Drive north on the 101 from Palo Alto toward the San Francisco airport, in a car equipped with the most sophisticated self-driving hardware available today, and two things are sure to happen. First, at least one other driver will look over in shock as the car pilots itself for long stretches at a time.
Next, as the time-worn lane stripes fade, as the concrete gets very light in color, the car will fire off warning lights and sounds as it drifts into another car’s lane. It can’t read the road. It can’t yet truly drive itself.
Automakers have promised self-driving cars will arrive as soon as next year. They’re only telling half the story: their half.
The infrastructure needed to support completely self-driving cars won’t be ready any time soon. If and when it does happen, that infrastructure is at least decades away–and it will come with a multi-billion-dollar pricetag.
Drivers have proven they’ll pay for the car hardware required for a limited level of autonomous driving. But will voters foot the huge bill–not only to catch up on long-deferred maintenance, but also to build the new generation of roads that self-driving cars will require?
Waymo self-driving prototype
What self-driving cars are
If you follow car news at all, you've heard plenty about electrification, connected cars, vehicle sharing, and self-driving vehicles from personal cars to semis. The four are on a collision course that could upend the auto industry and dramatically affect the way we get around in the near future.
If futurists are correct, gas-engine cars piloted by humans will go away, and autonomous vehicles will carry passengers on a revamped road system. By some estimates, up to 25 percent of all miles driven by 2030 could happen in autonomous, shared, electric-powered cars.
What will those vehicles be capable of doing? For starters, “self-driving cars” is the technically incorrect shorthand applied to a range of vehicle functions, all under the umbrella of autonomous driving.
Autonomous cars use suites of sensors and systems–cameras, lidar or radar sensors and ultrasonic sensors–to analyze and react to road conditions, to maintain lane control and speed control, and to shut the car down to avoid or prevent crashes.
Engineers describe autonomous or self-driving cars using five levels. Level 0 means no car intervention in the act of driving, while Level 1 means the car can intervene for safety, as many cars now do with forward-collision warning and automatic emergency braking systems. The car senses an obstacle, and at lower speeds, can stop before a crash.
At Level 2, the vehicle can control speed and steering for several seconds before it calls for the driver to take the controls. Vehicles such as the Volvo XC90 and Mercedes E-Class are capable of this. At Level 3 the car can be in total control, but will require the driver to stay at attention and to take over in case of an emergency. For safety reasons, most automakers expect to skip this level.
At Level 4, the vehicle will be able to control itself, and will stop completely if its systems fail or cannot perform adequately on the road. Finally, at Level 5, the driver is factored out entirely; the car is built to drive itself under all conditions, and need not have a steering wheel or brake pedal.
That last leap is a huge one. While many automakers have promised Level 4 self-driving cars by 2021, it’s not clear if they’ll be sold to the public, or only used by car-sharing services in tightly controlled ways, in very limited regions and circumstances.
Of course, Tesla Motors is a standout. It hasn’t specifically promised Level 5 self-driving cars, but it says it will introduce an enhanced Autopilot system that will permit it to demonstrate a self-driving car run from Los Angeles to New York by the end of the year. Tesla’s Elon Musk says the trip will be made without any driver input.
Pothole repair (via Wikimedia)
The problem: infrastructure
These automaker claims haven’t gone unchallenged, but they do lack any detail on the part of the equation that automakers don’t control: infrastructure.
Self-driving cars need better roads, better lane markings, better traffic-light timing, and better maintenance to behave predictably. Lane striping must be clear and regular. The road surface must contrast crisply with the striping. With today’s technology, the roads must be clear of snow and ice and so should the car’s sensors that are reading those roads.
Those conditions alone have limited self-driving car demonstrations to just a handful of locations. California and Nevada have led the way in grooming highways and streets for these demonstrations, and thousands of Tesla, Volvo, and Mercedes drivers accumulate piloted-driving miles on roads of varying quality.
The Department of Transportation has inaugurated an important test of conditions in 10 locations across the country. At sites from Pittsburgh to former military sites in Maryland, and the former Willow Run GM facility in Michigan, the DOT has given the green light to autonomous vehicle testing so that it can discover the benefits and problems self-driving cars will bring to the roads.
But to ensure a generation of self-driving-capable cars are worth their added cost, the roads themselves have to be maintained. And that’s a problem in the U.S., particularly in states where road maintenance is woefully out of date.
One city s about to learn how much it might cost to prepare for self-driving cars. The city of Atlanta is not one of the key DOT cities identified for self-driving cars, but it wants to be one. Later this year, the city will christen a smart-traffic corridor along North Avenue, which connects Georgia Tech with the busy Midtown corporate district.
It will upgrade the street with everything from traffic-sensing cameras to solar-powered trash compactors, all in the interests of encouraging more multi-modal transportation and self-driving car traffic. Using North Avenue as a case study, the city estimates it could take 50,000 environmental sensors, 20,000 pedestrian and mobility sensors, and 10,000 cameras on top of the the current 960 street lights and more than 50,000 street lights to upconvert its city grid to so-called “smart street” status.
City officials say the costs associated with upgrading North Avenue to a street friendly to self-driving cars are inherent in the “smart street” upgrades. They don’t describe how much that upgrade will cost versus a standard repaving project. In other cities, resurfacing a mile of road has been estimated at well over $1 million, with an additional $50,000 or more required in striping and in traffic-sensing cameras per intersection.
In 2015, the state of Georgia asked voters for a $900 million tax increase that would allow it just to catch up on deferred and increased future maintenance, including long-deferred road improvements. The bill, approved by voters, dropped the state’s popular $5,000 tax credit on purchasing or leasing electric cars and instead charged electric car drivers a road-usage fee, while it raised gas taxes on other vehicles.
As a result, the state committed to spend $2.2 billion alone on resurfacing 2,500 miles of roads, repairing and building new bridges, and improving intersections for smoother traffic flow.
That tax increase came before the city has factored in any other hardware upgrades, such as V2X receivers and transmitters. V2X (“vehicle to X”) technology allows vehicles to communicate with other cars as well as stop lights, traffic sensors, cameras, and temporary obstacles such as construction barriers. Connected by V2X, all those pieces form an “internet of things” that share information about the same physical environment at the same time.
Georgia isn’t alone in playing catch-up on roads. In 2015, California legislators produced a report that found 68 percent of the state’s roads were in “poor to mediocre” condition. In the report, the state estimated it would need $8 billion a year just to keep up its highways. It takes in $2.3 billion annually. Its backlog of road repairs could cost up to $130 billion. The state just raised registration fees and gas taxes to start chipping away at that number.
At the time of the California legislative report, the American Association of State Highway Transportation Officials estimated all but 12 states were considering ballot measures to fund major transportation projects–including catching up on deferred maintenance. And according to the DOT, half of the states have a majority of their roadways that qualify as poor or mediocre.
Waymo self-driving car prototype
Will it work?
Automakers are confident in their ability to deliver Level 4 self-driving cars to the roads within five years. People know what self-driving cars are now, and they want them.
A recent poll of drivers suggests many are willing to pay up to $700 for braking and steering systems that amplify the safety of their cars.
And, as roads and highways are rebuilt, "we think it would be very, very wise to build modern infrastructure with 21st-century capability in mind," said Paul Brubaker, president and CEO of the Washington, D.C.-based Alliance for Transportation Innovation.
As of yet, the movement lacks a driving force other than the canny marketing of Tesla’s Autopilot system–for which it’s being sued.
The leap to Level 5, to truly self-driving cars, is uncharted territory. It’s even more murky without a sufficient answer as to how self-driving cars will be able to navigate the hundreds of miles of roads under construction at any given moment across the country. Or, the thousands of miles of roads under blanket of bad weather. Without some major technological leaps, there will be huge swaths of roads that are effectively always offline to self-driving cars.
The uniquely American patchwork of roads and transportation networks would seem to bode poorly for the kind of cooperation and uniform regulations required so that self-driving cars can develop quickly and safely. There’s not even a single uniform way to measure how often an autonomous-driving system disengages–a critical measure of how safe it can be.
As difficult as the questions are for engineers, the cost to upgrade the nation’s infrastructure is almost a complete unknown. Will self-driving cars require dedicated lanes, or a nationwide blanket of V2X communication? Hundreds of billions of dollars, and decades of work, likely are at stake.
By the time the project concluded, the nation’s interstate system cost $128.9 billion dollars to build 42,795 miles of highways. Adjusted for inflation, that number now stands at more than $230 billion.
What will it cost to upgrade a nation’s worth of roads to the levels needed for self-driving cars? Whether it’s $50 billion, or $500 billion–or $1 trillion–it’s likely to take decades to get a nation’s worth of state-level governments just to agree on how to spend it.
As they evolve, the infrastructure for self-driving cars will be pitched as a life-saver for an increasingly distracted nation of drivers. As of yet, no one's said how much it will cost, or who’s going to pay for it.