The Department of Transportation just issued a comprehensive policy on self-driving cars

Automated vehicles (AVs), colloquially known as "self-driving cars," are no longer sci-fi speculation. They are happening. Multiple vehicles with varying levels of automation are already on the road, and numerous companies in and outside the traditional automotive market are working furiously on greater levels of automation. Several companies expect to debut fully automated (driverless) vehicles by 2020.

Right now it’s the Wild West out there. No one is quite sure how AVs are going to work, how they’ll be regulated, who will insure them or bear liability for their mistakes, or what sort of software or data protocols they might share, among other things. (The National Highway Traffic Safety Administration has released some statements on AVs since 2013, but nothing comprehensive.)

This uncertainty over standards, along with worries that every state will develop its own bespoke regulations, threatens to constrain the development of AVs.

So the US Department of Transportation is attempting to get ahead of the curve. On Monday, it released a surprisingly far-reaching "Federal Automated Vehicles Policy."

The policy attempts to do all sorts of things — we’ll get into the details below — but the overarching motivation is that DOT wants to accelerate the development and adoption of AVs.

DOT views AVs as a safety technology that could reduce some of the 38,000 traffic fatalities a year in the US, 95 percent of which are caused by human error. It also sees AVs as an accessibility technology that could provide personal transportation to whole populations (disabled, elderly, etc.) who have lacked it.

The DOT is not neutral toward AVs. It wants to get them on the road soon. That’s a big deal.

The policy comes in four buckets.

1) What the vehicles need to do to be safe

The "vehicle performance" section lays out a 15-point safety assessment, so that AV developers and manufacturers know the sorts of things that federal regulators will expect. It covers everything from cybersecurity to data collection to crash response. I won’t run through all 15 points, but a few terms are especially interesting.

Operational Design Domain has to do with when and where an AV will be fully in control — what conditions have to be met for it to be fully autonomous, and what circumstances might call for the driver to reassert control.

Minimal Risk Condition is the state an HAV reverts to if circumstances change such that it is outside its Operational Design Domain — in other words, if it can’t drive itself anymore. Will it pull over? Put on its hazards?

The Human Machine Interfac refers to how AVs communicate relevant information to their tragically unreliable wetware occupants. What common standards should all such interfaces meet? Should there be a universal set of signs and signals?

And then there are "ethical considerations." AVs will have to make life-or-death decisions. What sorts of factors should they take into account? How should they weigh human lives and property damage? How exactly will DOT regulate the ethics that engineers program into AV software?

There are lots more of these, some of which raise as many questions as they answer.

2) What federal and state governments need to do

The second section addresses the division of responsibilities and authorities between the federal government and state governments, and suggests a model policy that states can adapt for their own use.

The feds will retain their authority to set and enforce safety standards, communicate with the public about safety, and occasionally issue guidances about how to meet national standards.

States will retain their authority to license human drivers and register cars, set and enforce traffic laws, and regulate vehicle insurance and liability.

As administration officials clarified on a subsequent phone call, this basically means that when humans are driving, states are in charge, but when software is driving, the feds are in charge. Automation is part of vehicle safety, and that is NHTSA’s domain.

For states that want to regulate the testing and deployment of HAVs, DOT has some suggestions for state policy, from how to administer tests to licensing and registering HAVs to insurance and liability coverage.

3) How DOT will use its existing regulatory tools

There are three broad ways that DOT communicates about standards with automakers. It is promising to speed up all of them in regard to HAVs.

The first is "letters of interpretation," by which automakers inquire how existing statutes apply to particular kinds of automotive technology. These have traditionally taken "several months to several years" for NHTSA to answer, but in the case of HAVs it is pledging to reduce that to 60 days.

The second is exemptions, by which automakers can request that specific test fleets with new technologies be exempted from specific existing standards. These too have taken years; DOT has pledged to reduce it to six months.

The third is rulemakings, which is how DOT makes new standards. These may be required for genuinely novel automotive technologies; they too will be expedited.

4) DOT may need brand new regulatory tools to deal with AVs

DOT is considering a range of new authorities that may be necessary to properly regulate HAVs. There’s a whole list. Below are a few I thought were intriguing:

For HAVs, DOT is considering moving away from its current system of self-certification, by which automakers vouch that their new technologies meet federal standards, to "pre-market approval," by which automakers clear new technologies with DOT before they hit market. That’s a higher level of scrutiny and likely to be unwelcome among automakers and tech nerds. The challenges are discussed in the policy.

DOT is also considering "cease and desist" power, by which it could tell a manufacturer to immediate halt production if some unanticipated safety concern becomes evident.

Then there’s "post-sale regulation of software changes," which could become a very interesting area of policy over the next decade. In HAVs, software updates will arguably be more frequent and meaningful than hardware upgrades. That puts DOT in the position of regulating a strange new world, outside its usual ambit. (Though in its defense, nobody is expert on HAV software yet.)

It’s also considering new requirements for data collection, which will be a crucial and contentious area of policy itself. As HAVs gather data about road conditions and safety stratagems, do they share it with one another, so that they all get smarter? That seems like the socially optimal way to accelerate the safety benefits of HAVs, but it may cause friction with the intellectual property concerns of automakers. What data will be shared what proprietary?

A shared language for defining vehicle automation

One further note of interest, for the nerds: DOT has officially abandoned the NHTSA’s own levels-of-automation classification in favor of SAE’s, which is preferred by industry. This calls for a new Vox graphic!

(SAE International)

(See also SAE’s "Taxonomy and Definitions for Terms Related to On-Road Motor Vehicle Automated Driving Systems.")

Automated vehicles are getting realer and realer

So that’s how the feds are currently thinking about AVs.

People are always bashing government and highlighting government failures, so it’s worth saying that this seems like a judicious and farsighted move on the part of DOT.

It recognizes that AV developers and state regulators need guidance on what’s expected of them, and that no one wants to end up with a patchwork of inconsistent state policies. It also recognizes that it’s counterproductive to impose too many prescriptive rules and standards on a fledgling technology. (Silicon Valley and automotive companies begged the NHTSA for a light regulatory hand.)

To my eyes, the agency has pretty deftly walked a middle path, offering guidance substantial enough to bring some clarity to the market but broad enough to allow for plenty of innovation and competition. And it built in a yearly review of the standards, to ensure that they stay current. It’s a sign of a more nimble DOT, which will definitely be necessary in the coming age of AVs.

It is somewhat mind-boggling that fully automated vehicles are expected to show up on the market in the next two to four years — not just in our lifetimes, in our pets’ lifetimes. And they will fundamentally change how we get around. The questions DOT is pondering are soon going to become urgent, contested issues, with lots and lots of money and lives at stake. Let’s hope we can all keep up.