are whole books on congestion and it's an active field of academic study. At the introductory level, I recommend Traffic: Why we drive the way we do which also has a section on robocars. But most of the thinking in traffic management is not based around robocars, or even the potential of technologies that are coming soon.
There are also many efforts in trying to make "smart roads" for human driven cars, under the umbrella name of "Intelligent Transportation Systems." Some of those efforts will apply as much, if not more, to robocars. Indeed, for the best results, we want a world where even human-driven vehicles almost universally carry a networked navigation device -- in other words a smartphone or newer car.
At the core, half of congestion is caused by basic lack of capacity, and a further 1/4 by accidents. Solutions to these things are possible.
First let's look at some of the simpler ways that robocars might seriously reduce congestion -- and thus provide faster trips with few starts and stops.
Better traffic management
There are a number of other technologies which can do a lot to reduce congestion which can also be used by human drivers as well or almost as well as robocars can use them. Light timing, noted above, may be one of them. In fact, today most efforts you will find are in this "Intelligent Transportation Systems" (ITS) or "Smart Roads" field.
The biggest current ideas in today's congestion management are highway metering lights and congestion charging in crowded downtowns. Metering lights limit the flow onto a highway, and while they cause initial anger -- who wants to have to get into a slow line to get onto a highway that appears to be moving smoothly -- people have come to learn that the reason the highway is moving smoothly is that the metering is taking place. It is generally always the case that the short wait at the metering lights assures a much shorter trip overall.
Congestion charging involves charging a monetary fee to enter or be in a highly congested zone, such as a central business district during rush hour. The charge scares cars away, and makes people carpool or use transit or even bike and walk. This seems to work, though it has many privacy concerns, and the initial way it has been implemented (with licence plate cameras and electronic tag readers at the borders of the zone) makes it expensive to adjust the borders. There are other, simpler methods of doing this which don't affect privacy as much, such as charges to enter and leave parking lots during rush hour.
Toll roads can be a form of congestion charging. Indeed, one could argue that carpool lanes, which sometimes reduce congestion, put a cost of sorts on access. Fancy metering and charging systems can also potentially offer more access or free access to carpools. Nowadays, carpool lanes are being converted to "high occupancy or toll" lanes, known as managed lanes, to make sure they reduce congestion, rather than increase it as they sometimes do.
Routing based on congestion predictions
Today many drivers rely on GPS based mapping systems to help them route trips. These systems are starting to be more traffic aware. They download live traffic data from central computer systems, and they even contribute by getting reports on progress of cars with GPSs and the ability to send data to the network. This is going to grow until we have available real-time reports on speed and congestion for every main block of a city and even every lane of a street or highway, as well as historical data on typical speeds and travel times at particular times of day, or under particular traffic circumstances.
Over time this will get very smart, so that it can not just tell what the traffic is like now, but even predict what it will be like. Systems will learn that if one highway gums up, another highway gums up later, or that sideroads become clogged. It's not out of the question that the system could will model every car or group of cars on the road, combined with knowledge of traffic light schedules and the planned routes of some of the cars in order to predict future congestion with greater and greater accuracy as the prediction time gets closer.
Nav systems are also starting to use this data to calculate optimum routes that avoid traffic congestion. The car of the near future (human or robot) will be able to plan routes based on predictions, and choose streets that will be less congested.
Without any cooperation between the cars, this will naturally cause vehicles to spread out over different roads. Why take the predicted congested road when a another is predicted clear? Some roads will remain more popular of course, but bunching done in ignorance of patterns should lessen.
Reservations for use of road
It will also be possible, once a car's navigation system has picked a good, low congestion route for the trip, that the system tell the city computer, or block-control computers along that route, what it has chosen. This knowledge can help the computers predicting traffic know where cars plan to go.
The system could go further and allow the car to ask for a "reservation" to use various regions of road at approximate points in time. The more accurate the vehicle is in planning its trip and the more accurate the congestion data, the more precise the reservation time can be. In addition, as the car gets further along its route, the accuracy of the prediction would improve. It might at first ask for a reservation to be on the street to its office any time between 8:30 and 8:50, but as it got closer and closer, it would refine that to an exact minute and eventually an exact second when just a block away.
Once many cars have asked for a reservation the system can refuse or discourage further reservations. It could even get to the the point that going down some roads without such a reservation would result in a ticket, and abandoning a reservation could result in a cost.
This would be somewhat akin to metering lights, but done on a road segment by road segment basis. Roads would rarely become congested because only a limited number of cars would be given reservations. Others would delay their trips (take them later) or preferably take reservations for other roads. People with no choice on a route (for example, it is the block which contains their destination) would of course always need to get their reservation, and might have bought or been granted premium access on that particular block.
There are some research problems here of course. People must not, for example, stop in the middle of the street or cause congestion because they arrive too early for their reservation. If they will be too late for their reservation, the system needs to be informed as much in advance as possible, for if a vehicle has to wait to avoid causing congestion (as the cars in a metering line wait) this waiting has to be done in a place that can handle it. On-ramps work fine for this, but city streets do not.
The other way to solve this is to go beyond free reservations to ones that cost money, and use bidding.
It should be noted that academics have designed systems for very precise reservations of road space for use by predictable vehicles. It has been shown that a traffic intersection can be managed with no stopping or lights through reservations. In such a system, cars reserve a trip through the intersection, and the intersection control computer of course arranges so those trips never put two cars in the same place at the same time. Instead cars slow and speed slightly as they approach the intersection to make sure they move exactly in their reserved slot. In the meantime, cars coming the other way are also zooming through in what looks incredibly scary. It has been shown that such a system can handle fairly large numbers of cars (if they are assured to obey their reservations) with quite minor average delays, compared to stop lights which add very high average delays to trips that go through them. Here's a link to a java demonstration. This requires 100% robocars, so it is fairly distant in the future.
In some cities, there can also be pedestrian congestion. One notable city is New York, where pedestrians often ignore walk/don't-walk signals and cross anywhere. A limited amount of that is tolerable, especially during low traffic. Because robocars will always yield to pedestrians, it could actually encourage more jaywalking. This could reach a level where the constant need to slow down for pedestrians causes road congestion, and you can't meter the pedestrians, only make rough predictions on their numbers.
It is an unresolved question whether society will just tolerate clogged roads due to random jaywalking, or if it will decide to try to cut back on it. Methods are readily available -- since robocars equipped with cameras can photograph anybody they have to brake for -- though some of these solutions have privacy concerns and may push too much rigidity onto pedestrians. We don't want a world where pedestrians and cyclists must move as rigidly and obediently as the robocars, but we might push for it at rush hour, while allowing more free movement off-peak.
Payment for road use
Congestion charging in effect charges for road use during congested periods. Advanced systems could break that down as minutely as desired, even to the square-foot-second (ie. exclusive use of a particular square foot of road at a particular second.)
This could be done manually, where the city (or even private road owners) setting prices to rent pieces of road, and adjusting those prices until the roads are well used but not overused. It is also possible to use bidding to set the prices, including "Dutch (uniform price) auction" bidding and 2nd price auction bidding. In a Dutch auction, you might have 50 cars seeking to use a road with the capacity for 30. They all bid, and the top 30 win and pay the 30th highest bid price and get to use the road. The others must now bid instead for an alternate route, or use of that route at an alternate time.
A 2nd price auction (which also can be combined with the Dutch auction approach) has interesting consequences if you allow bids of zero. In a second price auction, the winner pays the price that the 2nd place bidder offered. (This is in fact how eBay auctions work, with the minor change that the winner pays the 2nd highest offer plus a small increment.) In a Dutch 2nd price auction all winners would pay the price bid by the highest bidding loser. If bids of zero are allowed, then if there is just one bidder who bids more than zero, that party wins but pays zero (the 2nd highest bid.) This can be a good way to use resources which, like a slot on a road, evaporate when they expire, and which are already paid for by taxes.
Under this system, in uncontested times, all the reservations are free. As soon as more than the maximum number of cars bid, however, we suddenly get a small cost, and it goes up as the bidders increase their offers and/or grow in number. Money only changes hands when there is actual contention.
All of this bidding would be done by computers based on simple guidance from their owners on budgets and time constraints. If there is enough road capacity to handle the traffic, most if not all bidders would be paying zero, or modest amounts to get exactly the route they want when they want if there is contention. Many will happily wait to another time or take another route to get it free or cheaper.
It is also possible to simply charge for all use of roads, and to entirely pay for the roads this way, rather than through taxes. In such a system there would be a minimum bid even for uncongested use of the roads, and people would end up paying a modest amount every time they drove. That price would ideally match the real cost of the roads, and the average person would pay no more than they pay today through other mechanisms like gasoline taxes and other taxes. In other words a few cents a mile.
From a commercial standpoint the price might vary based on how much of the road you take up, so that larger vehicles pay more. It might also depend on the weight of your vehicle since heavier vehicles cause the vast majority of road damage. And of course it could depend on when you use the road, in particular if you want to use it at more high-demand and high-congestion times.
The price could also be set to meet social goals, since the roads are right now almost entirely government owned. Aside from reducing congestion, the price could encourage some forms of transit or align with goals of urban planning. Since current forms of transit are not as energy efficient as light electric robocars, it may not be good to encourage them.
(It should be noted that long term when roads do compete with other methods of carrying vehicles, it makes more sense for users of the roads to bear their real cost and users of the other methods -- rails or guideways -- bear their real cost.)
Politically, charging for roads is difficult: Even congestion charging. People feel they paid for roads already in their taxes, and even when this is not true, pay-per-use roads are viewed as elitist, giving quick trips to the wealthy and making the poor wait. As such, alternate systems might be considered, such as a lottery for slots on congested roads. The lottery might be conducted a day in advance, so you would know the night before if you were going to get a perfect commute, or would have to leave early or late to get your slot on a freely-moving road.
Combining it all
With more cars per square foot of road, no collapse into congestion through driver irrationality and balancing of load to avoid getting close to stop-and-go collapse points, the result could be a truly immense increase in the capacity of existing roads to handle people. With 3x from smaller vehicles, 2x from closer spacing and all the other factors it's not hard to imagine as much as a 10x (1000%) increase in road capacity.
Such a huge increase is important, because urban planners have learned that building new road capacity often only temporarily relieves congestion. Within a few years, people start making use of the new capacity, and changing their commutes and the locations of their homes based on it. The new lanes are soon clogged. This is known as "induced demand." Working properly, fully metered roads should rarely get clogged outside of accidents. Greater demand just means people either have to shift when they drive, pay more, or wait longer to get access to them.
Urban planners will be able to experiment with how to allocate access to roads when there is too much demand. They could use waiting lots (like today's metering lights and on-ramps) or they could use congestion charging. They could also use seniority, so that those who have lived a long time along the route get to access it but newcomers must wait or by access from a departing local -- this would discourage development along routes which exceeds the capacity of the routes, as new arrivals would be strongly discouraged or pushed to use other routes and methods. (Politically this would be pretty "interesting.")
But it's just possible that a 10x increase in capacity might well be "enough" so that there is enough capacity for all. After all, we built our roads to try to supply the irrational and uncoordinated actions of human drivers, and while they are stretched beyond capacity in many places, they are not stretched 10x beyond capacity. Only time will tell on this question, and it depends in turn on what urban densities people settle upon.
The result can be a trip which matches a train in a private right of way. Your vehicle might wait a short time until it can be assured of reservations for a good route, than slowly accelerate you to enter that route. It would rarely hit a red light and probably avoid stop signs or go through ones that it has a reservation to blow through. As such it will rarely need to stop before gently slowing down at your destination, quickly and smoothly. You were probably reading your book and didn't even notice.
Empty vehicle moves
Robocars present one risk of increased congestion, because they allow vehicles to move while empty. Indeed, this is one of their great benefits, and enables taxi services and car sharing (which in turn reduce need for parking,) as well as self-refueling, road clearing and other good things.
Empty vehicles can increase congestion. The alternative to empty vehicle moves is just to have more vehicles available. For people with short commutes, the same car could carry several different people during a rush hour, going back and forth. In many cases, the reverse direction is not congested, so the empty vehicle move is not a congestion problem. In the situation described above, where streets are reversed, capacity is removed from the reverse-direction which increases the chance of congestion there.
A few empty vehicle moves will be in the commute direction, mainly some delivery of vehicles that commuters will take further. They won't all come from the anticommute path. In some cities, there also isn't nearly so much of an anticommute path -- people are going both ways. Empty cars are often not in any hurry, and can take slower or longer routes to avoid causing congestion.
Empty vehicle moves save on the cost of vehicles, but cost energy and congestion. In the end, especially if travel on congested streets is given an appropriate cost, there will be an economic balance. Delivery robots will face the same issue, and people will avoid scheduling them for congested times. But some people will still need things then.
Without care, the mechanisms for congestion charging and metering can turn into something that tracks all cars wherever they go and records it. This can be avoided with the proper design.
For example, congestion charging can be done with a smart transponder in a car which hands over anonymous digital "coins" or tokens. People can buy congestion tokens for cash, or buy them with credit cards but use cryptographic blinding to purchase tokens that do not connect with the purchaser.
Then, when entering a congested area that requires payment, the transponder would hand over a token. The virtual "toll booth" would recognize the token as valid and mark it spent, but would not know who used it. Only if the "booth" did not get a valid token would it take a photo of the licence plate to issue a ticket.
The same system can be used for metering and reservations, except the tokens are free and allow passage at the specific time. Again only if somebody bypasses the system would they be photographed and ticketed.
In some cases, the metering would be done by having some lanes for use by cars with transponders and virtual tokens. Other cars would need to wait for a metering light. (This requires there be a space for them to wait, as is the case in highway on-ramps.) There could also be real toll booths for those who want to choose between waiting and paying.
A car wishing to plan a route and reserve slots within it need not identify the car or occupants. As long as it has the right authentication tokens it can pass without recording who it is. However, this may not protect the privacy of the car since only one person likely plots a route from my house to my office. This requires that the route be broken up into chunks and the association destroyed after it is no longer needed. Most companies resist destroying information, they think they can find a use for it later. As such, customers or governments will need to insist on such destruction.
Car cooperation and centralization.
I have deliberately avoided discussion of cooperation between cars. I mostly avoid this topic because while it's cool, it is only practical once you have a lot of robocars ready to cooperate. As such it's the problem you solve later, not first. Most of the proposals above rely on traffic computers managing and reporting on individual blocks, and cars simply sending queries too and receiving broadcasts back from these computers. There is remarkably little need for the cars to communicate with one another, at least at first.
As time goes on, some things will benefit from car to car communication. For example, if you talk to and identify a trusted robocar near you, you might be able to follow it a bit more closely, or depend on it to swerve with you if you have to swerve to avoid the unexpected -- for example to pass the school of fish test. You could also ask other cars to fill in blind spots in your vision, letting you know what is behind them from your viewpoint. Handy, but it can't be necessary since you don't have it on day one.
Ideally even the city and street computers are highly decentralized. Streets can be receiving updates on their traffic and making predictions but this need not be one computer. It certainly will be a network of different computers when you get the scale of large metro areas or states. Chances are various competing services will offer traffic predictions which they will make by talking to computers that manage individual streets, and by talking to the cars.
Reservations for use of a block of street can be handled by a computer dedicated to just that block if desired, though usually we can expect many blocks to be handled by a typical system. Your computer just needs to know how to reach that computer in the "motornet."
The early techniques don't even involve the cars sending data up, it just involves getting broadcasts or fetching over a general network. That's good as it has a lot less complexity, and is already happening.
Just not enough capacity
It can of course be the case that, even with all the techniques outlined above, there just isn't enough road to move enough 1-2 person vehicles at rush hour. In this case the problem can be solved using the existing transit lines built in the 20th century, which when fully packed have very high capacities. (Indeed, these can be very efficient if you don't need to run them in off-peak hours.)
Another option is computer coordinated robo-buses and robo-jitneys. If you wish to go downtown at rush hour and the metering system indicates a long delay or high price to do so, it might also offer the chance to ride your robocar or robotaxi directly to a point where you will transfer with minimal delay to a group vehicle, which could be a train but is more likely to be a bus. The bus would take all of you along the route you were all already wishing to go, and drop you at a point close to your destination where a robotaxi (or in rare cases another bus or jitney) would be waiting for you, with another instant transfer.
Unlike scheduled bus lines, these bus trips would be created on demand given reasonably short notice. They might be full size 60 passenger articulated buses or 8 person vans. If the trip happens to match a rail line, it might be a train, though ad-hoc scheduling of trains is much harder as they don't usually have offline stations (ie. they block other trains while waiting at a station) and can't easily pass other trains.
If there are not enough buses to meet demand, small waits may be necessary, or even reservations in advance to use the bus.
In the extreme, the highways and major routes could become mostly fleets of closely packed robo-buses. 40 passenger robo-busses moving 80mph with 1/2 second inter-bus spacing (60') on a 6 lane highway (lanes can be safely switched in direction for robots) presents the theoretical ability to move 1 million people per hour fully utilized. That's 100 times greater than what the same 6 lane highway handles today with cars and 3 lanes in each direction.
Of course, that's a ridiculous number of buses, and much more than we need. It is an academic exercise to show what the capacity of the roads can be.
It is unfortunately likely that the earliest robocars and pre-robocars may lead to some increase in congestion. Already several car makers have announced "traffic jam assist" products which drive in traffic jams.
These cars, and the fully autonomous robocars, will make being in a traffic jam much more tolerable. It will not be frustrating and it will in some cases no longer be lost time. Today people who reschedule their day to avoid rush hour may find the balance shift and move their travel back to the peak times. If they do, it's a concern until these other approaches can reverse this.
In addition, empty robocars moving to pick people up feel no frustration from traffic, though it means they have to take longer and spend more fuel to reach their passenger.
Guideways for light vehicles
Robocars will come in many sizes, but I make an optimistic prediction that it may become common to travel urban trips in a small, single passenger vehicle. Such vehicles will be light as well as small. This opens up a few interesting options in the creation of roads meant only for these ULVs (Ultra-light vehicles.)
Elevated paths for ULVs are vastly cheaper than ordinary roads. Ordinary roads cost around $1.5 to $2M per lane-mile. At-grade ULV lanes are little more than bicycle paths and have been known to be built for $150,000 per mile. Elevated guideways are not common so there is not as much history on them, but due to the lighter weights they are much cheaper than elevated roads, and their footprint on the ground is very small. Elevated roads are also bridges, allowing intersections with no traffic control.
Even at a cost approaching roads, elevated guideways can be built above existing roads (with support poles on the median or shoulder) to increase capacity where building new full roads is not possible. And they can also be built in the most congested areas to add capacity where it is needed.
Roads for light vehicles are much cheaper to build and maintain as most of the wear and tear on roads comes from heavy vehicles. While it's difficult to have an area be only served by roads for light vehicles, a combination of ULV track and regular roads can be much cheaper and offer high capacity.
Tolerance for longer rides
The environment inside a robocar -- a comfortable seat with work desk, screen, internet, phone and more, or a sleeper, or a face-to-face meeting area will create tolerance for rides in congestion. Efforts at a smoother ride through different suspension systems and less jerky acceleration and braking will also offer more comfort, as can tilting to make sure all forces are vertical. This may make people more tolerant of slower streets and alternate routes, or slower movement on congested routes.
Consequences for Cities
The ability to move several times as many people per hour into and out of a city with the same infrastructure doesn't just mean a faster and more pleasant trip. It also changes the current constraints on the size of cities, and the urban densities that are workable. Cities may double or triple in size, with far-reaching consequences.
Theory and Practice
These theoretical increases in capacity and reduction in congestion have yet to be proven. In reality they will not be as high as the predicted maxima. Over time, research in simulators and eventually on real streets is needed to find out what the best policies and algorithms are. The social, personal and environmental cost of congestion is very high and the rewards are great for finding the algorithms that work.
Comments can be left at the Blog post on Congestion.