Driverless Cars: Not as Autonomous As We Thought

  

As part of an investigation by U. S. Sen. Edward Markey (D-Mass.), autonomous-vehicle operators Tesla, Waymo, and others have made public some details about how their supposedly driverless cars can be remotely assisted or controlled under certain circumstances.  In a report published on Apr. 8, Andrea Guzman of the Austin American-Statesman described how the companies reluctantly shared some but not all the information Sen. Markey requested.  Specifically, they refused to say how often such interventions take place.

 

It was a surprise to me, though it probably shouldn't have been, that autonomous-vehicle firms have staffs of remote operators ready to take control of a car or truck that gets into trouble.  It's a given that every such vehicle is in continuous wireless communication with "headquarters," whatever that amounts to.  And when the onboard systems determine that the situation the vehicle is facing is beyond its capabilities, it's reasonable that it sends an alert to headquarters to flag a human being to take a look and decide what ought to be done.  But in at least a few situations, the human-assisted decisions haven't been all they should be. 

 

For example, an investigation by a safety agency found that a remotely-operated Waymo vehicle illegally passed a school bus where children were boarding.  Apparently, the Waymo remote assistance operators do not take direct control of the vehicle, but merely guide it in making its own decisions.

 

Tesla's process of remote assistance seems to be more assertive.  In responding to Sen. Markey's inquiry, Tesla revealed that in rare cases, their remote assistance operators can take direct control of the vehicle, but its speed is limited to either 2 or 10 MPH, depending on whether the internal automated system grants direct control to the human controller.

 

That makes it sound like the final decision is up to the machine, not the human.  Whatever the details, the fact that human remote assistants are sometimes involved in driverless car control adds another layer to the already complex mixture of responsibilities involved. 

 

Both Tesla and Waymo emphasized that the remote assistants do not sit there watching each car passively.  For one thing, there are too many cars to monitor.  Waymo revealed that by spreading the work among staff in the U. S. and the Philippines, they can maintain about 70 remote assistance operators on duty at any given time.  Tesla's operators are all in the U. S., and presumably work day and night shifts.  They all have valid U. S. driver licenses for at least three years prior to employment, for what that is worth. 

 

The information that all companies queried refused to reveal was how often it happens that remote-assistance operators intervene in a vehicle's operation.  Perhaps justifiably, they claim that is a trade secret that might compromise their competitive position.  My guess is that they have a pretty good idea of what the competition is doing, but they don't want the competition to know that.  Until it becomes an issue in a court case, then, we'll just have to guess at whether the driverless vehicle you see cruising down the street is truly on its own, or is being supervised from Palo Alto or Manila.

 

The mixing of autonomous and human control is going to become more urgent in the future as more and more systems become capable of running themselves.  Less than two weeks ago, on Apr. 14, Ukrainian President Volodymyr Zelenskyy announced that for the first time, his soldiers captured an enemy position using only remotely-controlled equipment, namely unmanned drones and "ground robotic systems," meaning armed robots. 

 

This historic achievement speaks volumes for the progress that Ukraine has made in developing agile, cost-effective military technology that augments their strength in human troop numbers.  After four years of war against a numerically superior foe, the Ukrainians have become world leaders in such technology, motivated by one of the strongest reasons around:  survival.  It is a truism in the history of technology that wars advance technical developments faster than peace does, and the Ukrainians have shown its validity once again.

 

For years, much ink has been spilled in engineering ethics journals about the dangers of autonomous warfare.  While this may not seem to have much to do with remote assistance for Waymo and Tesla, the two cases are extremes on a continuum.  In the case of peacetime vehicles, you want the machinery to do what it's supposed to do benignly without hurting anybody, but in cases where there is potential danger, you want the human backup person to intervene.  In the case of autonomous warfare, you want to damage and destroy the enemy and its resources, but you don't want the autonomous systems going haywire and attacking your own troops. 

 

Getting back to remotely-assisted civilian vehicles, it appears that the operating companies have found a comfortable medium between the extremes of completely autonomous operation—no humans involved anywhere along the line—and what would amount to a remote-control cabdriver, namely one remote assistant dedicated to each vehicle full-time.  Their comfort zone appears to be a mix of mostly automated control, but some manual control in emergencies. 

 

While it would be nice to know more details, in a free market it's up to the firms themselves to decide the optimum mix, and they seem to want to keep the details to themselves.  Until there's a major lawsuit in which the remote-assistant issue becomes a critical factor, we may not learn a lot more about it. 

 

The question of responsibility is fundamental in engineering ethics.  With the rise of all sorts of autonomous systems that are not quite autonomous, the question gets complicated.  The best we can do is to learn as much as we can about how things are actually done, so when problems arise the public and regulatory agencies are not kept in the dark about where responsibility may really lie. 

 

And whether it's a Waymo car or an armed robot capturing an enemy soldier, it's a little comforting to know that humans are still at the top of the command structure, however remotely.  Unless I'm the soldier—then it may not matter.

 

Sources:  Andrea Guzman's article "Self-driving cars still rely on humans—but how often remains unclear" appeared in the Austin American-Statesman on Apr. 8, 2026 at https://www.statesman.com/business/article/tesla-waymo-remote-operators-self-driving-cars-22192151.php.  I also referred to an article describing the Ukrainian remote-controlled capture of a Russian position at https://www.politico.eu/article/volodymyr-zelenskyy-robotic-systems-russia-army-positions-ukraine/. 

Waymo Comes to Austin

 

In March, the robotaxi company Waymo announced that it was teaming for the first time with the ride-hailing service Uber to provide rides in Austin, Texas.  Waymo traces its roots to a secret project started by Google in 2009, with some key staffers having participated in the 2004 Stanford Self-Driving Car Project.  It is now a subsidiary of Alphabet Inc., the parent company of Google.  Uber, along with Lyft, provides app-based ride-hailing services outside the usual taxi structure.  Waymo's driverless taxis operated commercially first in San Francisco, and now provide service in U. S. cities such as Phoenix, Los Angeles, Miami, and now Austin.

 

Last week, I rode with a friend, who was driving, from north Austin to the University of Texas main campus just north of downtown.  Our route went along Burnet Road, which is in many ways the prototypical Austin street:  four lanes, a center median in some blocks, all kinds of one-story retail businesses along the way with multiple parking-lot entrances in each block, and fairly heavy traffic. 

 

We pulled up at a red traffic signal, and a few feet ahead of us in the lane to our right were two Waymo driverless cars, one in front of the other.  There were no passengers and of course, no driver.  They are easy to spot from a distance, even if you can't see whether anyone is in the driver's seat.  A thing like a black police gumball on top of the roof spins constantly, as do smaller gizmos on each side of the car, and it bristles with wide-angle camera lenses and less identifiable technology, as well as logos letting you know what it is.  The Wikipedia article on Waymo says the vehicles are equipped with 360-degree lidar (light-based radar) with a nearly 1000-foot (300-meter) range, radio-type radars, and an extremely sophisticated AI processing system.  Also, some stray comments on a Reddit page indicates there are human "remote monitors" who are responsible for several cars each.  So yes, they look completely autonomous, but somewhere in the background there's a human ready to intervene if something unusual happens. 

 

I didn't know all of that as I gazed at a technology that has been on the streets in some form for at least a decade.  But I have never actually seen a driverless car that close, let alone two of them. 

 

The car in front was facing the intersection when the red light changed to green, and sure enough, it knew to start going, just like somebody was driving it.  This is not a true-confession column, but I have to admit something here.  Seeing those driverless cars acting so normal gave me the perverse urge to try to mess them up somehow, to wave a hand in front of one of the cars or something just to see how clever it was.  It was only a brief stray thought, but its strangeness struck me.  I am normally a live-and-let-live type of person, not given to vandalizing-type ideas.  But there is a radical difference between reading about things and seeing them for yourself.

 

Philosophers make a distinction between two types of knowing.  Knowing by description is what you would learn about riding a bicycle, say, by reading a book on how to ride a bicycle.  You might understand the physics of bicycles, you could answer detailed questions on bicycle dynamics, but if you had never actually sat on a bicycle or tried to ride one, your knowing about bicycles is only the first kind of knowledge: knowledge by description.

 

On the other hand, knowing by acquaintance requires a personal physical encounter with the thing known.  Up to last Wednesday, I had known a lot about autonomous vehicles.  I have blogged about them probably dozens of times.  But all my knowledge was by description, not acquaintance. 

 

Seeing those two Waymo cars in the flesh, so to speak, was qualitatively different than any amount of reading or watching YouTube videos.  I was right there, not twenty feet away from them, and if something happened to go wacky with one of them, I could be personally in danger. 

 

Not that Waymo has had too much trouble along those lines.  Wary of how one spate of bad publicity can ruin an entire project, Waymo has been very cautious in choosing its operating locations and in keeping a nearly spotless safety record so far.  According to Wikipedia, the only fatal accident involving a Waymo driverless car happened when an unoccupied Waymo vehicle was involved in a multiple-car pileup set off by a Tesla driver who was going 98 MPH before the crash.  Waymo can hardly be blamed for that.

 

Austin is a good choice for Waymo's initial teaming with Uber, as it is full of technophiles who will take a driverless Uber ride just for the thrill of it.  The annual international futurefest called SXSW (originally South by Southwest) was held in Austin in March, and I'm sure many of the out-of-town participants were tickled to get Uber rides in Waymos, which was probably one of the main reasons they were rolled out that month. 

 

There are still some people like me who will have generally negative feelings towards robotaxis.  For one thing, you're not going to have the huge variety of interpersonal experiences that riding in a human-driven cab provides. 

 

Just for example, a couple of weeks ago I flew to Lawrence, Kansas, and the closest airport was in Kansas City, a 50-mile-or-so drive away.  The place I was visiting hired a limo service to pick me up at the airport, and for the hour or so drive to Lawrence I had a fascinating conversation with a native of Veracruz, Mexico, who was as full of local and international tourist-type info as he was curious about various exotic places I'd been. 

 

Waymo isn't going to do that.  All you're greeted with is an empty car with a few control buttons, and no conversation.  At least not yet, but maybe they'll add a Cab-Chat app for people who miss the old days of human drivers.

 

Sources:  The Associated Press carried an article describing Waymo's teaming with Uber in Austin at https://apnews.com/article/uber-waymo-robotaxis-austin-texas-988aba46988e649be8cf59979587a8e5.  I found a reference to human monitors of Waymo cars at https://www.reddit.com/r/waymo/comments/1f3ur68/current_waymo_revenue_per_car/?rdt=36096.  I also referred to the Wikipedia article on Waymo and the Stanford Encyclopedia of Philosophy article https://plato.stanford.edu/entries/knowledge-acquaindescrip/. 

Cruise Gets Bruised

 

General Motors' autonomous-vehicle operation is called Cruise, and just last August, it received permission (along with Google's Waymo) to operate driverless robotaxis in San Francisco at all hours of the day and night.  This made the city the only one in the United States with two competing firms providing such services.

 

Only a week after the California Public Utilities Commission acted to allow 24-hour services, a Cruise robotaxi carrying a passenger was involved in a collision with a fire truck on an emergency call.  The oncoming fire truck had moved into the robotaxi's lane at an intersection surrounded by tall buildings and controlled by a traffic light, which had turned green for the robotaxi.  Engineers for Cruise said that the robotaxi identified the emergency vehicle's siren as soon as it rose above the ambient noise level, but couldn't track its path until it came into view, by which time it was too late for the robotaxi to avoid hitting it.  The passenger was taken to a hospital but was not seriously injured.

 

As a result, Cruise agreed with the Department of Motor Vehicles to reduce their active fleet of robotaxis by 50% until the investigation by DMV of this and other incidents was resolved.

 

In many engineering failures, warning signs of a comparatively minor nature appear before the major catastrophe, which usually attracts attention by loss of life, injuries, or significant property damage.  These minor signs are valuable indicators to those who wish to prevent the major tragedies from occurring, but they are not always heeded effectively.  The Cruise collision with a fire truck proved to be one such case.

 

On Oct. 2, a pedestrian was hit by a conventional human-piloted car on a busy San Francisco street.  This happens from time to time, but the difference in this case was that the impact sent the pedestrian toward a Cruise robotaxi.  When the unfortunate pedestrian hit the robotaxi, the vehicle's system interpreted the collision as "lateral," meaning something hit it from the side.  In the case of lateral collisions, the robotaxi is programmed to stop and then pull off the road to keep from obstructing traffic.

 

What the system didn't take into account was that the pedestrian was still stuck under one of the robotaxi's wheels, and when it pulled about six meters (20 feet) to the curb, it dragged the pedestrian with it, causing critical injuries. 

 

California regulators reacted swiftly.  The DMV revoked Cruise's license, and the firm announced it was going to suspend driverless operations nationwide, including a few vehicles in Austin, Texas, and other locations.  There were only 950 vehicles in the entire U. S. fleet, so the operation is clearly in its early stages. 

 

But now, after Cruise has done software recalls for human-piloted vehicles as well as driverless ones, it is far from clear what their path is back to viability.  According to an AP report, GM had big hopes for substantial revenue from Cruise operations, expecting on the order of $1 billion in 2025 after making only about a tenth of that in 2022.  And profitability, which would require recouping the billions GM already invested in the technology, is even farther in the future than it was before the problems in San Francisco.

 

The consequences of these events can be summarized under the headings of good news and bad news.

 

The good news:  Nobody got killed, although being dragged twenty feet under the wheel of a robot car that clearly has no idea what is going on might be a fate worse than death to some people.  After failing to heed the warning incident in August, Cruise has finally decided to react vigorously with significant and costly moves.  According to AP, it is adding a chief safety officer and asking a third-party engineering firm to find the technical cause of the Oct. 2 crash.  So after clearly inadequate responses to the earlier incidents, a major one has motivated Cruise management to act. 

 

The bad news:  Several people were injured, at least one critically, before Cruise realized that, at least in the complex environs of San Francisco, their robotaxis posed an unacceptable risk to pedestrians.  I'm sure Google's Waymo vehicles have a less-than-perfect safety record, but whatever start-up glitches they suffered are well in the past.  Cruise does not have the luxury of experience that Waymo has, and is in a sense operating in foreign territory.  Maybe Detroit would have been a better choice than San Francisco for a test market, but that would have neglected the cool factor, which is after all what is driving the robotaxi project in the first place.

 

Back when every elevator had a human operator, there was a valid economic and engineering argument to replace the manually-controlled units with automatic ones.  The main reason was to eliminate the salary of the operator.  Fortunately, the environment of an elevator is exceedingly well defined, and the relay-based technology of the 1920s sufficed to produce automatic elevators which met all safety requirements and were easy enough for the average passenger to operate.  Nevertheless, some places clung to manual elevators as recently as the 1980s, as I recall from a visit to a tax consultant in Northampton, Massachusetts, whose office was accessed by means of an elevator controlled not by buttons but by a rather seedy-looking old man.

 

Being an old man myself now, I come to the defense of everyone on the street who would like to confront real people behind the wheel, not some anonymous software that may—may!—figure out I'm a human being and not a tall piece of plastic wrap blowing in the wind, in time to stop before it hits me.  Yes, robotaxis are cool.  Yes, they save on taxi-driver salaries, but this ignores the fact that one of the few entry-level jobs that recent immigrants to this country can get which actually pays a living wage is that of taxi driver, many of whom are independent entrepreneurs. 

 

Robotaxis may be cool, but dangerous they should not be.  GM may patch up their Cruise operation and get it going again, but then again it may go the way of the Segway.  Time will tell.

 

Sources:  An article from the USA Today Network in the online Austin American-Statesman for Nov. 16, 2023 alerted me to the fact that Cruise was ramping down its nationwide operations, including those in Austin.  I consulted AP News articles at https://apnews.com/article/cruise-general-motors-pedestrian-recall-software-crash-bf08c0c6e7914649750b4dde598af5fc

(Nov. 8, 2023) and at https://apnews.com/article/san-francisco-cruise-robotaxi-crash-e721a81c1366c71a03c0aa50aa2e98f3 (Aug. 19, 2023).