Waymo’s blog: ‘Advice, not control: the role of Remote Assistance in Waymo’s operations’ (Image: Waymo)
Lately, Waymo is in a deep defensive crouch, insisting that its remote assistance (RA) agents do not, in any way “directly control, steer, or drive the vehicle.” They only respond to requests by the Computer Driver (or Autonomous Driving System), if needed. Question is: Is this a difference or just a distinction?
The issue poses a huge dilemma of public perception. The company can’t afford investors, or even the public at large, thinking its robotaxis are not fully autonomous.
But the cats have departed the bag. Missy Cummings, George Mason Univ. professor, calls Waymo’s RAs “babysitters.” Even humorist Dave Barry has weighed in, writing about his first ride in a Waymo robotaxi, which he jokingly described as guided by “a 14-year-old boy somewhere in Asia — he goes by ‘Kevin’ — who steers your car remotely with a joystick.”
Assistants vs drivers
Certainly, remote operators — whether remote “assistants” or “drivers” — are a necessary backstop, as a matter of safety. But why is Waymo so hung up on terminology? Who cares if these human remote agents are called assistants, drivers, or late to dinner?
Nobody at Waymo is telling us, but it’s likely the company is hammering semantics with a weather eye on future litigation.
Further, this could also help them evade state control over AVs.
Remember, in the United States, driver’s licenses are issued by each individual state. State and municipal governments want to address the potential impacts of these vehicles on the road. Most states already have or are considering legislation related to AVs.
But if Waymo succeeds in its argument that their remote operators are “not drivers,” states could lose the ability to hold AV companies accountable for crashes involving botched remote operator interventions. That could ultimately combine with federal pre-emption of state-by-state AV regulations to result in no state oversight for AV safety, which is indeed a dream come true for many AV companies who prefer federal regulations.
But wait. Before venturing into the legal weeds, one question.
When a robotaxi crashes, bearing in mind that the driver’s seat was vacant, who do you think was driving?
Was the “driver” Waymo’s computerized autonomous driving system? Could a passenger be somehow involved? Perhaps a remote operator might have given misleading advice to the computer and steered the car into trouble?
An added complication is that the legal interpretation of who and what the driver is depends on the state where it happened.
But if the fault falls on a remote operator, the question of who was driving does not depend on whether Waymo calls its call-center minion an “assistant” or a “remote driver.”
The law doesn’t care.
Moreover, it might matter little whether Waymo’s remote assistants had a steering wheel, accelerator or brake pedals at their remote workstation.
Regardless of the remote assistants’ locale, said Phil Koopman, professor emeritus at Carnegie Mellon University, “If you are even only partially involved in the dynamic driving task (DDT), that’s sufficient to make you a driver in a practical sense.”
Smart Summons
Take Tesla’s Smart Summon feature, suggested Koopman.
Smart Summon, as defined by Tesla, is “designed to allow you to move Model Y to your location (using your phone’s GPS as a target destination) or to a location of your choice, maneuvering around and stopping for objects as necessary. Smart Summon works with the Tesla mobile app when your phone is located within approximately 65 meters of Model Y.”
Note the steering wheel is not mentioned. “Your app” gives you control of the vehicle with a simple “go/stop” control, noted Koopman.
Tesla has made clear that a driver using Smart Summon is responsible for his or her car. In case of an accident, normal liability rules apply in both fault and no-fault states.
This is an example of why— without use of a steering wheel, joystick or brake pedal—an “assistant” who is not even in the car could become a tele-driver.
Big red button
So, let’s agree that to even partially perform DDT, you don’t necessarily need a physical steering wheel in front of you. The control device could be an app, or a big red button on the autonomous dashboard.
In Tesla’s, robotaxis have access to two emergency buttons, according to Reddit.
One button, on the front main screen, reads “Stop In Lane.” This stops the car immediately in its lane if the car appears poised to do something wrong.
Button No. 2 a “stop now” Emergency Hard Brake, that appears to be on the door, reprogrammed to act as a E-Brake, aka kill switch.
Youtube videos have shown these stop buttons in use by passenger-seated safety monitors. It is not unreasonable to think that such a “stop now” button will be made available to retail customer passengers.
But matters get tricky.
In fact, some state may find a passenger liable for not hitting a robotaxi’s “stop now” button. Think of a “panic button”, a vehicle button that lets passengers end a ride early, or even just an “end ride” button on a passenger’s mobile phone app.
William H. Widen, a professor at the University of Miami’s School of Law, sees that the action — hitting an in-vehicle stop button — allows riders to have a certain level of control over the vehicle. He argues in his academic paper that this could make a level 4 passenger the driver, thus exposing them to DUI convictions even if they never touch the button.
Put it more plainly, “There’s a lot of uncertainty around the concept of a driver and who’s driving, and there’s a surprisingly low threshold for the amount of human control that is necessary to trigger this status of, if you will, of being a driver,” Widen explained to a reporter at Gazette,
Left in a gray zone
There is a possible legal motivation to Waymo’s insistence that its computer driver (or ADS) is responsible for all real-time driving tasks and decision-making.
Susanna Gallun, an attorney and transportation policy researcher at the University of Texas at Austin’s Center for Transportation Research (CTR), said to me, “AV statutes that define the ADS as the ‘driver’ complicate the situation and leave remote assistance in a gray zone.”
Emerging guidelines fomented by an industry group such as the Automated Vehicle Safety Consortium (AVSC) focus on remote assistance programs as a safety function. In her opinion, however, “They do not yet resolve, in a doctrinal way, who counts as the legal driver when remote assistance information changes the vehicle’s behavior.”
Duty of care?
Koopman has always maintained that “getting rid of the human driver 100% is not going to happen this decade, and probably not the following decade.” Some sort of help from humans, he concludes, will be necessary for a long time,
In that sense, it doesn’t matter what Waymo calls its remote operators. What matters, said Koopman, is the safety outcome.
“That human remote operator makes a contribution to driving safety and the risk of a potential loss event,” he added, regardless that the remote person is doing all the driving or simply contributing a small piece of the driving task.
Given that Waymo has already suffered a mishap caused by a human remote operator, it should be obliged to provide its hazard analysis and articulate the measures it has taken to mitigate similar risks in the future, Koopman pointed out.
The mishap Koopman referred to is an incident in San Francisco when a remote assistant told a Waymo robotaxi that a red traffic light was green.
Waymo might say that it examined this incident and decided it was an incident so minor that nobody should worry. If that’s how Waymo wants to build its safety case, that's their choice, Koopman said.
But what if there are numerous other incidents like this that did not happen to result in a reportable incident? Should Waymo have a responsibility to do better than simply deflect safety concerns over remote operators?
Product liability
What worries Koopman is that Waymo is claiming its ADS solely is responsible for all real-time driving tasks and decision-making. An ADS is not a human. It’s a computer.
AV operators like Waymo might prefer to keep “human” remote operators out of the discussion, because an ADS is not recognized as a legal “driver” with affirmative statutory duties, said Gallun. Negligence doctrine is built around human conduct. Tort law evaluates breaches using a “reasonable” person/driver standard. Since ADS is not human, it’s hard to argue the meaning of “reasonable automated driving” for holding ADS responsible for being negligent.
Moreover, an ADS is not a legal person. So it cannot be sued for negligent behavior. And even if it could, it has no money to pay restitution, and the ADS does not care if it goes to jail.
So, when an AV crashes, the crashed-into plaintiff has no choice but to argue “product liability.”
This is great news for AV companies. Corporations are legally comfortable “in the product liability lane, and they do not want to be in the negligence lane,” Gallun noted.
For plaintiffs, as Koopman has always said, “Product liability is the wrong tool for most automated vehicle crashes.” Plaintiffs are forced to shell out large sums to hire lawyers and expert witnesses and must often invest years of their time to prove that software code caused a crash, for example. Worse, what if it turns out the product is designed perfectly and the sole fault for a crash was a mistake by a remote operator, who was not legally the driver? Could that result in a fatal crash falling into legal limbo, with no recourse possible for victims?
Keep in mind that if a remote operator makes a mistake, their employer could be on the hook for negligence. Perhaps this helps explain why Waymo is so resolute about the semantics of the word, “driver.”
But not everything is in plain sight.
AV companies are already quietly stacking the deck gainst victims of self-driving car crash who take them to court. They are taking every step to make legal and regulatory systems work for them, which means keeping everything in the product liability lane, and closing off all avenues for tort negligence liability.
The AV industry is counting on consumers to remain uninformed and unaware of the game already being rigged. After all, why would ordinary consumers —who have been already led to believe that robotaxis are safer than human-driven cars — worry about serious crashes that haven’t even happened yet?
I get most people don’t want to hear this. But put bluntly, with “product liability” as their only recourse, consumers hit by AVs face greatly diminished chances to hold carmakers and robotaxi services accountable. This includes not only riders, but also vulnerable road users such as pedestrians who might be hit by one of these vehicles.
It’s time to pay attention to our own legal consequences in the future.
I'm struggling to understand how the way that Waymo characterizes its remote facilitators could make a difference for tort liability. Waymo is vicariously liable for the tortious acts of its agents within the scope of their agency relationship. That's black letter law. And whether Waymo calls a remote facilitator an "remote assistant" or a "driver" makes no difference to the agency analysis.
It *might* matter whether Waymo has an employer-employee relationship or an independent contractor relationship with the remote facilitator. But even if Waymo's remote facilitators are independent contractors, the Restatement suggests that Waymo might be vicariously liable because Waymo knows or should know that the activity of remote facilitation poses a "peculiar risk."
Here is section 59 of the Restatement (Third) of Torts:
"An actor who hires an independent contractor for an activity that the actor knows or should know poses a peculiar risk is subject to vicarious liability for physical harm when the independent contractor is negligent as to the peculiar risk and the negligence is a factual cause of any such harm within the scope of liability."
And here's what the comments say on the definition of "peculiar risk":
"Under this Restatement, an activity poses a peculiar risk when, if reasonable care is not taken, the resulting risk differs from the types of risk that are usual in the community. ... For instance, the activity of driving an automobile on the highway often leads to severe and even catastrophic injuries caused when reasonable care is not taken—when, for instance, a driver becomes distracted, falls asleep, or has inadequate or slow physical reflexes. These are not peculiar risks—even though they are often severe and preventable—because they do not differ in kind from types of risk usual in the community. On the other hand, transporting large fuel tanks by truck can pose a peculiar risk. When reasonable precautions are not taken, the risks—falling tanks, fuel spills, and explosion—are not usual in the community."
I think remote facilitation of an automated driving system--unlike conventional driving--at least arguably "differs from the types of risk that are usual in the community."
SAE J3016 states clearly in the Remote Assistance definition that providing information to the ADS is not being part of the DDT. Here is the full definition:
3.23 REMOTE ASSISTANCE
Event-driven provision, by a remotely located human (see 3.31.5), of information or advice to an ADS-equipped vehicle in driverless operation in order to facilitate trip continuation when the ADS encounters a situation it cannot manage.
NOTE 1: Remote assistance does not include real-time DDT or fallback performance by a remote driver. Rather, the ADS performs the complete DDT and/or fallback, even when assisted by a remotely located human.
NOTE 2: Remote assistance may include providing an ADS with revised goals and/or tasks.
NOTE 3: The remote assistance function does not include providing strategic instruction regarding selection of destinations or trip initiation timing (i.e., dispatch functions), even if the same person performs both remote assistance and dispatching functions.
This article was written without taking this clear definition into consideration. The stop button on the Tesla Summon app or the big red button for the attendants in the Tesla robotaxi service in Austin, for example, are part of the DDT. Provision of information or advice is not.