Robotaxi Teleoperators: We Know They’re There. What Do They Know?

It's time to discuss the substance of teleoperators. What are their responsibilities? What are their qualifications? What do they know? Are they "the drivers"? What's the robotaxi/teleoperator ratio?

Mauricio Peña, the Chief Safety Officer at Waymo, testifies before a Senate committee (Image: PBS)

At a Senate hearing Wednesday, Mauricio Peña, Waymo’s chief safety officer, acknowledged that his company’s robotaxi operations rely on “remote operators” located in the Philippines, when a Waymo vehicle “asks for guidance in certain situations.”

Peña’s admission that Waymo uses “teleoperators” based outside the United States, is something of a breakthrough. The stone wall has fallen.

To be clear, though, the real news isn’t that self-driving cars are NOT human-less. As Phil Koopman, professor emeritus at Carnegie Mellon University, explained a year ago on Substack, “All Robotaxis Have Remote Drivers.”

All Robotaxis Have Remote Drivers

The familiar hype has been about robotaxis completely getting rid of human drivers. This purge won’t happen this decade, and probably not the next one either. The actual challenges will focus on figuring out human drivers’ reaction time in a crisis and ensuring safety when that happens. Reduce human involvement enough and you might be profitable. Reduce it too much or in the wrong way, and one high-profile crash can put you out of business.

Now that everyone knows teleoperators are involved in robotaxi services, we have some questions. What are the responsibilities of those teleoperators? What are their qualifications? What do they know? What exactly do they do? How many are allocated for how many robotaxis? How many are located outside the United States?

The answers are critical to the eventual safety outcome of robotaxis.

After confirming that Waymo has a “human being that helps the vehicle to navigate difficult driving scenarios,” Senator Ed Markey (D-Mass.) told Peña, “For me, that’s fairly shocking. Waymo has critical safety employees — who may need to intervene in a split second if a Waymo encounters an unknown, dangerous situation located in the United States — but they are outside the United States.”

WATCH LIVE: Tesla, Waymo executives join safety experts in testifying on future of self-driving cars (Source: PBS)

The location of remote operators matters, because it poses concerns about communication channel latency and dropouts. This issue applies to a robotaxi both for remote assistance and remote driving.

Qualifying remote operators is also important. As Koopman asked, “Would you ask a five-year old if the traffic light at the intersection is red or green?” What if the teleoperator isn’t sober? Does he or she have a driver’s license? Where, in the Philippines or in the U.S.? Bottom line: “Can we trust this faraway stranger?”

During the hearing, when pressed by Sen. Markey if a majority of teleoperators are located abroad, Waymo’s Peña demurred, saying, “I just don’t have that number.”

Sen. Markey politely described Peña’s response as “curious.” I call it disingenuous.

I couldn’t help wondering why investors let Waymo get away with not answering nuts-and-bolts questions about the implementation and expansion of its teleoperation, at a time when AV advocates already acknowledge that the key issues of automated driving have “shifted from technology challenges to scaling and business model challenges.”

But I digress.

Follow-up questions

I asked safety experts Koopman and George Mason University professor Missy Cummings the follow-up questions they would have asked Waymo, after Sen. Markey’s line of questioning.

Cummings focused on transparency.

She said, “I first would have asked that if they care so much about safety and transparency, why have they not:

“1. Released the video of Waymo striking the child. (Editor’s note: She is referring to a recent incident where Waymo car hit a child near Santa Monica elementary school.)

“2. Made a statement in their safety case about maximum latency of remote operations

“3. Made a sworn statement about who many times a remote operator was involved in a crash

“4. Provided data on hard braking events

“5. Explained in detail why their operations on highways are extremely limited and provided a date for operations everywhere

“6. Provided the rate of immobilizations (per VMT).”

Who’s the driver?

Koopman explained that the most important question Waymo chose not to answer was about who’s driving.

“Who is ‘the driver,’ when a teleoperator gives a robotaxi advice that affects the vehicle’s motion?” asked Koopman.

He mentioned an incident in San Francisco when a remote assistant told a Waymo robotaxi that a red traffic light was green. The robotaxi entered the intersection, and this resulted in a mishap involving a crossing vehicle that had the right of way with a green light.

No one was seriously hurt, but if there had been injuries, who do you sue?

In his testimony, Peña tried to distinguish assisting from driving. He stressed, “Waymo asks for guidance in certain situations, and it’s an input, but the Waymo vehicle is always in charge of the dynamic driving task. So that is just one additional input.”

Okay. But when the teleoperator helps with additional information that turns out to affect the dynamic driving task, who’s driving?

Koopman said, “Usually whoever makes the decision to enter an intersection against a red light would be considered the driver and be blamed accordingly.”

If the remote assistant was not driving, Koopman asks, what exactly was he doing? Peña could have clarified that teleoperators in the Philippines are not drivers.

But he didn’t say that.

This matters because by leaving the question of who was actually driving a vehicle unanswered, Waymo creates an accountability void, leaving nobody responsible for future robotaxi mishaps.

Further, I worry about a cascading effect potentially triggered by teleoperators. Teleoperation could pose the danger of introducing more human error into the process of removing human error. As robotaxi companies lean on third-party teleoperator outfits to manage unanticipated safety problems, they might push liability issues downstream. When a teleoperator in the Philippines makes a grave error resulting in a crash, the victim might be compelled to hunt down a culpable teleoperator who’s not in the United States. But by that time, the teleoperator in question might have been long gone.

Remote driving vs. remote assistance

In a paper entitled “What self-driving companies should learn from drone remote operations” currently under review, Cummings argues that self-driving vehicles— despite improved performance—still struggle in uncertain and dynamic situations. This makes necessary self-driving systems’ reliance on human oversight through remote operations.

The point is not whether teleoperations should be deployed.

“Remote operations are a safety‑critical capability whose design will strongly influence the near‑term viability of self-driving vehicle deployments,” she argues.

Instead of leaving teleoperation’s role ambiguous, AV promoters should be discussing “latency constraints, the importance of human‑centered workstation design, the challenges of managing operator workload, the need for systematic operator training, and the necessity of robust contingency planning.”

Referring to her paper, Cummings maintains, “It is important to understand that remote driving and remote assistance are two different things.” However, she added, “They both directly support safety. They are both affected by latency, although remote assistance less so (but it still can be a big problem.”

Bottom line

Teleoperators deployed by robotaxi companies should come as no shock. Instead of getting caught up in a race to reduce the number of teleoperators deployed in robotaxi services, the companies who want to run these businesses must define teleoperators’ responsibilities while coping with the technical challenges involved in designing safe teleoperation centers.