There’s no driving test for self-driving cars in the US — but there should be

by Ehsan

There’s no driving test for self-driving cars in the US — but there should be


Tesla’s “Full-Self Driving” (FSD) driver-assist feature does not offer what its name promises.

Although Tesla CEO Elon Musk portrays the software as a prelude to fully autonomous vehicles, a quick online search shows videos of FSD-enabled Teslas in North America aiming straight at pedestrians, crashing into parked cars, and slowing down on a highway for no discernible reason. Unsurprisingly, more than a few observers have declared FSD to be a menace to traffic safety.

But if you look for similar videos of FSD-enabled Teslas stumbling their way through Europe’s cobblestone streets, you’ll come up empty. It’s not because Tesla’s system is better able to navigate San Sebastián than San Francisco; there are simply no FSD videos from Europe at all.

Why? FSD isn’t approved for public use there.

Tesla cannot deploy FSD anywhere in the European Union unless it first obtains a green light from regulators. To obtain that approval, Tesla must convincingly demonstrate that cars with FSD are at least as safe as those without it. At least so far, it hasn’t.

Unlike their European peers, American car regulators do not require — or even offer — any kind of safety preapproval for a new car model or technology. Instead, car companies “self-certify” that their vehicles comply with federal guidelines pertaining to everything from steering wheels to brake fluids. But no such rules address the driver assistance and autonomous technologies that are critical to the car’s future — and to the safety of everyone who walks, bikes, or drives.

Facing no significant oversight, automakers like Tesla can legally deploy any advanced driver-assist system (ADAS) they like, regardless of how dangerous it may be. According to federal law, only if the Department of Transportation’s National Highway Traffic Safety Administration (NHTSA) observes a pattern of dangerous problems can it launch an investigation (which NHTSA is now doing with Tesla), potentially culminating in a recall. Until then, the cars under investigation can continue zooming along American roads and streets.

Musk himself summed up the transatlantic distinction during a speech earlier this year in Berlin: “In the US things are legal by default, and in Europe they’re illegal by default.”

With its blurring of lines between driver and vehicle, the automation of cars is forcing US regulators to rethink their traditional approach. A federal official, who was not authorized to speak publicly, said that NHTSA is currently exploring how it could structure a preapproval process for autonomous technologies, a move that could finally force US carmakers to ask permission to deploy a new technology — rather than beg forgiveness after something goes wrong.

For the United States, it’s an opportune time to ask a fundamental question: is it wise to wait until after disaster strikes to protect Americans from dangerously designed vehicles?

Riding in an American automobile was a risky proposition during the early 20th century; 16 Americans were killed per 100 million miles driven in 1929, more than 10 times today’s rate.

Despite the carnage, federal officials paid minimal attention to car safety until outrage followed the 1965 publication of Unsafe at Any Speed, Ralph Nader’s explosive bestseller. A year later, Congress enacted the first comprehensive federal rules for car safety, and NHTSA was born in 1970.

Even as federal officials weighed various automotive regulatory frameworks, they never seriously considered forcing carmakers to obtain preapproval for a new vehicle model or component.

Lee Vinsel, a Virginia Tech professor of science, technology, and society who wrote a book about the history of auto regulations, said that he found no records of public leaders of the era even raising that possibility. Instead, federal officials chose to put car companies in the regulatory driver’s seat, and the process they established 50 years ago remains intact today.

Here’s how it works: the encyclopedic Federal Motor Vehicle Safety Standards (FMVSS) establishes rules for any car sold for use on public roads, touching on everything from the strength of door hinges to the frequency of windshield wipers. NHTSA can and does update FMVSS to incorporate new technologies and features, but the process moves glacially.

“The expectation is that a new FMVSS element will take 10–15 years before coming into effect,” said Daniel Hinkle of the American Association for Justice, a trade group for trial lawyers.

Manufacturers show their adherence to FMVSS through a process known as self-certification, which works largely as it sounds: car companies simply affix a label to each vehicle signaling compliance. Although NHTSA conducts spot checks, carmakers seem largely to abide by FMVSS; NHTSA launched only 90 investigations into FMVSS violations in 2020.

“It’s unusual to see vehicle crashes that are specifically due to FMVSS non-compliance,” Hinkle said.

NHTSA only brings the hammer down with a recall if an investigation uncovers a pattern of safety problems on public roadways. Such investigations take months; meanwhile, Americans continue to drive the faulty or dangerously designed vehicles.

Notably, the US adopts a much more proactive safety posture toward aviation safety. When an airplane manufacturer wants to build a new kind of plane or alter a component, the company must work with the Federal Aviation Administration (FAA) to obtain approval prior to deployment.

Crash data suggests that that framework has served flyers well. According to research by Northwestern University economist Ian Savage, the United States experienced 0.07 aviation fatalities per billion passenger miles from 2000 to 2009 — around 1/100th the rate for those driving in a car or truck. But the FAA preapproval system doesn’t come cheap; according to a 2014 Department of Transportation workforce assessment, the FAA employs over 6,000 people working on vehicle safety while NHTSA had just 90. To put another spin on those figures: FAA employed over 10,000 enforcement staff for every 100 aviation deaths, while NHTSA had 0.3.

“It’s not unusual to have a year where there are zero deaths in commercial aviation in the United States,” said Transportation Secretary Pete Buttigieg at the SXSW festival earlier this year. “But on the roadways, we basically take it as a given, as normal, as the sort of cost of doing business, that thousands and thousands of people will die every year.”

We also take it as normal that the FAA’s regulatory model, centered around preapproval of new planes and technologies, is not applicable to automobiles — despite an American roadway death toll that is horrific compared to other developed countries and only getting worse.

Across the Atlantic, Europeans are far less likely to die in a car crash than Americans. In France, for instance, the per capita roadway death rate is only a third as high as in the US.

There are numerous reasons for the gap, including higher transit ridership in Europe (traveling by bus or train is far safer than going by car), slower urban car speeds, and less driving per capita. Europe has also adopted automotive regulatory standards that are stricter than those of the FMVSS.

For instance, new vehicle models sold in the European Union must include intelligent speed-assist technology that sounds an alarm or applies resistance to the accelerator if a driver exceeds the posted speed limit. No such requirement exists in the US.

From a process perspective, the European system of automotive regulations works more like the US’s hands-on approach toward aviation than its laissez-faire oversight of motor vehicles.

Before a new vehicle or component can be released to the public, the car company must obtain preapproval known as type approval from an EU member state. Obtaining that green light requires time and money as regulators review data and conduct inspections and tests to confirm compliance with EU law. And if a car company wants to deploy a technology that isn’t yet regulated by the EU, the automaker must demonstrate that the proposed feature is at least as safe as a vehicle without it — a requirement that has a profound impact on the deployment of advanced driver-assist and autonomous technologies.

European car rules have real teeth; in 2019 they forced Tesla to adjust several elements of the company’s Autopilot ADAS system, such as limiting how sharply the steering wheel can turn when Autopilot is active, and the EU recently forced Tesla to make certain Autopilot functions available to all drivers, not just those who hit gamified benchmarks. European regulators have yet to approve any deployment of FSD.

Antony Lagrange, the team leader for automated and connected vehicles and safety at the European Commission, makes no apologies for Europe’s regulatory assertiveness. “We’re talking about very complex, safety-critical products,” he said. “We need to ensure that these products are safe for production — and that they continue to be safe after they’ve been released.”

The same holds true for over-the-air updates, an increasingly common method of updating car software that avoids forcing the owner to visita dealership. In America, automakers are free to issue such updates whenever they wish, but in Europe, they must obtain preapproval from European regulators.

That requirement implicitly encourages automakers to limit the frequency of over-the-air updates, which may itself make cars safer. “One thing that type approval has going for it is that it forces car companies to measure twice and cut once,” said Ed Niedermeyer, a co-host of the Autonocast podcast. “In the United States, there is no regulatory cost of putting out a mediocre product and iterating. It makes sense to try and force the companies to get it right the first time.”

That sentiment was echoed by Jascha Franklin-Hodge, chief of streets for the city of Boston who previously worked in Silicon Valley. “My worst fear,” said Franklin-Hodge, “is that carmakers adopt a software development mentality, with an acceptance of errors that constantly require correction.”

Is the European regulatory system of automobile type approval superior to the American norm of self-certification? It depends whom you ask.

Type approval processes are expensive, both for governments employing a small army of engineers to test vehicles and for car companies forced to navigate complex regulatory systems before they can sell to the public.

American carmakers would prefer the status quo. “The current self-certification model should be preserved, since this framework has worked well in the US,” said John Bozzella, head of the Alliance for Automotive Innovation, which represents almost all of the legacy automakers.

Finch Fulton, who served as Deputy Assistant Secretary for Transportation Policy in the Trump administration, agreed. (Finch now works at Locomation AI, an autonomous trucking company.) “If you were to shift NHTSA towards a type approval model, there are no clear benefits, but the increase to NHTSA’s budget would likely be astronomical,” he said.

Missy Cummings, a former Duke engineering professor who now works at NHTSA, previously praised the FAA’s type approval model as a “comprehensive” approach to safety, while noting that it can “also be extremely costly and add significant time to the deployment of new technologies.”

Of course, it might not be a bad thing to add a few more engineers to NHTSA’s ranks or to slow down the launch of a work-in-progress system like Tesla FSD. The arguments in favor of preapproval strengthen as automobiles morph into three-ton computers on wheels, where a system malfunction could have catastrophic consequences. Indeed, technologically complex, partially automated cars seem similar to the airplanes that the US has long regulated with type approval — with a sparkling safety record, especially compared with the carnage on US roadways.

In 2016, NHTSA issued guidance about regulating autonomous vehicles that explicitly cited the FAA’s use of preapproval as a model, noting that “its challenges seem closest to those that NHTSA faces in dealing with highly-automated vehicles.”

Six years later, the agency seems to be giving preapproval new consideration: a source said that NHTSA is now developing a pilot that would empower the agency to weigh in before a new automated technology is deployed on public roads.

Asked for comment, an NHTSA spokesperson would say only that “the agency is committed to continually evaluating its authorities and processes to help ensure the safe development of advanced vehicle technologies.”

Kelly Funkhouser, Consumer Reports’ manager of vehicle technology, said she would support such a move. “Our government just can’t move fast enough for self-certification to be an effective strategy,” she said “We need pre-approval to be brought into the conversation.”

Still, the very idea of questioning self-certification is so sensitive that Buttigieg quickly changed the subject in an interview earlier this year, and NHTSA declined an invitation to comment on the record for this story.

The topic may be uncomfortable, but it is becoming unavoidable. Beyond the surge in American roadway deaths, the country’s existing regulatory framework is ill-equipped to handle the ascent of automated car technology. One way or another, things will have to change.

Consider how ADAS systems like Autopilot are already muddying the traditional line between state-licensed “drivers” and federally-approved “vehicles.” In 2016, a Tesla with Autopilot activated was traveling at 74 mph near Williston, Florida, when it slammed into a turning trailer, slicing off the roof and instantly killing the driver.

Following an investigation, the National Transportation Safety Board cited the Tesla driver for failing to pay sufficient attention as well as Tesla itself for designing Autopilot to allow “prolonged disengagement” from the driving task and making it easy to activate Autopilot on road types where it isn’t meant to be used. Should federal or state officials take the lead in preventing a recurrence? It isn’t clear.

The state / federal divide grows even murkier with Level 4 vehicles, capable of being operated without any engagement from a driver on certain roadways. States like California require AV operators to obtain a permit to deploy a Level 4 vehicle on state roads; cities like New York City have dabbled in such regulations as well. But NHTSA sees Level 4 vehicles as falling under its own purview; in 2020, the agency forced EasyMile to halt its self-driving shuttle operations after an injury in Columbus, Ohio.

As we enter an era of automated driving, how can regulations best protect American road users? To date, conversations about adjusting car regulations for autonomous technology have ignored the weaknesses of self-certification, focusing instead on removing obstacles that might slow new deployments. For instance, federal legislation has been introduced that would preempt states from regulating autonomous vehicles and empower NHTSA to exempt more AVs from FMVSS safety requirements. That’s hardly a recipe for ensuring safe technology deployment.

Still, Buttigieg does seem to recognize the need to chart a different path for automaker oversight. “So many of our regulations to keep cars safe are based on how cars always used to be,” he said earlier this year. “We need to make sure that they’re based on how cars are going to be.”

Any movement toward type approval would be a heavy political lift, facing opposition from a recalcitrant auto industry as well as fiscal hawks wary of adding to NHTSA’s budget. But the transition need not happen all at once; it could begin with a handful of preapproval checks for obvious safety problems, like phantom braking. Faced with a more assertive NHTSA, carmakers might rethink the wisdom of putting work-in-progress ADAS or autonomous systems on public roads.

With road deaths already at a 20-year high, we need fewer risks — not more of them — on American streets and highways. Type approval could be a critical tool to help navigate the autonomous era to come.

Lucas Peilert provided research assistance


Source link