There aren’t many ideas in the U.S. today that have the combined sociopolitical momentum and potential for radical disruption as autonomous vehicles do. They could prevent accidents. They could open up acres upon acres of space in cities for nontransportation uses. They could transform the automotive market to a shared-ownership model.
But even as the country rushes to embrace the technology, it’s not quite clear what an autonomous vehicle even is.
Generally speaking, an autonomous vehicle can both sense its environment and navigate without human input. But people working on all sides of the field — regulatory, software, automotive design — have pointed out that even slight discrepancies in the definition of an autonomous vehicle could lead to headaches when developing the technology. Manufacturers have asked state regulators in California to adopt private definitions so that they know exactly which of their products will need to meet regulatory standards and which ones won’t.
So what, exactly, is an autonomous vehicle?
It depends on who you ask. Look toward state-level definitions such as what's in place in California and proposed in New Jersey, and there’s a black-and-white approach: A vehicle is either autonomous or it isn’t. Turn to national sources like the National Highway Transportation Safety Administration (NHTSA) and the Society of Automotive Engineers (SAE), and there are documents outlining shades of autonomy.
The common ground between the two approaches is the general agreement that there is a threshold separating different types of vehicles, and that threshold exists at the place where vehicles are capable of controlling certain key functions of driving — steering, accelerating and braking chief among them.
The Black-and-White Approach
The California DMV, arguably the furthest along in terms of state agencies regulating autonomous vehicles, says an autonomous vehicle is “any vehicle equipped with technology that has the capability of operating or driving the vehicle without the active physical control or monitoring of a natural person, whether or not the technology is engaged, excluding vehicles equipped with one or more systems that enhance safety or provide driver assistance but are not capable of driving or operating the vehicle without the active physical control or monitoring of a natural person.”
Brian Soublet, legal counsel for the DMV, said that definition’s closest analogue within NHTSA’s definition is level three.
While representatives of the automotive industry have urged the state DMV to adopt the SAE levels — and that’s the approach of the Virginia Legislature in a bill it’s considering — California DMV representatives say their hands are tied.
“Our statute, which is an enabling statute for the [regulations], did not itself adopt the SAE levels,” Soublet said. “It just told us what is autonomous and what is not autonomous.”
That issue aside, Soublet said he sees no reason to cozy up to a level-based definition.
“Whatever we do has to be within the ability of the public to understand,” he said. “So you start throwing in things like ‘This is a … NHTSA level four vehicle,’ then you have to explain what a NHTSA level four vehicle is.”
And DMV Deputy Director Bernard Soriano admits there have been some instances when people developing autonomous vehicles were confused about whether their projects meet the department’s definition. Tesla Motors and BMW, for instance, approached the DMV to ask whether their self-parking systems make their vehicles autonomous in the state’s eyes.
They don’t, Soriano said — even if there’s nobody in the car when a human sends it off to park itself. The reason is that the person is still monitoring the vehicle.
That’s part of the California DMV’s approach, he said. If an AV developer is ever confused about the state’s definition, they can approach the department for clarification.
“That door is open for the manufacturers,” he said.
Even so, Soriano said, neither the NHTSA levels of autonomy nor the SAE levels offer perfect clarity either.
“Oftentimes there are situations where it’s ambiguous where something falls within level three or level two,” he said.
The Shades-of-Gray Approach
Though the language used is different, NHTSA and SAE both use a system of levels to describe automated driving. NHTSA offers a level zero for non-automated vehicles followed by four levels of automation, while SAE has a zero level and five automated levels.
A big difference between the two, according to SAE and Volkswagen representative Barbara Wendling, is that SAE’s levels focus on what the group calls the “dynamic driving task” — that is, all aspects of vehicle operation that aren’t strategic. Things like braking, steering and accelerating can all be handled by a machine, but determining where a car goes is still up to humans.
Between levels two and three is where the biggest shift happens. At level two, a human is still performing part of the “dynamic driving task.” At level three, the vehicle performs all pieces of that task itself — under certain conditions.
NHTSA’s levels follow a somewhat similar arc as SAE’s, but are written in simpler language and with less specificity, Wendling said.
Both NHTSA and SAE introduced their level descriptions under the banner of clarity. The distinction between vehicles based on functions of operation, SAE wrote in press material, should eliminate confusion between stakeholders. It is also based on trends among the various companies testing various aspects of autonomous driving.
“We need precision, and the SAE levels do provide precision when it comes to questions of scope,” said Wendling, who chairs SAE’s task force that crafts the descriptions of the levels of autonomy.
But the level-based approach has its doubters as well. Brad Templeton, an AV consultant who’s worked with Google’s self-driving car team, believes NHTSA’s and SAE’s approach only adds to public confusion. That’s because the level-based structure implies that vehicles will advance from one stage to the next in a neat and orderly line, he said.
“The thing that’s wrong about having levels is that if they do get into regulation, they have an idea that is almost surely false … that there is this progression and that things will happen that way,” he said.
But that’s not necessarily true — companies are skipping around from one level to the next. Templeton pointed to the Navio autonomous shuttle being used on college campuses in France as an example.
While Wendling said it’s possible that some people could expect autonomous technology to advance along the lines of SAE’s levels, she said that would be a mistaken assumption — the organization’s official document explaining the levels says that “they imply no particular order of market introduction.”
“If you actually read the levels and you take the time to understand the supporting definitions, you don’t come to that conclusion,” she said.
SAE, along with auto industry representatives, have been pushing for California to adopt its levels so as to avoid confusion about when regulations apply.
“There’s a continuum of technologies that are encompassed by your regulations,” said Emily Brascavoli, a representative of Ford Smart Mobility, at a California DMV workshop on its proposed AV regulations. “Some of the requirements you propose make more sense when talking about SAE level three vehicles than they do when talking about SAE level four vehicles.”
The United Nations’ Organisation for Economic Co-operation and Development has adopted SAE’s levels, as have stakeholders in Europe, Wendling said.
However, she admitted that the original SAE levels do involve a certain level of ambiguity when it comes to the lower levels of autonomy — so there are some cases where it wouldn’t be entirely clear which level a vehicle would fit into. The society is working to fix that, she said, adding that she is hoping to see a clearer version published in 2016.
“I hope that the people who criticize the first version of [the levels] will be pretty happy that we’ve tightened up and clarified things in the second version,” Wendling said.