When Regulating Self-Driving Cars, Who’s at the Wheel?
In order for driverless cars to conquer the road, someone has to write the rules for their use. Right now, it’s not clear who that someone will be.
The storm that rolled through Ann Arbor, Mich., in late November brought nine inches of snow and an experimental opportunity too good to pass up. A team of Ford engineers working to develop self-driving vehicles decided it would be a good time to put their modified Ford Fusion sedans to the test.
Snow, like rain, can be especially tricky for automated vehicles. Precipitation makes it harder for driverless cars to know where they are. Their cameras can’t see lines on snow-covered pavement or in the reflections of puddles. Falling precipitation interferes with radar. Piles of snow make finding the curbs and road edges harder, even for the cars’ laser-powered mapping devices. On top of that, snow is something of a novelty for self-driving cars. Most of them have been confined to sunny locales in states like California, Nevada and Texas, where rain and snow are rarer.
So the Ford team jumped at the chance to test their vehicles in the Michigan winter. Rather than heading to Ford’s proving grounds in Dearborn, they went to Mcity, a 32-acre test track in north Ann Arbor. It’s a shared track that’s operated by the University of Michigan and used by automakers and the state transportation department to try out autonomous and connected cars. Mcity includes elements you wouldn’t expect to find on most test tracks, things like stoplights that broadcast information to vehicles, a railroad crossing, a bus stop, highway on-ramps and gantries, a small hill, gravel roads, sidewalk crossings, stop signs, a simulated tree canopy and overpass, roundabouts, vandalized traffic signs, and a mockup of downtown city blocks.
Jim McBride, Ford’s technical leader for autonomous vehicles, says the company’s cars handled the snow well that day. They stayed in their lanes, followed the rules of the road and managed the slippery conditions.
The Ford engineers bring their test cars to Mcity often. In fact, they started using the facility months before it was completed last summer. For them, it’s a place where they can test their cars’ reactions to a situation over and over again. “We can set up instances where we can push mannequins on carts out between parked cars,” McBride says. “Instead of encountering those once every million miles driving, maybe we can encounter them once every five minutes.”
Ford engineers recently tested a modified Fusion at Mcity, a 32-acre track operated by the University of Michigan. (Ford)
Mcity is itself something of an experiment. It is the most concrete indication of how the state of Michigan and its industries imagine that autonomous cars will become reality. Their hypothesis is that the march toward these vehicles will be completed step by step, not in a single leap; that cars will become better at talking with each other and their surroundings even as they include more advanced autonomous features; and that these potential society-scrambling innovations will be made by the existing auto industry.
For Michigan transportation officials, that requires a light regulatory touch. “The public sector has a role in this,” says Kirk Steudle, the director of the Michigan Department of Transportation (MDOT). “What we’re trying to figure out is: How do we enable the technology to happen? It’s growing, it’s flourishing. How do we make sure the government is not in the way?”
But the competition to host the companies that will produce the cars and trucks of tomorrow is fierce. There is no guarantee that Michigan will remain the engine driving the automotive industry, especially if companies in Silicon Valley have anything to say about it. As states compete for business and the federal government works on providing national guidance, the rules of the road for driverless vehicles are not turning out to be consistent state to state. Those rules -- along with factors like technology-ready infrastructure, state leadership and even, yes, weather -- could go a long way in determining where Ford, Google or any of their competitors develop and build their autonomous vehicles.
The Mcity track in fairer weather. (University of Michigan)
Michigan trailed California and Nevada in passing a law explicitly authorizing driverless vehicles. Michigan’s law didn’t get on the books until 2013. But as the center of the U.S. auto industry, the state also has a lot more experience than most others with experimental vehicles. Its new law made clear that people other than car manufacturers and universities could build and test these vehicles, as long as they had permission from the state.
Michigan has also encouraged automakers to develop technology that lets vehicles and infrastructure communicate with each other. The state and other partners tested some of that technology for more than a year in Ann Arbor. About 2,800 cars, trucks and buses took part in the pilot project, allowing them to get information from traffic lights, curve warning devices and other sensors along roadways. Even though the testing phase is now over, the city of Ann Arbor and the University of Michigan are still expanding the network of connected stoplights and roadside sensors.
Meanwhile, the state of Michigan and automakers are rolling out devices to communicate with cars and trucks along 120 miles of roadways, particularly interstate highways, in the Detroit area. The types of sensors vary along the network, and are often tailored to the needs of nearby automakers. While the car companies use the devices to develop their own connected vehicles, the state is using them for applications that would discourage red light running or warn travelers about work zones. The devices also feed information about traffic and weather to the state’s traffic center in downtown Detroit.
To Steudle, connecting cars to the roads they drive on is an important part of developing self-driving cars. “Some say the autonomous car can go do its thing, that it doesn’t need any infrastructure,” he says. “Yeah, that’s like having a classroom of really smart people who don’t ever talk. If they talk to each other, everybody gets smarter. That’s where we see the connected piece of automation coming together as well.”
Mcity is a proving ground to test how all of those pieces fit under realistic circumstances, including inclement weather. “If you want something that’s going to have mass appeal, it’s going to be something that works at all speeds, that operates in all weather conditions,” says Jim Sayer, the director of the University of Michigan’s Transportation Research Institute, who helped design Mcity. “What you don’t want to be doing is putting the public at risk unknowingly and creating problems in traffic.”
In other words, the Michigan group is taking a very different approach than the one used by Google.
Seeing a car drive itself is nothing unusual in the neighborhoods around Mountain View, Calif., where Google’s headquarters is located. Half a mile from the main Google campus, on a modest suburban side street, sits X, formerly known as Google X, that company’s not-so-secret research and development lab for major “moonshot” projects. Right now, the most elaborate of these projects is the driverless automobile.
The X fleet comes in two different models: an older version of a Lexus SUV retrofitted with cameras and other sensors, and the newer version, a white two-seater that can operate without a steering wheel or pedals.
Google initially tested its autonomous vehicles on freeways, but in the last two years it has shifted its focus to neighborhood driving. It now sends its driverless vehicles onto public streets, alongside regular cars, pedestrians, cyclists, baby carriages, dog walkers, traffic cops, school buses, construction equipment and everything else drivers must navigate around. A Google car once encountered a woman in an electric wheelchair chasing baby ducks in the road.
One sunny December afternoon, two X engineers took one of the Lexus models out for a spin. Sitting in the front seat, the engineers monitored the car as it tooled around on a preprogrammed route, passing the tidy lawns and modest ranch houses in the neighborhood around the X headquarters building. Riding in a Google car is at once thrilling -- there’s no one operating the wheel! -- and stultifyingly mundane. It’s like riding with an overly cautious grandmother. (Last year, a Mountain View cop pulled over a Google car for going too slow, but the car got off with only a warning.) On that test drive, the Lexus always traveled at or below the speed limit, coming to a complete stop at every stop sign and accelerating slowly when it started up again. When the Lexus approached a motorcycle parked harmlessly along the curb, the car jolted to a full stop and then gave the bike an exceedingly wide berth before continuing down the street.
In fact, one of the only reminders that the car is driving itself is the display on the computer that one of the engineers holds in her lap. It shows the readout from the lidar sensor onboard, which maps the surrounding area in much the same way that radar does, only using lasers instead of radio waves. The lidar display is essentially what the car “sees.” It’s a mesmerizing jumble of geometry, an endlessly redrawn set of lines that jitter and jump as the car renders the landscape around it. The X cars are remarkably good at identifying the objects, people and other vehicles they encounter.
So far, Google’s other model, the two-seater, has only taken to the California streets with a steering wheel and pedals that employees can use to take over whenever necessary. But someday, when the technology gets good enough, Google wants to put cars on the road that don’t have a steering wheel and pedals -- or even rearview mirrors. There’s one obstacle to that goal that Google so far has not been able to overcome: the California Department of Motor Vehicles (DMV).
In 2012, California became the second state (after Nevada) to pass a law expressly allowing driverless cars on its roadways and regulating them. The law required the state DMV to issue two sets of rules: one for testing the cars and another for their general use by the public. Google and 10 other manufacturers have operated under the testing rules since late 2014, but Google, in particular, had hoped the long-stalled rules for the sale of the cars and for their use by the public would be relatively lenient.
That hasn’t been the case. The biggest flashpoint was the state’s requirement that cars used by the public must include steering wheels and pedals to allow occupants to take control of the vehicle. “Right now, we don’t have any data on autonomous vehicles being used without a driver. So we couldn’t possibly let the public use it without a driver,” says Jessica Gonzalez, a spokeswoman for the California DMV. She says her agency plans to revisit its testing regulations soon. That could open the door for letting companies test cars without steering wheels, and, perhaps eventually, enabling them to gather enough data to convince regulators that the cars are safe.
But a Google spokesman says the company was “gravely disappointed” by the requirement. “This maintains the same old status quo and falls short on allowing this technology to reach its full potential, while excluding those who need to get around but cannot drive,” says Chris Urmson, the head of Google’s driverless car project.
Chris Urmson, Google's director of self-driving cars, discusses proposed regulations on autonomous cars at a public hearing. (AP)
Google insists on developing a car without a steering wheel partly because it contends that people often don’t pay attention while their cars are operating autonomously, a fact other developers have noted as well. Google recorded one of its volunteer “drivers” reaching into the back seat while his car cruised along a highway at 65 mph. In fact, there’s a whole slew of videos on YouTube shot by the “drivers” of Tesla Model S cars while their vehicles are in “autopilot” mode (an advanced form of cruise control). Many of the videos show hair-raising near-accidents. In other words, Urmson says, the better the cars get at driving, the worse their human drivers perform. The way to solve this problem, in Urmson’s view, is to take driving responsibility away from the user altogether.
Many regulators and safety advocates, though, are sticking by the requirements for steering wheels so occupants can take control of their vehicles when they need to. John Simpson, director of the nonprofit Consumer Watchdog’s privacy project, says state-mandated reports from Google and other manufacturers testing vehicles in California show that, while their technology is improving, human intervention is often required. “That’s what the tech industry always says any time a sensible rule or regulation is being proposed: ‘Oh my God, you’re going to stifle innovation!’” he says. “I don’t think that’s true. When public safety is at stake, that should be the primary concern. Any regulations that are offered need to protect public safety.”
The rules from the California DMV for public use of autonomous vehicles, which are months away from being finalized, certainly take a cautious approach. The rules specify that manufacturers would have to pay a fee of up to $50,000 to sell their cars in the state. Manufacturers could only lease, not sell, their vehicles; they would have to get the vehicles and software -- including any software updates -- approved by a third-party validator; they would be required to share data with the state about any accidents, cyberattacks or switches away from the autonomous mode because of safety concerns.
Jude Hurin, the Nevada DMV’s top expert on autonomous vehicles, says California regulators are “overstepping their bounds” with such heavy-handed rules, taking what he calls a “federal approach” to regulating vehicles.
Hurin believes the California proposal would treat market-ready vehicles with the same wariness as it treats experimental vehicles. “You’ve allowed [manufacturers] to test on your highways for years,” he says. “So when they bring it to the consumer, you should have some kind of partnership or understanding that they’re not going to introduce something that’s totally threatening to individuals.”
Compared with California’s draft rules, Nevada’s regulations for public use of self-driving vehicles are modest. They require that the vehicles follow the state’s traffic laws, store crash data in black boxes, alert occupants if the driving software fails, and come to a stop if the occupants don’t respond. Both the cars and their owners must get state approval.
The rules fit in with Nevada’s embrace of companies developing the more ambitious version of driverless technology. Gov. Brian Sandoval has ridden in both a Google driverless car and a prototype of a driverless truck built by Daimler. The governor also announced earlier this year that he is creating a center for autonomous vehicles in Nevada’s economic development agency.
Daimler Freightliner self-driving trucks were demonstrated last spring in Las Vegas. (AP)
But the rush to regulate self-driving cars at the state level, outside of the first few pioneering states, has largely abated. For one thing, vehicle makers, particularly Google, have stopped pushing for state laws. They see state laws and regulations getting in the way of progress, rather than encouraging it. If every state issued a separate set of rules for the new cars, they argue, it could be nearly impossible to abide by all of them.
The federal government eventually came to the same conclusion. In January, the U.S. Department of Transportation and the National Highway and Traffic Safety Administration announced an ambitious goal of proposing uniform national best practices for both the auto industry and state policymakers within six months. To get there, federal regulators are likely to look at the regulations in California and Nevada as models. For manufacturers, though, state laws and regulations have become less important. For the most part, they have stopped asking for the blessing of state governments before testing their autonomous vehicles. They have decided they won’t need it. Texas lawmakers have never explicitly authorized self-driving cars, and yet the state’s capital city of Austin is Google’s No. 2 testing ground for its autonomous fleet.
A Google driverless car makes its way through Austin. (AP)
Bryant Walker Smith, a law professor at the University of South Carolina, believes automated vehicles are already legal in the United States, and wrote a 110-page law review article aimed at proving his point. To reach his conclusion, Smith combed through state laws, federal regulations and even international treaties. “It comes down to a question of interpretation or enforcement discretion,” he says. “A state or a community that is really supportive of automated driving may well conclude that the laws already support it. A state or community that is resistant to automated driving will likely reach the opposite conclusion.”
Supporters of driverless cars, Smith argues, can bolster their case without passing a law like the ones in California and Nevada explicitly regulating their use. “What I recommend,” he says, “is to start by doing a close, careful legal audit of existing law. Identify all of the potential provisions in the vehicle code and other laws that are relevant to automated vehicles. Then decide if those need to be changed.”
Many of the most difficult questions are deep in state vehicle codes. For example, New York requires that drivers always keep one hand on the steering wheel. There are laws against obstructing rearview mirrors, rules governing drivers’ conduct after an accident and regulations limiting the number of hours a driver can drive. Many of these would lead to absurd results when applied to autonomous vehicles. Laws against abandoning vehicles could inadvertently apply to self-driving cars in search of a parking spot. In fact, as Smith points out, it is unclear in many cases who the driver really is: the person who pushed the “go” button, the software engineer who programmed the car’s autonomous functions, or even the car’s absentee owner.
Beyond legal clarity, leaders who want to promote autonomous and connected vehicles in their communities can take practical steps to facilitate the process. One important step may simply be to remind the public about the potential safety benefits of automated cars. Roughly 34,000 Americans die every year in car crashes, and, by some estimates, human error contributes to more than 90 percent of those deaths. “The public should be concerned about automated vehicles,” Smith says, “but terrified about human drivers.”