Who Will Regulate Autonomous Vehicles Best?

It’s not clear yet whether the AV START Act will pass the U.S. Senate and become a law. What is clear is the first effort at creating a national safety standard for connected and automated vehicles (CAVs) wasn’t the most effective tactic.

The bill requires the National Highway Traffic Safety Administration (NHTSA) to pre-empt regulations and progress achieved by the states wasn’t the most effective tactic. But pre-empting work done in the states to get a handle on autonomous-vehicle trials has raised some complaints. And offering CAV developers permission to leave important safety gear off the vehicles they put on the road – as a way to cut costs and red tape and therefore make development faster – is generating a lot more complaints.

In fact, more than 100 organizations have raised concerns, ranging from associations of mayors and governors to coalitions of clubs for bicyclists, hikers and others worried about being run over and left behind. Add to that list scores of consumer-advocacy groups that are deeply concerned about the danger of having their hometowns flooded with CAVs that are not yet smart enough to avoid every accident, but which also may be driving around without important safety equipment.

NHTSA is roughly two years behind where experts say it should be, and bills like AV START are showing up with no agency support largely because “the federal government was caught flat-footed by autonomous vehicles and it has no idea what kind of advice to give,” according to Roger Lanctot, director of automotive connected mobility at Strategy Analytics.

Big legacy car companies backed the rule because they want the liability protection that government regulation can provide to them, but some states are headed toward the opposite end of the spectrum when it comes to regulating CAV trials. “States like Florida, Arizona, Virginia are saying, ‘Sure, come on in with the appropriate prudence. You’ll have to bear any responsibility, but go right ahead with your AV test,’” Lanctot said.

State departments of transportation split control over the car and driver with NHTSA, which maintains the Federal Motor Vehicle Safety Standards (FMVSS) that define the safe operation of the car itself.

States are responsible for the car registrations, and for the drivers – training them, licensing them, ticketing and traffic schooling them, and are quite good at making ordinary people less dangerous in a car than they would otherwise be, according to Philip Koopman, associate professor of electrical and computer engineering at Carnegie Mellon University, who specializes in safety testing and validation of autonomous vehicles.

New construction
But CAVs are new, and those who know them best are those who have lived with them longest. That includes state departments of transportation, which have been learning alongside of the CAV companies what makes a self-motivated vehicle tick. The Pennsylvania DoT, based close to the birthplace of self-driven cars at Carnegie Mellon University, has been on the job longer than most CAV-friendly agencies, said Koopman.

So it would be useful to have a look at the experience of Roger Cohen, senior advisor to the secretary of transportation for the Commonwealth of Pennsylvania, who Koopman described as “the best guy, leading the best autonomous car program of all the states.”

Pennsylvania’s first CAVs rolled out of Carnegie Mellon University in Pittsburgh more than a decade ago on the way to winning the DARPA autonomous driving challenge, which gave Cohen and company the sense of inevitability about the whole thing.

“You can’t actually do as much as you think to keep autonomous vehicles out of the city if you want to,” Cohen said. “So when it became clear the private sector was coming in, we designed model legislation to suggest to the general assembly to authorize full, automated vehicle testing at all levels, with PennDoT designated as the agency charged with oversight for safety. We thought it was a critical part of the legislation that we be given authority to oversee the process, and we were—but only through the mechanism of policy, not regulation. “

The Pennsylvania legislature has never been quite sure what to make of autonomous vehicles, so it has been cautious about peremptory decisions, authorizing the CAV program that allows Cohen a measure of influence but not the ability to make the rules.

“As it turns out, policy can leave you a lot more nimble and flexible and better able to keep pace with the increasing advancement of tech than regulation, which takes a year two to get approved and is then very rigid,” Cohen said. “There was an upside and a downside. We could respond with appropriate policies and create a framework we could later present to be formalized as regulation. But we also could build a collaborative relationship with stakeholders and with all six testing entities that are active in the Pittsburgh area and around the county.

“PennDot drafted some policies that looked like regulations, but were only policies. But I laid those in front of them and spent six months of meetings, twice per month, finding common ground and giving room for people to state a dissenting opinion,” Cohen said. “We came up with areas of common ground that gives us a much better idea of what other stakeholders are up to or interested in. We collect information from testers at a certain point in the process so we know what’s going on, and we offer voluntary guidelines.”

Low-key as the approach is, Cohen wasn’t afraid to offer NHTSA a little advice along with the feedback it requested on AV 3.0, which is the third version of a gradually expanding outline. The agency considers this to be an ongoing discussion, in which NHTSA suggests where the market should go but doesn’t decide anything.

Fig. 1: NHTSA’s AV 3.0 plan. Source: U.S. DOT

In addition to feedback on guidelines, the PennDOT suggested NHTSA keep in mind that it is important to have more than just regulations and the threat of enforcement in the toolbox. It’s also necessary to know what all those CAVs are doing and if their developers have described them correctly, which means it is incredibly important to have a skilled independent third-party validation of the ADAS and CAV systems an agency is supposed to regulate. It is not enough to be able to set regulations and then get people to swear they will comply.

Validation, as it’s been done traditionally, focuses on too narrow a target to be the ideal way to confirm how well a CAV can see what is in front of it, let alone what it will choose to do with that information, according to Marques McCammon, vice president and general manager of connected vehicle solutions at Wind River Systems.

Validation is the only way to really know that something works, but it’s better to expand beyond one component or system and look at the way they interact – such as the interaction among
LiDAR, radar cameras and the central processor that uses that data to decide whether it is looking at a pedestrian walking a bike or a bike parked by the side of the road.

“It’s all too structured and process-oriented, but you have to build your code, take your process and code to a third party, and have that third party certify that it is compliant with the right ISO standard,” McCammon said. “There are different schools of thought about how to validate artificial intelligence, but a bottom-up approach seems more accurate. You can’t use AI to predict whether your AI systems are functionally safe. That’s tantamount to asking a baby to plug a laptop into a light socket. There are better ways to get it done.”

Advice to NHTSA
PennDOT also recommended the NHTSA put more effort and resources into developing or sourcing a good third-party validation scheme to confirm the safety of ADAS systems and testing protocols.

And it reiterated the important role of state DoTs for their expertise and ability to provide on-the-spot enforcement. “PennDOT applauds the Department’s commitment to achieve regulatory and operational consistency, but cautions against prematurely fixing a regulatory regime or operational standards while those technologies are still emerging and both their capabilities and complexities remain incompletely understood.”

State DoTs are valuable as partners of federal agencies, both for their local and subject-matter expertise and their local enforcement authority, according to the PennDOT response.

The document also pointed out the value of connected vehicles that operate at low levels of autonomy for slow-speed shuttles or safety vehicles, which describe PennDOT rules for registration and disclosure, or which set out long-term goals for CAV programs, a description of the planning process, roles, relationships, and details of CAV’s job and support activities in Pennsylvania.

PennDot published a pair of guidelines last summer — Automated Vehicle Testing Guidance, and its Joint Statewide Connected and Automated Vehicles Strategic Plan.

Still, neither pretends to be able to tell if an autonomous car can drive safely. Both are packed with detail on CAV testing, collaboration, managing both a test project and the people involved with it. Both also use Cohen’s conservative approach to risk, which depends on the timely use to safety – like using two safety drivers rather than one on any vehicle that travels faster than 25mph.

They also try to lay out the applications of CAV systems for things other than self-driven taxis and electric sports cars. Trolleys that travel at less than 25 mph are easier to build, validate, manage and survive than one designed to go 50 mph, for example. The key is matching speed and capability to the task.

Platooning of trucks is low hanging fruit at Level 1. This includes self-driving workzone vehicles with truck-mounted attenuators that block intrusion by cars into a work zone in order to protect road workers while they cut grass or brush or repair a roadway, Cohen said.

“That is a big deal to us,” Cohen said. “We lose a lot of people—more people than the state police. Now we can put people in those trucks with an attenuator following and guarding a paint truck, or whatever. Those trucks get hit once a month, and we did send a couple of people to the hospital, but we haven’t lost anyone yet. That feels like we benefit directly from the technology.”