Will Politicians Block Our Driverless Future?
Autopia is within our grasp-if government doesn't screw it up.
Faced with innovation that operates outside of their control, many lawmakers can't stop themselves from grabbing the steering wheel. Sometimes literally.
"So, I'm in the Tesla," Sen. Ben Nelson (D–Fla.) told a March hearing of the Committee on Commerce, Science, and Transportation. "And we're coming back across the Anacostia River and getting up on—on the bridge then to get on to the ramp on to 395. And I'm instructed in the driver's seat: 'Engage the autonomous switch.' I click it twice. 'Take your hands off the wheel.' And so all of a sudden the car is speeding up and they say: 'It automatically will go with the flow of the vehicles in the front and back.' But now we are approaching the on-ramp on to 395 and it is a sharp turn. And the vehicle is still speeding up. And they said, 'Trust the vehicle.'"
But the senator could not quite manage to do that. "As we approach the concrete wall," he said, "my instincts could not resist. And I grabbed the wheel, touched the brake and took over manual control. I said, 'What would have happened?' They said, 'If you'd left your hands off of the wheel, it would have made that sharp turn and come on around.' And so I am here to tell you that I am glad I grabbed the wheel."
Making the senator's panic even more metaphorically apt was the title of the Senate hearing: "Hands Off: The Future of Self-Driving Cars." While many lawmakers therein expressed considerable enthusiasm about autonomous vehicles, most members of the committee made it clear that they'll be hard-pressed to stick to a hands-off approach to regulating driverless cars. As Nelson put it, "In the federal government we have a critical role to make sure that the regulatory environment and legal environment in which American business does business is able to develop and manufacture these vehicles. And also it means that we're going to have to—in our case—exercise responsible oversight."
Nelson's words should strike fear in the hearts of those who are looking forward to playing Scrabble, applying makeup, or reading a Kindle on the freeway. Leading automakers such as Ford, Toyota, and Volvo, plus contending tech companies like Tesla and Google, forecast that self-steering cars will be widely available by 2020, with fully autonomous vehicles less than five years later. Driverless cars have the power to make us richer, less stressed, more independent, and safer.
Unless lawmakers and regulators manage to screw everything up.
Robots, Take the Wheel
"We should be concerned about automated vehicles," University of South Carolina law professor Bryant Walker Smith told the Associated Press in March. "But we should be terrified about today's drivers."
Self-driving cars aren't perfect and they never will be. They are, however, significantly better pilots than today's distracted, maladroit, and sometimes drunk human beings. In 2014, some 38,000 Americans died in traffic crashes. The National Highway Traffic Safety Administration (NHTSA) estimates that 94 percent of automobile accidents in the U.S. are due to human error. A Virginia Tech Transportation Institute study commissioned by Google estimated in January 2016 that human-driven vehicles crash 4.2 times per million miles traveled whereas current self-driving cars crash 3.2 times per million miles, a safety record that's likely to keep improving as robocars gain more real life experience on the roads.
California, unfortunately, didn't get the memo about comparative human/robot safety records. The Golden State's Department of Motor Vehicles (DMV) wants to mandate that all self-driving cars have steering wheels, pedals, and a licensed, specially trained driver in the front seat, according to the 22 pages of draft regulations Sacramento released last December.
As the autonomous vehicle pioneer Brad Templeton wryly summarizes: "We don't understand this, so let's not allow it until we do understand it." There's a cost to such pre-emptive prohibitions. "Once you ban something, it is hard to unban it," Templeton, a software engineer who sits on the board of the Electronic Frontier Foundation—and formerly consulted on Google's self-driving car project, says. "A regulator doesn't want to be the regulator who unbans something and then something goes wrong."
To Templeton's point, a February open letter from a consortium of California tech business associations to the secretary of California's State Transportation Agency decried the "draft regulations that explicitly prohibit the operation of fully autonomous vehicles in California."
Google's vehicle division, in a December statement, declared that it was "gravely disappointed" with the draft regulations, especially the requirement for a licensed driver to be inside the car. In a public blog post, the company's self-driving car chief Chris Urmson urged, "Instead of putting a ceiling on the potential of self-driving cars, let's have the courage to imagine what California would be like if we could live without the shackles of stressful commutes, wasted hours, and restricted mobility for those who want the independence that the automobile has always represented."
But regulators are busy slapping shackles on this technology before it has a chance to deploy. Testifying to the March Senate committee hearing, Urmson noted that in the past two years 53 bills aiming to regulate self-driving cars have been introduced in 23 states. Five states have already adopted laws regulating autonomous vehicles, he said, and "none of those laws feature common definitions, licensing structures, or sets of expectations for what manufacturers should be doing." He added that unless a unified approach is crafted, these disparate state regulatory schemes "will significantly hinder safety innovation, interstate commerce, national competitiveness, and the eventual deployment of autonomous vehicles."
"What we have found in most places," he said, "is that the best action is to take no action. And that in general the technology can be safely tested today on roads in many states."
Send in the Feds?
All of the industry participants at the Senate Commerce Committee hearing strongly urged Congress to pre-empt the developing patchwork of state regulations on self-driving vehicles. "We propose that Congress move swiftly to provide the Secretary of Transportation with new authority to approve life-saving innovation," Urmson said. "This new authority would permit the deployment of innovative safety technologies that meet or exceed the level of safety required by existing federal standards, while ensuring a prompt and transparent process."
In contrast with California regulators, the feds at the NHTSA responded positively in February to Google's request that the agency legally recognize the company's on-board computerized self-driving system (SDS) as the driver in a self-driving vehicle. "If no human occupant of the vehicle can actually drive the vehicle, it is more reasonable to identify the 'driver' as whatever (as opposed to whoever) is doing the driving," the NHTSA letter declared. Even so, the agency noted that it would have to embark upon a rulemaking to figure out how to make sure that the SDSs' sensors can monitor and operate such federally mandated safety features as turn signals, headlight dimmers, rear visibility, and so forth.
At the North American International Auto Show in Detroit in January, Secretary of Transportation Anthony Foxx announced that his agency would develop within six months federal guidance on the safe deployment and operation of autonomous vehicles and devise a model state policy on autonomous vehicles outlining a path to a consistent national scheme. In addition, the Obama administration proposed in its 2017 budget allocating nearly $4 billion to funding various unspecified autonomous vehicle pilot projects.
Not So Fast
So is federal intervention the solution? Not exactly.
There are two "equally important components that will determine the future of autonomous vehicles," Lyft's vice president for government relations, Joseph Okpaku, said at the March hearing. "The first is the interaction of everyday people with these new vehicles, and the second is the much more unpredictable interface of the government with this entirely new transportation resource."
University of Texas engineer Kara Kockelman notes that traditional automakers tend to "see the transition to self-driving as a very natural, a very normal process adding over time features like GPS, adaptive cruise control, cameras, lane keeping assist systems, dedicated short range communications, and so forth." Such semi-autonomous vehicles can safely operate only in predictable traffic environments, so some manufacturers are suggesting that dedicated additional infrastructure such as separate highway lanes be built for them.
But "special lanes are a bad idea," says Kockelman. "They would be incredibly expensive and constraining." Planners, politicians, and regulators may think that establishing dedicated infrastructure and rules for self-driving cars is helpful, but Templeton notes that "such rules could easily lead to them not being allowed in the ordinary lanes."
Kockelman argues that semi-autonomous vehicles, or what NHTSA calls "limited self-driving automation," present a big safety problem. With these so-called Level 3 vehicles, drivers cede full control to the car for the most part but must be ready at all times to take over if something untoward occurs. The problem is that such semi-autonomous cars travel along safely 99 percent of the time, allowing the attention of their bored drivers to falter. In an August 2015 study NHTSA reported that depending on the on-board alert, it took some drivers as long as 17 seconds to regain manual control of the semi-autonomous car. "The radical change to full automation is important," argues Kockelman. "Level 3 is too dangerous. We have to jump over that to Level 4 full automation, and most manufacturers don't want to do that. They want protection; they want baby steps; they want special corridors; they won't get that."
Consequently the first law of the robocar revolution, according to Templeton, is "that you don't change the infrastructure." Whatever functionality is needed to drive safely should be on board each individual vehicle. "Just tell the software people that this is the road you have to drive on and let them figure it out," Templeton says. "Everything you must do is in software or you lose." Some self-driving shuttles confined to specific areas—airports, pedestrian malls, college campuses—will be deployed, but they are not the future of this technology.
Another infrastructure mistake would be mandating the deployment of "smart roadside infrastructure" such as traffic lights and sensors to monitor conditions like icing on bridges and communicate the information via radio to autonomous cars. In 2015, Sens. Debbie Stabenow (D–Mich.), Gary Peters (D–Mich.), and Lamar Alexander (R–Tenn.) embraced this idea when they introduced the Vehicle Innovation Act, which included spending more than $300 million on various favored tech, including vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications systems.
Before embracing such external information systems, keep in mind that the U.S. Department of Transportation estimated in 2007 that 75 percent of the nation's 330,000 traffic lights are mistimed or use obsolete control systems. "If city and county street and road agencies can't keep traffic signals up-to-date, how long would it take them to install and upgrade smart road systems?" Randal O'Toole asked in a 2014 Cato Institute study, "Policy Implications of Autonomous Vehicles." It's all most states and cities can do to fix potholes, much less deploy and maintain sophisticated networks of roadway sensors.
Other regulators and politicians want to require automobiles to be equipped with V2V communications tech using dedicated short-range communications (DSRC) protocols. The idea is that the cars could talk with one another to provide warnings of traffic jams, accidents ahead, or vehicles in front that are braking. They might even cooperate with one another to get through intersections. A good bit of the Obama administration's promised $4 billion dollars would be earmarked for V2V research and development.
"Some cellular experts feel DSRC will be obsolete by the time it's required on new vehicles," says Kockelman.* "Regulators simply can't write down a communication standard that will be useful for a long time." Templeton agrees. "People outside the industry think it's essential and the car companies are just going along with it to keep them happy," he says. "It's something designed in 2000 [that] wouldn't be fully deployed until 2030 or later." The bottom line: "Mandating V2V connectivity is stupid and a waste of time."
Templeton cites the internet as a model for how to roll out the technologies that enable self-driving cars. "The internet is a dumb network that connects smart devices," he explains. "You want smart cars running on stupid roads." Dumb networks push innovation to the edge, giving end-users control over the speed and direction of change.
Winning Hearts and Minds
Despite an embrace of other life-changing tech—who can even remember life pre-internet?—Americans remain rather terrified of driverless cars. In March, the American Automobile Association asked 1,800 Americans what they thought of self-driving automobiles. Seventy-five percent were not yet ready to trust the driving to robots. More women (81 percent) than men (67 percent) were afraid of allowing an autonomous vehicle to drive itself with them in it. Baby boomers (82 percent) were more afraid of self-driving vehicles than younger generations (69 percent). On the other hand, most drivers did say that they would like their next cars to be equipped with semi-autonomous features such as lane-departure warning, adaptive cruise control, automatic emergency braking, and self-parking technology.
Regardless of the suspicion, America's love affair with the individually owned automobile may already be waning. A January 2015 study by Michael Sivak and Brandon Schoettle at the University of Michigan Transportation Research Institute reported that the percentage of Americans with a driver's license decreased between 2011 and 2014, across all age groups. For people aged 16 to 44, that percentage has been falling steadily since 1983. "In 1983, 46 percent of 16-year-olds obtained a driver's license. In 2014, that figure has dropped to 24 percent," noted Lyft's Okpaku at the Senate Commerce Committee hearing. "That's a 50 percent change in something that I was 100 percent certain I wanted more than anything else when I was 16."
An underappreciated benefit to self-driving vehicles is that they would provide mobility to millions of Americans who can't or don't drive now, such as disabled people, children, and the elderly. Fleets of shared driverless cars could cut the average consumer's transportation costs by as much 75 percent, according to a 2013 Columbia University study by mobility specialist Lawrence Burns. A 2014 report by the investment firm ARK Investment Management noted that private vehicles in the U.S. are in use less than 4 percent of the time, or about one hour each day. Given a U.S. fleet of 260 million vehicles, that's a daily waste of about 5.8 billion hours of potential productivity.
More autonomous cars also mean less time in traffic. A January 2015 University of Texas study concluded that if travelers take advantage of robotaxis, one autonomous vehicle could replace as many as seven to nine privately owned vehicles. The Texas researchers, simulating a shared autonomous vehicle fleet in Austin, calculated that typical traveler waiting time would be around a minute. Even at rush hour, the average wait would be under four minutes.
Squadrons of electric robotaxis would also massively cut greenhouse gas emissions. A 2015 Lawrence Berkeley National Laboratory study published in Nature Climate Change reckoned that the per-mile greenhouse gas emissions of a battery-powered robotaxi in 2030 would be about 90 percent lower than a 2014 gasoline-powered private vehicle. A consumer switch to a fleet of shared self-driving cars could also free up lots of land currently devoted to parking—about 3,000 square miles, an area the size of Connecticut.
In 2007, University of California, Los Angeles (UCLA) urban planner Donald Shoup estimated that as much as 30 percent of central business district congestion arises from drivers cruising around in search of a parking space; that would be largely eliminated with driverless cars. Automated vehicles travel more efficiently on highways and streets, thus further reducing congestion. The old-fashioned Detroit business model, in which car manufacturers peddle automobiles to individual owners, will be steadily superseded by on-demand public transport in which companies sell commuters mobility rather than vehicles. Municipalities will soon recognize that this impending consumer switch will make their inflexible and costly public transit systems obsolete.
Once the technologies have proved themselves, the rollout of driverless vehicles will be rapid, because of the enormous cost savings. One highly cited analysis, put together by the global financial services firm Morgan Stanley in 2013, suggests that autonomous vehicles could bring America $1.3 trillion in annual cost savings: $488 billion in accident avoidance, $507 billion in productivity gains, $158 billion in fuel savings, $138 billion in productivity gains from congestion avoidance, and $11 billion in fuel savings from congestion avoidance. It also estimates that self-driving long haul freight trucks would cut costs by $168 billion annually.
Google in February patented an autonomous package delivery platform. Driverless vehicles will alert customers on their cellphones that their package has arrived; then people will stroll outside, open the appropriate bin with a credit card or pin number, grab the package, and send the car on its way.
Given the initial high costs of the technology, self-driving vehicles will probably be first deployed as fleets of robotaxis and long-haul freight trucks. In January, General Motors invested $500 million in the ride-sharing service Lyft with plans to develop an on-demand network of self-driving cars. Consumers will quickly accept their new robot chauffeurs. At the Senate Commerce hearing, Urmson observed, "The first five minutes are often a little tense, but after that people think, 'That car drives better than me.' We're fairly confident that once people try it out, they're going to enjoy it."
Hacked on the Highway
A 2014 Harris Interactive poll probed Americans' attitudes toward self-driving cars and reported that 59 percent were concerned about liability issues, especially who would pay when a car crashed. Some 52 percent feared that self-driving cars might be maliciously hacked, and a third worried about having their privacy invaded by government, insurers, or advertisers.
Templeton is fairly sanguine about liability issues, noting that the entire industry is hyper-aware of how safety failures would impact their businesses. "Developers don't need to prove the safety of the vehicles to the government, but first to their boards of directors and customers," he said. Moving to a no-fault insurance system would resolve most liability issues. Since human drivers rarely balance ethical issues when they are involved in an accident, self-driving cars should be programmed to cause as little damage as possible while protecting their passengers.
Is hacking a problem? "Vulnerabilities in autonomous vehicles are not a whole lot different from the sort of cyber attacks that can be unleashed on modern vehicles that are not automated today," observes the Berkeley transportation security researcher Steven Shladover. In 2015, two cyber security researchers famously hijacked from 10 miles away a Jeep being driven by Wired senior writer Andy Greenberg, by killing the engine and steering it into a ditch. And yet, even as computers have proliferated in our cars, the primary car-hacking crime is stealing vehicles by cracking their keyless entry systems. Otherwise, there are very few instances where cars are known to have been hacked by people other than researchers.
At the March Senate hearing, Sens. Ed Markey (D–Mass.) and Richard Blumenthal (D–Conn.) tried to get the industry representatives to support their Security and Privacy in Your Car Act of 2015. The SPY Car Act would authorize the NHTSA to conduct a rulemaking on motor vehicle cybersecurity, aiming to protect against unauthorized access to electronic controls or driving data, including information about the vehicle's location, speed, owner, driver, or passengers, whether transmitted or stored on-board the vehicle. The act would also prohibit manufacturers from using collected information for advertising or marketing purposes without the owner's or lessee's consent. All of the industry panelists declined to endorse the bill, arguing that it was way too early to start setting standards.
One of the biggest hacking vulnerabilities will be through the same V2V communications systems that some in Congress want to mandate. Hackers could seek to smuggle in malicious code as cars talk to one other. In a 2015 study, UCLA computer researchers simulated a vehicular botnet attack via vehicular ad hoc networks that could snarl autonomous vehicles into a massive full-stop traffic jam within 20 minutes of being launched. Researchers are busy working on ways to make sure that information transmitted between cars can be validated. In any case, autonomous vehicles will have their own on-board sources of data from their radars, laser range finders (lidars), cameras, altimeters, gyroscopes, tachymeters, GPS, and map databases. If V2V information conflicts with what the car is sensing, it will rely on what it sees rather than on what it is being told.
While acknowledging that V2V might be useful, Templeton is basically against promiscuous chatting between vehicles. "The first thing you teach your children is not to talk to strangers," he said. So why would you want your car to talk with a jalopy with an unknown reputation? While concerned about private hackers, Templeton is more worried about state actors. Computer attacks against fleets of autonomous vehicles by the Chinese, Russians, or Iranians could create chaos in major cities. In 2014, University of Michigan engineers and computer scientists reported that they had gained control of 100 wirelessly networked traffic lights in an undisclosed Michigan city and were able to change the state of the lights on command. They entitled their study "Green Lights Forever."
Even as lurid worst-case scenarios are conjured, it is good to keep in mind that our already vulnerable traffic system has never been significantly hacked.
What about data exhaust—that is, the excess data collected by an autonomous car, especially information about its passengers? Robocar privacy is a hard problem. Fleet owners will reasonably want to monitor their cars as they move about, for purposes of rebalancing their locations and recalling them for charging. They will also want to make sure that their vehicles are not being abused, and so would need to verify and hold accountable those who are using them. Passengers may prefer to keep their movements to themselves.
One possible solution is to require fleet operators to discard data after a specified short period of time. This will be resisted, since passenger data could be monetized in various ways, such as advertising. Templeton has suggested that cars could be designed so that passengers could physically shut off in-car cameras using an outside switch before entering a driverless vehicle. They would then open the shutter after exiting, so that the car's interior state could be inspected.
As in so many other areas of our lives, most Americans may be willing to surrender their personal data in order to gain access to cheap and convenient mobility services. A February poll by Morning Consult found that only 18 percent of respondents were confident that Uber would keep personal information and data secure. Yet Fortune reported in December that the number of Uber's daily rides doubled in the past year.
The best solution will be to adopt legislation and for courts to uphold Fourth Amendment privacy protections requiring law enforcement to obtain a warrant to access our travel data.
Are We There Yet?
If such regulatory roadblocks are avoided, how rapidly will self-driving cars be adopted? Let's compare two technology rollouts: old-fashioned people-driven cars vs. cellphones.
In 1900, according to the Census Bureau, there were just 8,000 automobiles registered. By 1925, the number was more than 20 million, a 2,500-fold increase. Around 260 million vehicles are registered in the U.S. today.
In 1985, there were 340,000 mobile phone subscriptions in the U.S., according to CTIA—the Wireless Association. In 2010, the number of subscriptions had risen to more than 300 million, a nearly 900-fold increase. In 2015, U.S. cellphone subscriptions exceeded 355 million.
In fact, the cellphone revolution might have kicked off a decade earlier had the Federal Communications Commission (FCC) gotten out of the way. A 1991 analysis by National Economic Research Associates argued that "had the FCC proceeded directly to licensing from its 1970 [spectrum] allocation decision, cellular licenses could have been granted as early as 1972 and systems could have become operational in 1973."The economists estimated that this regulatory delay until 1983 cost the U.S. economy $86 billion ($205 billion in today's dollars).
Autonomous vehicles' market penetration is likely to happen much more quickly, because the model of individual ownership is likely to change. How many self-driving cars would it take to supply the cheap mobility services of 260 million U.S. passenger vehicles? Using the University of Texas' low-end calculation of each shared self-driving car replacing the services of seven individually owned vehicles, the answer is 40 million self-driving cars. Let's accept Google's projection that fully autonomous cars will become available in 2020. Assuming something like cellphone adoption doubling times, an initial fleet of 100,000 fully self-driving cars in 2020 would increase to 6.4 million by 2026 and replace the services of nearly 45 million conventional automobiles. By 2029, the self-driving fleet could number more 50 million.
In less than half a generation, Americans will benefit from a massive change in how we commute, shop, and travel. Autopia is within our grasp, if only we don't allow over-cautious regulators, outdated infrastructure demands, and our fears to kill off the self-driving car.
*This quotation has been altered to clarify Kockelman's views on the obsolescence of DSRC.