Autonomous cars likely to increase congestion

CleanTechnica

Congestion is a serious issue in developed countries, but autonomous cars are likely to be as much a part of the problem as part of the solution. Consider these US numbers from a studypublished in late 2015:

The top 30 bottlenecks are each responsible for more than one million hours of lost time annually. Drivers stuck on these roads experience total delays of about 91 million hours every year, the equivalent of 45,500 person-work years. The lost value of time to the economy from congestion just in this handful of locations is upwards of $2.4 billion annually.

Many advocates frequently allude to traffic congestion as one of the key things which autonomous cars will help solve, but it’s worth looking at the actual causes of congestion and how likely patterns of autonomous car use will play out in the real world. As this graphic from the US Department of Transport’s 2012 Urban Congestion Trends report shows, 40% of congestion comes from bottlenecks. While there are other contributors, the biggest single contributor to bottlenecks is too many cars on the road. Almost everything that can be done to the roads will drive only marginal improvements, while increases in car miles traveled will have a direct and non-linear increase in congestion.

Typically, autonomous car advocates point to two elements which will definitely decrease congestion, but ignore the factors which are likely to increase it.

Reducing collisions reduces congestion — One major advantage of autonomous cars is that they won’t get in accidents with anything like the frequency or severity of human-driven cars. This will reduce overall congestion, while also seriously reducing the number of auto body repair and paint shops required. But they will still get in collisions. The evidence on this is fairly clear:

The Insurance Institute for Highway Safety (IIHS) has seen a 7 percent reduction in crashes for vehicles with a basic forward-collision warning system, and a 14 to 15 percent reduction for those with automatic braking.

That’s excellent, but the automatic braking is one of the primary collision avoidance tools for cars, and it achieves only 14% to 15%. Further, traffic collisions contribute only 25% to congestion. Solutions which decrease collisions but increase road miles traveled as well could have a larger impact with the road miles than with the collisions.

Increased packing of cars on the road reduces congestion — One potential benefit of autonomous vehicles is that they can communicate with each other and traffic management systems and permit greater numbers of cars to get through a particular stretch of road in an hour without volume-induced traffic jams. This is hypothetical in that it hasn’t been proven with real-world traffic and real autonomous cars yet, but it has been well modelled for decades, and there is little reason to doubt that in general this is true. The chart accompanying this shows varying degrees of increased flowthrough of traffic above the level at which congestion occurs due to the vehicles have varying degrees of autonomous and cooperative driving. It finds that, with 100% cooperative adaptive cruise control — Teslas with Autopilot that talk to one another — an increase of 100% in the normal volume could be managed. This would have a very large impact if car miles driven didn’t merely increase to consume the capacity.

And that’s where the rub is with autonomous vehicles. There are differences of opinion about how autonomous vehicles will actually operate as opposed to idealistic models.

---

There are different scenarios on this category and its impacts, and differences of opinion. In the absence of any actual data, let’s look at the drivers of behaviour:

Price point reduces transit and increases congestion — Autonomous taxi or Uber equivalents are presumed to be cheaper than taxis or Ubers driven by a human. Eliminating the labour cost from taxi equivalents will likely reduce their cost to consumers. This would make them a more viable option for more trips. But what trips exactly? The Uber / Lyft experience indicates that they are competing to a great extent with transit and increasing total car miles driven. If they become cheaper, then transit gets gutted of riders more, making it less and less economically viable, and congestion increases. If more trips are taken in single passenger vehicles, more vehicles will be in those bottleneck areas, and there’s a humdinger of a problem with that:

Congestion is a non-linear function, so as a road approaches its maximum capacity, small changes in traffic volumes can cause proportionately larger changes in congestion delays.

That’s right, just small numbers of additional vehicles will create traffic jams in free-flowing traffic. Autonomous cars somewhat mitigate this effect, but that doesn’t matter if autonomous vehicles means more vehicles on the road at the same time.

Further, ride shares have to drive further than someone in a private car would in order for the passenger to make their trip, and will often have to travel through bottlenecks empty in order to pick up their customers or return to their designated drop off points. Completely empty autonomous cars on the road are not likely to reduce congestion while requiring more driving.

Consumers stay attached to private cars — Many people like having cars and use them as extended backpacks and purses. They leave fitness clothes, files, items to be repaired or returned, purchases, and the like in them as they go about their daily rounds. They are status signifiers to neighbours. They are places of privacy. They are vehicles whose sunk costs are turned into travel conveniences, with weekend trips to skiing and annual road trips south made economical by having a car for commuting. Early anecdotal indicators from Tesla drivers with Autopilot are showing much more relaxing long drives, and the Tesla model has free Superchargers, so long-distance driving is becoming much cheaper and easier with autonomous cars. While many automotive theorists predict radical changes in human behaviour related to car ownership, there’s little reason to believe that people will become rational just because of a new technology. There will be an impact on private vehicle ownership, but not nearly as big as some advocates suggest. And if this is the case, then there is little reason to expect a reduction in total car miles driven, the largest indicator of congestion.

Reduction in family vehicles increases total miles driven — While in general there is little reason to believe that there will be substantial decreases in ownership, let’s explore that model, in case it comes true. Let’s identify a model family of mother who works downtown, father who does piece work at home, and two children in school. In this model, the family’s single autonomous car delivers the mother to work downtown, then swings back and takes the kids to school, then swings back and takes the father to a mid-day yoga class, then unspools this at the end of the day. This does take a two or maybe even three car family down to a one car family. Except it does it by up to doubling the total miles driven by the family, as the car goes two ways. Once again, congestion is strongly correlated to total miles driven by cars, and in this model an empty car is driving around with no one in it a lot of the time and the total miles driven by the family vehicle go up. That likely increases congestion.

Parking outside of the city core increases total miles driven — Right now people who drive to work park at or very close to work. That’s pretty useful because they have access to their vehicle and it stops driving once they reach their destination. Under one model posited for autonomous cars, the car wanders away to cheap parking outside of the core. The problem with this is that it once again puts more vehicle miles on the car which is strongly correlated to increased overall congestion. Imagine a bunch of empty cars on the roads at rush hour trying to get through city streets to cheap parking.

Passenger comfort reduces intersection throughput — There’s another downside to autonomous cars related to congestion. People mostly will want them to be as comfortable as trains, accelerating and decelerating smoothly. Studies have suggested strongly that this means that car behaviours will slow traffic through intersections, increasing passenger comfort at the expense of overall efficiency. This can be seen by the innumerable stories about Google Cars driving like short-sighted, very polite, very timid people who have all the time in the world. So in addition to intersections having a bunch more cars in them, many of them empty of all passengers, the autonomous vehicles, especially the ones with people in them, will be taking from 4% (best case) to 2000% longer to clear an intersection. While autonomous cars increase highway vehicle capacity, they could decrease intersection vehicle capacity by 18% to 53%.

Longer commutes become viable, increasing total miles driven — At present, commuting by car is highly unproductive time. Commuting by transit is also problematic due to lack of power outlets, WiFi, and ability to have private conversations. There’s a model of the future, however, in which autonomous cars provide a private productivity zone, allowing people to live even further from work. Autonomous cars promoting suburban living and long commutes as a viable alternative increases traffic congestion, not the inverse.

Induced demand increases volume — Induced demand says that, after supply of something is increased, demand increases as well. This is a well understood economic behaviour that can be observed in efficiency programs and traffic. Study after study after study shows that when you increase road carrying capacity, the roads just fill up again.

Duranton and Turner say building more roads results in more driving for a number of reasons: People drive more when there are more roads to drive on, commercial driving and trucking increases with the number of roads, and, to a lesser extent, people migrate to areas with lots of roads. Given that new capacity just increases driving, they find that “a new lane kilometer of roadway diverts little traffic from other roads.”

There is no reason to believe that increasing road carrying capacity due to autonomous cars being able to drive more closely together with fewer volume-induced slowdowns will not be subject to exactly the same effect. Ignoring all other aspects of adoption, if more vehicles can get through a specific road segment in an hour, it will probably fill up to that maximum volume with resulting congestion bottlenecks reoccurring.

It’s also worth, of course, pointing out that all of those additional vehicle miles would produce much more carbon and pollution emissions in the absence of any other changes. Thankfully, electric vehicles are leading the charge in this space, and there is a strong likelihood that significant penetrations of fully autonomous vehicles will match the necessary electrification of those vehicles.

So, the traffic congestion scenario is not as rosy as many autonomous car proponents suggest. For the purposes of discussion, I’ve put together some projections of different scenarios. In each, I’ve reduced collisions by 40%. Most other factors don’t change, bottlenecks due to volume swings substantially with increased vehicle miles and the non-linear effects discussed earlier.

At best, the greater number of cars which would be able to get through a given point without congestion ensuing would be consumed by the greater number of cars both full and empty driving through that point on average. This is behaviour which is remarkably similar to building new roads, in that relief from congestion is always temporary. “If you build it, they will come” is the reality of traffic. In this scenario, there is still some benefit from reduced collisions, so overall traffic is better.

In the median scenario, all benefits from reduced accidents are consumed by greater numbers of vehicles on the road, leading to exactly the same amount of congestion as before, just balanced to more cars rather than more collisions.

And in the worst case, there’s a potential for much worse congestion due to all of the vehicles on the roads.

---

As I look across all of the systemic implications of autonomous cars, I see equal or greater congestion on our roads, not less, and more societal negative externalities than advantages. Autonomous vehicles are not a silver bullet for traffic problems. They are coming, and as a person who drives, I welcome them, but good policy and adaptation to the transformation requires clear eyes.