The big question about self-driving cars is “when?” On one hand, there are rumors that Google will start selling its self-driving cars next year. While even the Antiplanner doesn’t think that’s realistic, Ford is promising self-driving cars in 2019 and other manufacturers are saying 2020.
On the other hand, many are saying that, due to liability concerns and technical problems with such factors as rain and snow, it will take much longer than that. Another study predicts that, even if the first self-driving cars enter the market in the next decade, it will take several decades after that for them to dominate the roads.
The Antiplanner has written on this before, but the more I learn, the more I am convinced that the first self-driving cars will be for sale by 2020 and that they will be the dominant form of travel within not much more than a decade after that.
Partly the difference in predictions is between advocates of two different models of self-driving cars. One model relies heavily on vehicle-to-vehicle and vehicle-to-infrastructure communications. The problem with that, as a recent GAO study indicated, is that a single infrastructure station is likely to cost over $50,000. Putting one at every intersection and every freeway on- and off-ramp would cost billions of dollars. Updating them as new technology becomes available would be a nightmare. The advocates of this technology (which mostly includes people who expect to make money from it) are the most pessimistic about when self-driving cars will be available.
Google, Ford, Volkswagen, and the other companies experimenting with self-driving cars aren’t relying on such infrastructure. Instead, they are designing cars that sense the world around them and respond to what they sense. Sure, their sensors could be overwhelmed by a hard snowstorm, but they will work at least as well as human eyes in such situations.
The problem that many engineers worry about is what happens when a self-driving car can no longer handle the road conditions? Suppose you are engrossed in a novel and relying on your self-driving car to take you home when all of a sudden you enter a fog bank and your car decides it can’t drive any further. How many seconds warning can it give you that you will have to take over, and will you be able to make the transition in time?
One answer to this question is that a self-driving car will do what any rational human driver would do if he or she can’t see the road ahead, and that’s pull over to the side and stop. But another answer, which seems to be Ford’s and Google’s answer, is that they won’t market a car until it can handle all situations.
Instead of relying on infrastructure-to-vehicle technologies, the Google/Ford/Volkswagen etc. model is to use extremely detailed maps, known as high-definition or HD maps. Such maps measure exact lane widths, the heights of curbs, the exact locations of traffic signs and signals, and so forth.
The premiere high-definition mapmaker turns out to be a subsidiary of Nokia called Here. Recently, Nokia sold Here to a consortium consisting of Volkswagen, Daimler, and BMW for more than $3 billion. The companies said their goal was to guarantee that the maps would be available to everyone (meaning, at least, German auto manufacturers) rather than monopolized by one high-tech company *cough* Google *cough*.
Here says it has already mapped 27 million miles of roads, with close to 2 million of them in the United States. Since the U.S. has 2.7 million miles of paved roads and another 1.3 million of unpaved roads, Here is about half done here. Not only is it mapping more every year, it makes literally millions of updates to its maps every day. Uber, meanwhile, has purchased Microsoft’s mapping division, and Google has its own mapping program.
Regardless of the source, such high-definition maps will be the heart of the self-driving cars of the near future. When Ford says it will have a car on the market in four years, it means that car will be able to drive anywhere that has high-definition maps, and human drivers would have to take over only when leaving mapped areas.
Although VW-Daimler-BMW wants the maps to be widely available, they probably won’t be free. Instead, auto owners will probably subscribe to a mapping service that not only updates the maps daily but also in response to where the auto is going. High-definition maps of the entire United States would probably fill many terabytes of memory, so your car will probably only have a subset of maps at any given time. If you plan a trip to other places, you’ll have your car download more maps, which may or may not be a part of your regular subscription.
Based on this, I find it completely reasonable to think that, by 2020, you’ll be able to buy a car that is completely self-driving on many roads and streets, but that may require you to take over on some little-used roads. Moreover, I think these cars will penetrate the market much faster than pessimists think, both because companies like Uber will make them available for-hire at rates of about 25 cents a mile, and because many cars built between now and 2020 will be easily upgradable to be self-driving cars. And, contrary to some expectations, such self-driving cars should almost completely replace transit (at least, outside of New York City) within a few years after they are introduced.