WASHINGTON — As autonomous vehicle testing advances without cohesive federal guidelines, companies are operating under inconsistent or nonexistent rules for how and whether a vehicle is monitored and controlled on the road.
There are no federal rules requiring companies testing self-driving vehicles on public roads to have the ability to control them remotely. And of the five states where companies most frequently test autonomous vehicles — California, Michigan, Arizona, Pennsylvania and Florida — Pennsylvania is the only one to require that a human have the ability to take over controls if need be, mandating that a driver be in the car while it’s tested.
Safety advocates argue the complex and varied rules for remote monitoring, guidance and control are emblematic of a larger problem: Autonomous vehicles in the U.S. remain largely unregulated, making it difficult to track safety progress as the technology advances.
Companies developing autonomous vehicles say there is a need for federal regulation — for example, a built-in system for mass testing of AVs, rather than seeking a limited waiver that allows companies to deploy just a small number — but that requirements for remote control should not be one of them.
While federal rules for remote control technology are nonexistent, AV companies encounter different rules depending on where they test.
Michigan, Arizona and Florida require the vehicles meet a “minimal risk condition,” an industry term that doesn’t have a strict definition but usually means the vehicle can safely pull itself out of traffic in the case of an emergency or malfunction, stop and turn on its hazards.
Thirteen other states — Washington, New Mexico, Utah, Colorado, Nebraska, Wisconsin, Texas, Ohio, Georgia, Virginia, Maryland, Massachusetts and New York — have also reported that they have tested autonomous vehicles at some point. None of these states require remote control capabilities either, though Washington and Ohio require remote guidance abilities; New Mexico, Maryland, New York and Massachusetts require a human operator behind the wheel; and Georgia and Utah require vehicles to meet a minimal risk condition.
California has a similar rule for vehicles being tested without a driver. But it’s the only major AV testing state that monitors how companies are testing, requiring that they send detailed information to the state describing any accidents that happen on public roads.
“The reality is that no federal legislative or regulatory structure has been built around the testing or deployment of self-driving vehicles on public roads,” said Jason Levine, executive director of the Center for Auto Safety.
“As a result, oversight has been left to the states and as one might as expect, it is a patchwork, which at best asks manufacturers to promise their vehicle will minimize risk when autonomous technology encounters a situation its programming is not prepared to handle, and at worst remains entirely silent. There is little rhyme or reason to explain which states have allowed this technology on their roads without drivers either in the vehicle or operating remotely.”
Autonomous vehicle developers who spoke with The Detroit News said requiring remote control technology would be expensive, ineffective and potentially delay innovation.
Most companies already monitor and have the ability to communicate with their vehicles on the road, experts say, and total remote operation isn’t feasible because the cellular connection isn’t always fast enough or totally reliable.
“Direct control can bring more harm than value,” said Artem Fokin, head of business development at Yandex SDG, a Russian AV company testing in Michigan, Israel and Russia. The tech inside the vehicle should be able to safely control it, he said.
“That’s because no remote connection is perfect. There are no absolutely 100 percent reliable remote connections out there in the world. And if you are doing some actual steering or driving during this situation when the connection breaks, you put the entire vehicle in a potentially critical, dangerous, harmful situation.”
Instead, Yandex and most other AV companies use a type of human assistance in which the vehicle indicates that there’s a problem and awaits instruction. Argo AI, like Yandex, says it uses a remote guidance approach and not remote operation.
Kyle Vogt, president of Cruise LLC, an autonomous vehicle company wholly owned by General Motors Co., recently explained during MIT’s EmTech Digital Conference how Cruise’s technology works in an unusual situation like “phoning a friend.”
“The goal is not for AVs to come up with the correct answer for what to do next 100 percent of the time,” he said. “The reality is there are certain scenarios where the AV can navigate on its own but it might take a while, and especially in early phases of this technology while people are getting used to the idea of being in a car without a driver there’s going to be some awkward social interactions on the road.
“We want to bridge that gap for people by having the ability for a remote operator in the rare cases where a vehicle is stuck or in an unusual situation to direct the video game. This doesn’t mean the remote operator is driving the car like in a video game; they’re providing bread crumbs for decisions on whether to reroute and how to get around a blocker.”
This guidance allows a person in a control center to look at the situation the vehicle faces and give it “hints” on what to do, said Sam Abuelsamid, principal research analyst for Guidehouse Insights. That’s different from remote control when someone is literally driving the vehicle from afar.
Remote control would be difficult to scale, requiring human operators to manage multiple vehicles at once, Abuelsamid said. For example, if an AV doesn’t know how to navigate a construction zone, the operator would have to drive each vehicle around it. “Whereas, with a guidance system, you could broadcast that information to the fleet,” he said.
Fokin also argued that requiring AV companies to be able to control their vehicles remotely would only add cost to up-and-coming companies without providing a safety benefit. Human error causes most crashes, so retaining human control would retain that unpredictability, he said.
A requirement for remote control capability “would stop technology from developing and moving forward,” he said. “It doesn’t make sense from safety, and it doesn’t make sense from a business perspective.”
It’s been years since major players in the AV space began investing billions in self-driving tech. But federal safety guidelines still require most vehicles to be built to accommodate human drivers, featuring steering wheels and pedals.
Industry and safety advocates alike are pushing for Congress or the National Highway Traffic Safety Administration to develop new rules for AVs that would allow them to be deployed and provide consistent nationwide safety guidelines that provide stability for manufacturers.
The National Transportation Safety Board has criticized its sister agency for taking “a nonregulatory approach” to AV safety. Late last year, NHTSA said it would explore changing that, issuing a notice of proposed rulemaking and asking for public input on potential rules.
In the meantime, NHTSA has voluntary guidance that recommends automakers achieve a minimal risk condition before putting a vehicle on the road, but there are no binding federal rules specifically for autonomous vehicles in the United States.
NHTSA has an online tool showing where some companies are testing AVs, but participation is voluntary. The Center for Auto Safety has likened the agency’s approach to collecting safety data from companies as choosing “to accept whatever table scraps of data (the) industry chooses to provide.”
AV companies are required to “ensure that their vehicles do not pose a risk to motor vehicle safety,” and make sure they adhere to applicable safety standards, NHTSA spokeswoman Lucia Sanchez told The News via email. “NHTSA’s top priority is safety. As new technology emerges and more data is available, the agency will consider updates to its guidance for innovators and states.”
Some companies believe NHTSA’s approach to Nuro Inc., a robotics startup that produces low-speed goods delivery vehicles, indicates the agency does plan to require remote monitoring in any federal AV safety standards. Last February, NHTSA approved the company to operate up to 5,000 vehicles over the next two years.
In its petition to regulators, Nuro wrote the vehicles would be constantly monitored and that trained employees would be “able to take over driving control” if needed. In its approval, NHTSA wrote that “Nuro’s suggestion to use the remote operator as a stand-in for the driver, for purposes of compliance certification, is reasonable.”
Abuelsamid says the expectation is there will be some degree of AV remote technology called for in federal safety standards for vehicles: “That would make sense, and I think that would be the preferred thing so that it’s consistent everywhere.”
When that time comes, the industry will be prepared, he added: “It definitely needs to be done eventually, and pretty much every AV company is working with the expectation that that is part of what they’re going to need to incorporate in their system at some point.”
Story by Riley Beggin and Kalea Hall, The Detroit News.