Legal Expert Bryant Walker Smith Identifies Strategies for States on Autonomous Vehicle Policy

CSG convened the Autonomous and Connected Vehicle Policy Academy June 12-14, 2017 in Detroit. A group of state policymakers from around the country attended the event. The academy included a special briefing by Bryant Walker Smith, assistant professor at the University of South Carolina in the schools of law and engineering. He spoke about the legal and regulatory landscape for autonomous vehicles.

Seven years ago, Bryant Walker Smith wanted to determine whether it was legal for autonomous vehicles to test on public roads around the country so he took it upon himself to read the state vehicle codes of all 50 states. His conclusion was outlined in a white paper originally issued in 2012. Its title: “Automated Vehicles Are Probably Legal in the United States.”

“Automakers and others have been testing their systems on public roads now for years with a trained safety driver in the front seat paying attention and intervening when necessary and that, almost unambiguously with a couple of exceptions in some states, is perfectly consistent with existing law,” Smith told policy academy attendees on June 14. “It’s one of the reasons why automakers and others have belatedly said ‘y’know what states, we’ve actually got the testing thing and we don’t really need any new laws specific to that in our state.’ Because unless you’re in New York, which still requires you to have one hand on the wheel at all times while the vehicle is in motion, there are very few laws that are inconsistent with simply letting your car drive while you’re paying attention.”

But Smith said there could be a couple of different reasons many automakers have been telling state policymakers to steer clear of autonomous vehicle legislation that could prove too restrictive in the future.

“That might be because they don’t see any changes that are needed and you can make a good faith argument that existing law is actually consistent with even more aggressive forms of automated driving,” he said. “You could even make an argument that in some states existing law is consistent with a vehicle operating with nobody in it. … But what you might be seeing from automakers as well is uncertainty about what they want—companies that don’t yet know exactly what their technologies, applications and business cases will be and until they figure that out, they’d prefer that you do nothing rather than do something that will ultimately turn out to be inconsistent with their vision.”

One notable exception to the automakers walking the halls of state capitols trying to quash legislation in 2017 has been General Motors.

“GM has been a really interesting case study because … they have decided that they know what they want and they’re aggressively pushing it,” Smith said. “That’s a sea change for GM, which in 2012 was instrumental in getting the first defeat of automated driving legislation in Arizona with the argument that this is too soon, this is not necessary.”   

Despite admonitions from many of GM’s competitors however and the legal opinions of those like Smith who say existing background law is probably sufficient, many states have chosen to move ahead with legislation, primarily as a means of attracting attention, Smith said.

“Many states have approached this as a question of economic development to say ‘if we pass a law, we can get self-driving cars to come to our state. If we pass a law, then everyone will realize that we have snow and that makes us a great place to test and then we’ll get all of the action,’” Smith said. “The reality has been a bit different. In many cases, the states with automated driving activity don’t map very neatly into the states with legislation.”

Smith noted that Washington and Arizona are among states that have not had legislative efforts on autonomous vehicles but have taken a broader approach.

Just the week before the policy academy convened, Washington Gov. Jay Inslee signed an executive order requiring state agencies to support safe testing and operation of autonomous vehicles on public roads in the state, establishing an interagency work group and enabling pilot programs throughout the state. On July 12, the governor announced the state’s first certified autonomous vehicle pilot test with a Virginia Tech-affiliated company called Torc Robotics.

Arizona Gov. Doug Ducey signed an executive order in 2015 that was similar in scope to Washington’s.

Smith said the approaches of these states and others demonstrate it’s not necessary to equate encouraging automated driving with passing a law that may prove more restrictive than existing background law or that may not even engage with existing law and ultimately raise more questions than it answers.

“So for example, you have a state that declares automated driving to be legal as Florida does but then doesn’t address what that means for the user or the driver of this vehicle,” Smith said. “Does it mean that they are prohibited from doing the kinds of distracted activities in a car that state law might otherwise prohibit? How does that map the obligations of the vehicle code onto some entity—the user, the engineer, the company responsible for the system? And that kind of engagement has often been missing from these initial state law efforts.”

Smith also pointed to two other states that have proven to be cautionary tales in their approaches to autonomous vehicle policy: California and Michigan.

“California stands as an example of a very slow (regulatory) process that has been counterproductive in some ways,” he said. “At one point, the department of motor vehicles, which was responsible for developing regulations, was about three years late in one of its regulatory packages. One of the things that is particularly fraught is initiating a slow-moving, complex rulemaking process at the state level that involves resources and expertise that may be well beyond the abilities of a state department of motor vehicles.”

Some industry officials were particularly concerned about California requiring a permit to deploy autonomous vehicles on public streets even if the vehicle has already been approved by the National Highway Traffic Safety Administration (NHTSA) and giving the state DMV the authority to yank a permit if the state deems the vehicle unsafe. Smith said that’s similar to what states can do now with conventional vehicles through the registration and state inspection processes. But for autonomous vehicles there are some distinct differences.

“When you register a vehicle individually, a state determines whether that vehicle is safe and DMVs only register reasonably safe vehicles and they are often empowered or even directed to deregister unsafe vehicles,” he said. “Unfortunately, what that tends to mean is ‘do you have a door on your car and is your mirror taped sufficiently to the side of it’ rather than ‘is the automated driving system or advanced electronics performing appropriately in a reliable way?’ Evaluating that second question is just not something that states are going to be in a good position to do.”

Nevertheless, Smith does see some value in having a mechanism at the state level to enforce the safety of individual autonomous vehicles.

“I would prefer to locate that in registration and through registration indirectly impose requirements on developers to provide information about their systems, to identify in a broad sense the capabilities and limitations to a state or states, to provide a central clearing house so that a (vehicle identification number) is linked to a database that is accessible to all states that has information about that particular vehicle for the purposes of enforcement, inspection and investigation,” he said. “But I would not create a separate regime like California has done for a separate permit and I certainly wouldn’t replicate that in all 50 states. The registration process, given that it already has some notions of reciprocity, could be used to coordinate that effort among states. A separate permitting process likely could not.”

In addition to California, Smith had little love for Michigan, where lawmakers approved a series of measures last year aimed at putting the home of the automotive capital of the world on the autonomous vehicle map.

“Michigan is a state that did not envision a regulatory process but did put together a mesh of laws that are so confusing and self-contradictory that I read them dozens of times and still can’t figure out what they actually mean or what they actually do,” he said. “They are a true bewildering mess. This is again a reflection of the complications and pitfalls of moving so quickly legislatively in this field.” (For more on Michigan’s approach, read this article summarizing the remarks of policy academy keynote speaker Kirk Steudle of the Michigan Department of Transportation).

Autonomous Vehicle Strategies for States

So what actions can state governments take to put down their markers in the autonomous vehicle marketplace of ideas? Smith identifies about 50 strategies for states in his recent article “How Governments Can Promote Automated Driving.”

Smith said executive orders, advisory opinions, memoranda and other documents can give a state something to demonstrate that they’re doing something but in a flexible way.

“Something that a developer of these systems if they’re looking for that confidence can go and read and have and show to say ‘we have a state that is committed to us but not necessarily one that is going to entangle us with rules that might be difficult to change or not particularly applicable to our vision for automated driving,’” Smith said.

Before taking any action it’s important for state officials to be on the same page about what they want to achieve, whether that is clarifying the legal status of automated driving, promoting automated driving or preparing for a future that includes automated driving, Smith said.

“One of the best ways to promote automated driving is to identify a structure within your state, vertically from the governor’s office down to the local police chief and horizontally within the community,” he said. “All of the safety organizations, disability rights organizations, transportation planners, business developers who have an interest in and enthusiasm for this technology so that they are connected to each other and so that a developer interested in these technologies has somebody to call, has a state Sherpa to help navigate them through the legal framework in the state and provide them with the level of certainty that they may need or want to conduct an activity in the state.”

But promotion of automated driving should not be the end goal itself, Smith believes.

“States should have an open discussion about what the broader policy goals are and how automation can serve that and that way automation is not being billed as a panacea and an ideal in and of itself into which we should all put great faith but rather as a tool to be used toward things that people can agree that they want,” he said. “The goal should be to use automated driving as a tool to advance broader societal goals. … Do you want safer roads? Do you want greater mobility for people? Do you want to reduce congestion? Do you want to provide services to people who otherwise would not have them because of ability, disability, income, other needs?”

When states do consider autonomous vehicle legislation, Smith said, there are specific areas policymakers should seek to avoid. Vehicle and software design standards for example are best left to NHTSA. And while fully autonomous vehicles may present profound ethical conundrums and the opportunity for intriguing parlor games, those should have no place in state legislation, Smith believes. Case in point is something called “the trolley problem:” A vehicle is moving forward and can go either left or right. It can’t stop and in its path on the left is a young child. On the right is an old lady.

“Who do you kill?” said Smith. “Do you kill the young child or the old grandmother? And people will say ‘this is the fundamental problem with automated driving. This is why we can’t have it.’”

But Smith believes this ethical puzzle about how to best program autonomous vehicles to make that Sophie’s choice has been much overhyped.

“When you are drafting legislation, your legislation probably does not need to specify who should be killed because the reality is that these systems are not going to be omniscient or omnipotent enough to make these granular determinations in a prospective, predictive way,” he said. “Furthermore, at the point that a system needs to choose who to kill, chances are it’s already made a mistake earlier on. Chances are the speed is too high, a risk has been taken that should not have been taken that got the system into that situation. So broadly speaking, we need to think about ethics in design and ethics in appropriate risk taking but that is a much broader set of questions than this narrow problem of ethics.”

Smith does see another important role for state policymakers as autonomous vehicle technologies roll forward: managing consumer expectations.

“First, so that people continue to see the potential benefits of these systems even when there are unfortunate crashes and incidents when public opinion fluctuates wildly that there is that credibility within government and within developers to explain the benefits in a realistic way,” he said. “It’s important in the other direction that expectations aren’t unrealistically high, societally that people think ‘oh the technology is ready everywhere all the time’ and individually so that people don’t misuse the systems at lower levels of automation.”

Policymakers can also emphasize the importance of public engagement and transparency by the industry, Smith said.

“The notion of the public safety case … where developers are expected to share their safety philosophy to be very candid about what’s working and what’s not in a way that matures the discussion,” he said. “To force that kind of transparency and honesty is part of the regulatory process. That provides the candor that I think the public needs to understand not just the benefits but also the challenges of the technology.”

However, when it comes to the concern some have that consumer acceptance of autonomous vehicles could be a barrier to achieving an autonomous future, Smith is unconvinced. He uses the analogy of the smart phone as a product consumers didn’t think they needed or wanted—until they did.

“Companies are going to recognize that that’s an essential part of their business case and are going to make the argument directly to the public,” he said.

But what will the autonomous vehicle industry actually be selling and how will they be selling it? The answer to that question may have significant policy implications for states as well, Smith said. The traditional model has been that a driver buys a car, drives it for about 12 years and perhaps sells it to another owner or disposes of it in some way and then buys another car.

“There are some (auto) companies that are still committed to that approach for automated driving long term although even those companies are likely to have some kind of service aspect—maybe a subscription navigation service, maybe a coordinated network for sharing your vehicle with others, maybe an asserted ownership interest in the vehicles so that they can take it back or simply shut down some of the functions at some point in the future when they deem it no longer safe,” Smith said. “But many other companies are at least tiptoeing into automated driving with a much more service-oriented model: shared rides, fleets, other systems that allow these companies to carefully manage the maintenance, the use, the deployment, the recall and ultimately the termination of these services. That matters from the regulatory perspective because as you’re examining the relationship of existing or new laws to these systems, it’s not just a question of regulation of the vehicle sale or the vehicle registration. Now it implicates taxis, transportation network companies, limos, maintenance, fleets, dealer models, all of which are subjects of state or local law.”

Evolution of Autonomous Vehicles

One of the other reasons policymakers may face a challenging task in regulating the industry is that autonomous vehicle technologies are evolving in a number of different ways simultaneously. The Society of Automotive Engineers (SAE) International has identified six levels of driving automation for on-road vehicles ranging from Level 0 (no automation) to Level 5 (full automation). But Smith noted the road to Level 5 is not necessarily going to be a linear path.

“Level 5 was really defined as an ideal, an aspiration to which we might never really get but we put it there for completeness—the opposite of level 0,” Smith said. “But level 4 is a system capable of driving itself without any human intervention within a limited subset of conditions. Those conditions might be as small as one road on one university campus during the day when it’s not raining and when it’s not too sunny at 25 miles an hour. Those conditions could be as large as the entire southwest of the United States.”

Smith said that could be one of the paths to automation everywhere all the time: concentrating on Level 4 automation in larger and larger controlled environments—or operational design domains as they’re called—until the technology becomes more demonstrably reliable to implement on a larger scale.

“There are no level 4 vehicles out today,” he said. “The reason for that is that every vehicle that is on the roads is supervised by a trained safety driver so these systems may aspire to be level 4 but they’re not and this is something that the public and I think legislators as well often miss—the recognition that there is a huge difference between a system that is being supervised by a driver and a system that is not.”

But concentrating on achieving level 4 automation may not be the only path to full automation.

“Another approach is to say ‘we can actually get some low-hanging fruit by assisting the driver’ and these are the (Level 2, Partial Automation) driver assistance systems that … (are) commercially available now in some vehicles,” he said.

Under level 2, those systems may automate some aspects of steering, acceleration and deceleration using information about the driving environment but the human driver is required to perform all remaining aspects of the driving task.

The question now for some is whether getting from level 2 to level 4 (and presumably, eventually level 5) will require a stop at level 3, also known as conditional automation. Under level 3, the automated driving system controls all aspects of the driving task with the expectation that the human driver will respond appropriately to a request to intervene should something go wrong.

“Some (automakers) will say ‘it’s very important that there always be a minimal risk condition, that the driver never be expected to resume actively driving once the system has started,’” Smith said. “Others will say that’s making a business case determination—that is legislating a particular technology rather than being technology-agnostic because those companies do see a safe use case for these level 3 systems, where somebody need not pay attention but under certain circumstances may need to resume actively driving.”

Tesla has deployed level 3 technology with the Autopilot feature of its operating system. A driver was killed last year in Florida when his Tesla Model S rammed into the back of a semitrailer. An investigation by the National Transportation Safety Board found that the driver had been using Autopilot for more than 35 minutes while barely touching the steering wheel despite repeated visual warnings from the driving system that flashed the message “Hands required not detected.” While Tesla considers Autopilot a safety mechanism, the driver in this case used the feature as if it were a fully autonomous driving system and that dichotomy between industry intent and consumer understanding is where danger lies, some believe. Ford and Google are among the companies now saying autonomous vehicle development should skip level 3 altogether because it leaves too much to chance. 

“So this is another area where you’ll see these tensions play out underneath the surface in the kinds of requirements that a company might suggest are appropriate for legislation—whether or not there needs to be a minimal risk condition,” Smith said.

For those state officials still interested in how to structure autonomous vehicle legislation, Smith has recently authored a "Model State Automated Driving Law," which is availabe at his website, Law of the Newly Possible.