Are fully functional self-driving cars right around the corner? After years of optimism, we're starting to see people wonder if they might be further away than people thought. And skeptics point to mapping as a key obstacle.
Writing at the Marginal Revolution blog, for example, economist Tyler Cowen argues that "mapping the territory, reliably, will remain the key problem. Until that is solved, driverless cars will be a form of mass transit — except without the mass — along predesignated routes."
Creating detailed and comprehensive maps is difficult in the sense that it takes a lot of work, but it's not a hard technical problem. Google has already done it for roads around their corporate headquarters in Mountain View and some of its competitors likely already have the same capabilities. Expanding these maps nationally doesn't require a conceptual breakthrough, it just takes money — and Google has a lot of money.
We can expect Google and its rivals to start their mapping projects in major cities where the bulk of Americans live. That means that most of us are likely to get access to door-to-door self-driving car service long before the last mile of rural American roads has been mapped.
Google's cars drive freestyle
Computers have been able to beat human beings at chess since 1997, when the IBM's Deep Blue beat human champ Garry Kasparov. But in his 2013 book Average Is Over, economist Tyler Cowen pointed out that (at least at the time he was writing) mixed teams of humans and computers — known as freestyle chess teams — were even better at chess than computer software alone. Humans provided valuable strategic insights to complement the massive computing power of the machines.
Google's self-driving car technology works on the same principle. A computer inside the car is responsible for making second-by-second driving decisions. But the car is in constant contact with Google headquarters, where a large team of human analysts — backed up by the vast computing power of Google's data centers — maintains an extremely accurate, detailed, and up-to-date map of the streets where Google's cars are driving.
Google described its effort to build this map in its most recent monthly update on the self-driving car program.
"Before we drive in a new city or new part of town, we build a detailed picture of what’s around us using the sensors on our self-driving car," Google writes. "Our mapping team then turns this into useful information for our cars by categorizing interesting features on the road, such as driveways, fire hydrants, and intersections."
Google's self-driving cars are able to navigate city streets pretty well even without this kind of detailed map. But when people's lives are at stake, "pretty well" isn't good enough.
The human-annotated map provides an extra margin of safety, allowing a car to know its location within about 4 inches. And identifying permanent, immovable road features ahead of time, the map allows a car's onboard software to quickly focus in on objects that aren't labeled in the map. These new objects tend to be people, animals, or vehicles that are likely to move, requiring the car to be extra cautious.
Google will need a huge staff to update its maps
There's a lot that software can do to speed up the process of identifying objects like street signs and fire hydrants, but Google still employs human analysts to do much of this work. When a single mistake could lead to an accident, it's better to be safe than sorry.
Right now, Google only has this kind of detailed maps for a small fraction of the country's roads — primarily in the area around Silicon Valley and Austin, Texas. Taking these maps national will be expensive. Only Google knows exactly how expensive, but we can make some educated guesses by looking at how much online mapping companies are spending to maintain their maps today.
A rough 2012 estimate found that maintaining the data for a global mapping service costs $1 billion to $2 billion per year, a figure that's in line with industry rumors.
We should expect the maps used for self-driving cars to be even more expensive because they're going to have to be a lot more detailed. The maps that power Google's self-driving cars are going to have to mark a lot of features — fire hydrants, driveways, street signs, and bushes — that aren't relevant for merely providing turn-by-turn directions.
Ironically, then, the effort to automate driving may actually create a lot of jobs, especially in the early years as self-driving technology is being rolled out. As Google and its competitors expand their self-driving vehicle programs nationwide, they're going to have to hire thousands of human analysts to produce the detailed maps that enable cars to drive safely.
And this won't just be a one-off development, either. Landscapes are changing constantly, with changing speed limits, new construction, and trees growing and being cut down. So while maintaining maps may require less manpower than creating them initially, self-driving car technology is likely to employ a lot of people for the foreseeable future.
This is one reason self-driving cars are likely to be rented, not owned
If a company had to build a nationwide map before it could bring its self-driving technology to market, that could be a major obstacle. But companies don't have to do that.
The alternative is to introduce the vehicles as an on-demand service rather than a product customers can buy. For example, Google might just map the city of San Francisco and then offer a self-driving car service that competes with taxis, Uber, and Lyft within the city limits. As the service grows in popularity, Google could expand its service territory, first to other parts of the San Francisco Bay Area, then to other major metropolitan areas.
It might take many years before Google manages to offer services in outlying rural areas. But as Uber has demonstrated, there's a ton of demand for on-demand rides restricted to major metropolitan areas. And without the need to pay a human driver, we can expect Google's self-driving cars to be dramatically cheaper than Uber's, which will mean even greater demand for the services.
Maps may be a key strategic asset for the self-driving car industry
Google seems to believe that detailed maps are an essential resource for allowing cars to drive themselves safely. Of course, it's possible that some of the other companies working on self-driving car technology — Uber, Tesla, Apple, and several major car companies are all rumored to be working to develop self-driving technology of their own — will find ways to build fully self-driving cars that aren't reliant on maps. But it's also possible that maps will be an essential resource for self-driving systems for the foreseeable future.
If that happens, it will provide a big strategic advantage for companies that have experience managing map data. That includes Google, of course. It also includes Apple, which has a mapping app for the iPhone. It may include BMW, Daimler and Audi, which jointly paid $3 billion for Nokia's mapping division last year. And it may also include Uber, which has been buying up mapping assets.
The expense of managing maps may also be a major reason why GM teamed up with Lyft earlier this year. Right now, GM's business model is to sell cars to people who expect to be able to drive them anywhere they want nationwide. But if self-driving cars need detailed maps of everywhere they go, then this business model would force GM to build a detailed nationwide map before selling its first self-driving car, an extremely daunting prospect.
Instead, the Lyft partnership gives GM the opportunity to build self-driving cars that, like Lyft's service, only operate in major metropolitan areas. Collecting the mapping data required to operate in these limited areas is a much more manageable problem.
Still, the importance of maps to the self-driving market is another reason that car companies may struggle to remain market leaders as the industry shifts to fully autonomous technologies. Google, Apple, and Uber have a lot of experience collecting, analyzing, and distributing vast quantities of fast-changing geographic data. Ford, GM, and Toyota don't.
This also may explain why car companies have been focusing on developing partially self-driving technologies like adaptive cruise control, emergency braking, and self-parking. These relatively simple self-driving capabilities don't rely on maps, and they're compatible with car companies' existing business models. Car companies hope that these will provide customers with enough of the benefits of self-driving technology to provide a competitive alternative to fully self-driving products from Google and others. But in the long run, this approach seems unlikely to work that well, as the benefits of fully self-driving cars will be massive.