If Self-Driving Cars Aren’t Safer Than Human Drivers, They Shouldn’t Be on Public Streets

AV crash

More details are emerging about how an Uber self-driving car struck and killed a woman in Tempe, Arizona, raising red flags about the testing of autonomous vehicles on city streets.

Tempe police rushed to absolve Uber since the victim, Elaine Herzberg, was outside a crosswalk, according to the Phoenix New Times. But the paper’s Ray Stern reports that it’s common for people to cross midblock at that location — something a human driver may have anticipated.

Police also said the Uber car was exceeding the speed limit, traveling 38 mph in a 35 mph zone, according to the San Francisco Chronicle, and that neither the vehicle nor the person behind the wheel, who is supposed to take control to prevent collisions, engaged the brake “significantly” prior to impact.

Beyond the particulars of the crash, which is still the subject of an open investigation, transportation officials and safety advocates warn that the incident highlights the dangers of allowing autonomous vehicle testing on public streets without clear safety standards or guidelines.

Uber’s program has only logged a few million miles in self-driving mode — crossing the 2 million mile threshold in September while adding a million miles every 100 days, according to Forbes. Meanwhile, human drivers in the U.S. traveled about 86 million miles for every traffic fatality in 2016, reports the Insurance Institute for Highway Safety. (Other companies, including Tesla and Waymo, have compiled more mileage than Uber.)

Arizona has become a testing ground for autonomous vehicles in part because of its hands-off approach to regulation. And last month, the California DMV passed a rule that would allow autonomous vehicles on public streets with only a remote operator to intervene in case of emergency.

At the federal level, U.S. DOT has only issued voluntary guidelines for autonomous vehicle companies. Legislation pending in Congress right now would allow companies to not just test AVs but sell them to consumers. The bill, AV START, passed unanimously in the House but has stalled in the Senate.

Autonomous vehicles have the potential to be safer than human drivers, but these testing arrangements are proceeding with no agreed-upon safety standard to assess the technology.

The National Association of City Transportation Officials said in a statement that firm guidelines should be established because “the current model for real-life testing of autonomous vehicles does not ensure everyone’s safety.”

“In order to be compatible with life on city streets, AV technology must be able to safely interact with people on bikes, on foot, or exiting a parked car on the street, in or out of the crosswalk, at any time of day or night,” said NACTO Executive Director Linda Bailey. “Responsible companies should support a safety standard and call for others to meet one as well.”

Meanwhile, in the absence of such a standard, “the American public is serving as crash test dummies,” said Cathy Chase of Advocates for Highway and Auto Safety, an organization founded by the insurance industry and consumer watchdogs. “I see this as an urgent call to action.”

Chase wants to see an across-the-board pause on further loosening of autonomous vehicle regulations until the National Transportation Safety Board, which is investigating the Tempe crash, releases recommendations.

As the Tempe crash illustrates, detecting people walking or biking is a known weakness of self-driving cars. A number of active transportation advocacy organizations objected to the federal AV START bill on the grounds that the technology is not advanced enough to safely react to pedestrians, cyclists, or people in wheelchairs.

Advocates for Highway and Auto Safety has 12 recommendations to improve the legislation. One thing the public should insist on, says Chase, is a “vision test” for vehicles operating on public roadways. Just like a human driver would need to be able to spot distant objects and react to them, so should autonomous vehicles.

Relaying information from Tempe police, the SF Chronicle reported that Herzberg, the crash victim, was walking a bike “laden with plastic shopping bags,” implying that Uber was not at fault. But that’s the kind of scenario that arises on city streets all the time in real life.

“If an AV cannot properly react to that type of situation it should not be on the road,” Chase said.

247 thoughts on If Self-Driving Cars Aren’t Safer Than Human Drivers, They Shouldn’t Be on Public Streets

  1. the passengers derive all the benefit from the new technology

    As a pedestrian and cyclist who is very tired of having near-misses by drivers who aren’t looking at the road while driving, and who knows several people who have been hospitalized by careless drivers, I couldn’t disagree more.

    Killing innocent bystanders is never OK when testing new technology.

    Why is it okay to kill innocent bystanders by allowing humans to drive cars despite all the evidence we have that it doesn’t work very well? If we want to ban all cars because of the unacceptable cost in innocent lives, then great, let’s do that, and then come up with an effective minimum bar for allowing cars on the road again and subject every control system–human or robot–to that validation. I’m on board.

    But if we’re going to continue to let humans drive, saying we can never try out any other option because it might be worse, so we’re stuck with never improving, is absurd.

    But robot cars could be far worse.

    Or they could be far better. What is your suggestion for finding out without ever letting them on the streets? It’s not like this is the first step in testing AVs (which would indeed be crazy), after all.

    No one really knows how neural networks work.

    We know a lot more about them than we know about how brains work. And we know absolutely nothing about how to make human brains better at driving. But we keep letting them drive.

  2. It’s funny that everything you mentioned is tested heavily in closed-system, removed from the public environments, to ensure it is within safety parameters before the general public ever encounters them (and in the medical field, with volunteers that are extremely well informed on the risks, and elect to participate anyway). Meanwhile, we had Uber simply start placing cars on the streets and interacting with the general public under the idea of “trust us, the AVs are ready, and will be safer,” and in many instances choosing locals that would require them to do the least safety reporting, etc.

    Boeing would be crucified if they tested a new flight system by simply placing it in a fully-loaded 747 on a cross-country flight without doing 1000s of hours of testing to determine all of the systems apparent flaws and shortcomings, so why does Uber get to seemingly skip that important step?

    These systems need to be tested and developed more thoroughly in closed environments before they’re put on the streets, and the only reason they’re not being tested in that manner is there’s a huge payoff for whomever gets to market first.

  3. That’s not ‘the big picture’.
    You don’t realize that the *vast* majority of Americans don’t have the same repulsion towards cars that people who might frequent this blog do.
    Having cars in their vicinity isn’t a quality of life issue. You may note that those cars are not on their own. They are there because people have purchased them at great expense and are currently occupying them when on the road. People. Love. Cars. The number of car trips has gone *up* with the advent of car share. Not down.
    When those cars aren’t required to have an error-prone human behind the wheel who vapes and runs red lights people are going to want to ride car share *even more*.
    Transit advocates are whistling past the graveyard on this. Their preferred way of getting around is quickly becoming outmoded. And instead of attempting to find something new, or innovating in any way transit fans are doubling down, demanding more subsidies, demanding more good money after bad. You’re going to lose this one. Might as well be prepared for that.

  4. the only reason they’re not being tested in that manner

    There are AVs on the road that have been tested in exactly that manner. For instance:
    https://www.google.com/amp/s/www.wired.com/story/google-waymo-self-driving-car-castle-testing/amp

    I’m not trying to defend Uber. I’m very, very clear on where they fall on the scale of being willing to trade other people’s safety for profit. Their recent earlier AV driving in CA gives me no confidence that their systems, in particular, are ready for on-road testing.

    I’m just responding to the general position that I keep seeing that until AVs have reached an unattainable standard of perfection (or even some far, far better than human standard rather than a more reasonable better than average driver standard), having them on the road constitutes reckless testing.

  5. I agree. AVs can be grouped in a manner that allows for continuous flow through of all directions at an intersection as well as a gap long enough to operate a scramble crossing phase for pedestrians.

    From the rider’s point of view, it is a continuous trip. From a pedestrian’s point of view, every two minutes there is a gap that allows for safe crossing of the streets.

  6. come on you NIMBY. Self-driving cars still kill less people per mile than any other form of transportation

  7. The investigation is still ongoing. Some of the comments written about this issue are very disturbing. It may be determined that this incident was unavoidable for a human driver or autonomous. Based on the hundreds of thousands of miles tested by autonomous vehicles, I would like to understand the percentage of pedestrian accidents by human drivers. It would seem that autonomous are significantly safer than human drivers. Here is a question we can all ask ourselves. If there were two vehicles coming at you while crossing the street – one you knew was autonomous (with numerous cameras and sensors), and one with a human driver (who could be texting, fiddling with a phone, talking to a passenger, looking for something in the back seat, taking a sip of something – or distracted in many other ways), which one would you feel stronger to avoid?

  8. A robot will “care” about whatever it is programmed to care about, without biases, unlike a human. And it won’t flee the scene like so many human drivers.

  9. It’s a fact that self-driving cars, even in their current relatively primitive form, have been safer than regular cars. The number of injuries and fatalities per mile driven (either to vehicle occupants or to those outside the vehicle) is much lower than with normal cars.To insinuate otherwise is counterfactual to the point of being willfully deceptive and flat out immoral.

    You might as well argue that autopilot on commercial aircraft shouldn’t exist because it can’t deal with 100% of situations, despite the fact that the overwhelming majority of crashes are caused by pilot error.

  10. But you’ve got nothing to say about the fact that they’re statistically far safer for everyone than a car controlled by a person. Have some perspective.

  11. Tell us, do you know EXACTLY how autopilot on a plane works? No, you absolutely don’t. And yet you continue to fly. Do you know for an absolute fact that the tap water you use is 100% free of harmful substances? No, you don’t. But you continue to use it anyway.

  12. Good thing you aren’t doing crash investigation, because your review is demonstrably wrong. Let’s take a closer look at facts:

    1. None of the articles I’ve seen had statements with police saying she crossed from the median. Every one I’ve read said she had been on the sidewalk. That doesn’t preclude a media outlet from wrongly reporting something however.
    2. The point of impact is clearly visible in videos and pictures of the scene. It was in the vicinity of where the right turn lane taper begins, ~270′ from the intersection, on the right side of the street. That is consistent with her stepping off the sidewalk.
    3. You “assumed” the vehicle positioning, but had you actually done a real analysis you wouldn’t have to assume, and you would have come to a different conclusion. The POI is also consistent with the damage to the hood on the right, as you noted, however your assertion that the vehicle was in the inside lane along the median is impossible due to the POI (side of the debris field) and the location where the vehicle came to rest (in the right turn lane). Clearly the vehicle was in the right lane, and either pulled into the turn lane after the crash, or was entering the turn lane at the time of the crash. I’m guessing the latter.
    4. The major damage to the bike was limited to the front wheel, also consistent with her coming from the sidewalk. Had she been crossing from the median the damage to the bike would have been at the rear wheel (being hit by the right front corner of the car).
    5. The remainder of your investigation is thus rendered moot since she wasn’t in the roadway for the period of time you suggest (having come from the right, and was immediately struck).
    6. Thus, from the crash scene that is readily available there was virtually zero temporal separation between the time she started crossing and the time of impact. The POI is at the right curb. That is consistent with a lack of reaction from the vehicle or backup driver.
    7. The vehicle’s stopping point is ~85′ beyond the POI. Just the pure stopping distance is 70+ feet, not including perception and reaction time of a human driver, so it appears the vehicle immediately braked upon impact. Again, considering detail #6, the braking WAS virtually immediate if indeed she stepped off the curb, or was stopped along the curb, turned, and started crossing.
    8. The media keeps reporting this as a 35 MPH zone, but in fact it is a 45. It increases from 35 to 45 at the highway underpass. However, the speed limit drops to 35 in the southbound direction on the opposite side of the street from the crash location, which is perhaps the cause of statements by the media and perhaps even by the police.

    One main question is whether Ms. Herzberg was on the left or right side of her bike. When someone is struck broadside while walking a bike, if the bike is between them and the car it can result in a much more serious crash. The bike gets slammed into them and pushes them downward, catches on them and flings them, etc. That isn’t intended to blame the victim, but it adds to the crash dynamics and energy imparted to the human and can significantly alter the crash outcomes, often tragically.

    Lastly, it is a crash scene, not a crime scene. Your hyperbole betrays your bias. Had it been a hit & run or something similar, then crime scene would pass muster.

  13. Phenomenally poor. See below. The cheerleading on Streetsblog is often quite disconcerting.

  14. ABS doesn’t leave skidmarks. The vehicle came to rest in a very short time/distance from the point of impact, indicating an immediate response upon collision. And the speed limit continues to be improperly reported. It is a 45 zone, not 35. It drops to 35 in the opposite direction on the opposite side of the street, but the northbound lanes are 45 MPH.

  15. Your analysis is simply typical blame the victim that we hear all the time from suburbanites.

    Your entire premise is rooted in the false belief that motor vehicles somehow have the highest priority of roadway use. Your analysis presupposes Humans are interlopers on the street.

    That’s entirely wrong.

    Streets are for people

  16. I enjoy driving.

    1) Traffic jams are fun.
    2) Traffic lights are cool.
    3) Highways are already fast. For longer distances planes or high speed rail is better.
    4) Kids shouldn’t be by themselves.
    5) The car would be more expensive because of the extra feature.
    6) This accident results in a much higher rate for SDCs.

  17. Vooch: Your usage of “Streets are for People.” is creating a Liar’s Paradox. Streets are for people. People operate and ride in motor vehicles. Since streets are for people and people use motor vehicles, streets need to accommodate motor vehicles.

    Along with AVs will come the ability to right size automobiles. Cars will no longer have to be designed to survive a crash at freeway speeds. Without the worry that a “City Car” may travel on a freeway, two-seat golf cart styled (remember the Gem) can become commonplace in urban environments.

  18. Right. And expect the video to be processed and presented in a way that communicates that the collision was unavoidable. Subtle changes can be done to the imaging presented via many “legitimate” methods that will fly under the radar of the layperson.

  19. hulking death machines are not people

    streets are for people

    the onus is always on the killer operating a hulking death machine ( HDM ) who chooses to operate his HDM among people.

  20. Both motorists who hit me (not at the same time) said “I never saw you”, even though I was riding legally, visibly, in daylight, and believed I had made eye contact with the sausage truck driver stopped at the cross intersection before he rolled through it into me. Now it appears AV’s will be saying the same thing, perhaps in a sweeter tone.

  21. the cyclist’s cloak of invisibility to motorists seems to function equally well for AV’s

  22. It is not the job of the pedestrian to avoid the vehicle. It is the job of the driver or cyclist to avoid the pedestrian.

  23. Wut? I really do not think the people designing these new transit means have any idea how human beings work. Pedestrians will never be robots.

  24. It is doubtful that a police officer (or any layperson for that matter) has the expertise to analyze video evidence and determine whether or not somebody “in the shadows” could be seen.

  25. It is impossible that there will no traffic lights. That is just ridiculous. And it is impossible that there will be no cross traffic. You will always have to stop and allow others to take their turn. Pedestrians will never work the way you imagine they will.

  26. By the testimony of the “driver” and police who viewed the video, the Uber vehicle was unable to recognize the victim and stop before it killed her. By my calculation, the victim was in the path of the vehicle for two seconds before it killed her. If that is the limit of the recognition and reaction time of the vehicle, then it shouldn’t be allowed to operate at the speed that it was moving.

  27. Actually people are repulsed by other people’s motor vehicles, whether from noise, dirt, parking space taken, traffic lane jammed, just not by their own, unless the fast food wrappers build up. It actually is a quality and quantity of life issue, but not one most have thought deeply about. I think 50,000 Americans killed a year from crashes, both in and by cars reverberates widely, and that doesn’t include the early deaths from noise stress, air pollution and lack of exercise/obesity.
    Many don’t even like driving, just need to for transportation since 90+% of “transportation” dollars have been spent motor vehicles over the last century, so alternatives are either bleak or dangerous.I’m not a transit activist, because most are nostalgic attempts to recreate some mythologized past, but do advocate for innovative human powered transportation and transit. see my FB page Velorution2020.

  28. that doesn’t agree with the article- since this death happened after at most 10-11 million miles of AV travel, while human operated fatality rates are 1 per 85 million miles. They may become safer, but apparently aren’t there yet.

  29. Well, it seems that the people who want to change how vehicles travel will need to do so by changing all the laws that protect pedestrians from vehicles. The way that cyclists imagine their relationship with pedestrians can or should be is an abject failure. Pedestrians are having a terrible time dealing with the crazy ideas of cyclists.
    We cannot change the laws to make things more convenient for vehicle operators at the expense of pedestrians. Pedestrian deaths are now the highest they have been in thirty years. While these new travel people want to constantly blame pedestrians, it is really a change in the attitude of many of those on wheels. They have lost all sense of responsibility for safeguarding pedestrians.
    There is no way that sense can be programmed into a robot.
    It is impossible that cyclists or robotic cars will never, ever have to stop for pedestrians, but that is clearly the goal.

  30. just looked up the numbers for fatalities in the link to insurance industry. “only” 36,000 deaths from traffic crashes including pedestrians and cyclists. Used to be 50,000 a year in the 70’s, but improved vehicle safety (not for those they hit) and reduced drunk driving have reduced that.

  31. “you’ll pry this steering wheel out of my cold, dead hands!” Probably more true for cars than guns.

  32. flying cars? promised in my lifetime from Jetsons on up. I’m 60, and don’t expect them any sooner than full penetration of AV’s.

  33. They’re not going to share their driving info with other companies to make these machines safer either, so deaths and injuries will have to accumulate with each company before their algorythms improve. Peds will be most at risk I think.

  34. Check your facts. The overall mortality rate for pedestrians/bike riders is ~2 deaths per billion vehicle (car) miles traveled (VMT) in the US. Self-driving cars in the US have driven less than 10 million miles and have killed one pedestrian. So the current rate for AVs is >100 deaths per billion VMT. If you restrict this comparison to Uber autonomous vehicles, the mortality rate is ~ 500 deaths per billion VMT.

  35. obviously they need to install zombie or vampire features. They can smell brains or blood from a distance, like sharks.

  36. I think you’re correct on one important point: The crash appears to have occurred in the curbside lane, i.e. “Lane 2”, and not in “Lane 1”. This is based on review of a photo (which I hadn’t seen before) that clearly shows the debris field in the curbside lane.
    However, this is an even greater indictment of the AV technology in the Uber vehicle. I now believe that the victim was in the roadway for at least five seconds before being struck and killed.
    Regarding the direction of travel I continue to accept the police statements that consistently describe her movement “from west to east” or “from the median”. Can you provide any official (police) statements that contradict these statements?
    The SF Chronicle source said she: “walked from a center median into a lane of traffic” (//www.sfchronicle.com/business/article/Exclusive-Tempe-police-chief-says-early-probe-12765481.php)
    A video source states “the pedestrian was walking a bicycle from the west side of the road to the east side of the road”, according to Sgt. Ronald Elcock, Tempe police spokesman, in the imbedded video at (https://www.azcentral.com/videos/news/local/tempe-breaking/2018/03/19/tempe-police-talk-self-driving-uber-accident/33099805/)

  37. That discretion does not match the damage. The damage to the SUV if the front right. The damage to the bike is the front tire.

    I have a feeling that in the rush to get information out. mistake were made in the statement.

  38. If you step in front of a vehicle without providing them ample time to see you and stop then it’s your fault, not the drivers fault, especially if you aren’t crossing in a cross walk or are crossing against the light at a controlled intersection.

  39. I don’t see where I described this as a “crime scene”. (I know that I typed “crime scene” once, but I remember correcting that Freudian slip.)
    However, it’s conceivable that this crash scene could end up being a crime scene if Uber is found to be criminally negligent or that they exhibited a criminal disregard for public safety.
    If it turns out that my analysis is accurate (I know you have your doubts), and that the vehicle failed to “see” and react to a person who was in front of the vehicle for at least two seconds, that would be evidence that the Uber vehicle isn’t safe to operate at 38 MPH in a dark (shadowy?) environment.
    If the Uber engineers and executives were aware of the limitations of the vehicle’s sensory equipment, and they still approved its use in that physical environment, they should be held responsible for the consequences of their decisions. (Dare I say hubris and greed?)

  40. The video is out. Anyone who used phrases like “jumped out” in front of the car has been proven to be full of it. The woman had been in the road a long time, and was almost to the other side.

    It was dark, and she seems to suddenly appear, but a decent driver would have swerved and missed her.

    But a vehicle with radar shouldn’t have had to swerve. It should have seen her long before she becomes visible to the human eye. This is exactly the scenario where an AV is supposed to OUTPERFORM a human driver.

    And as for the snide comment below, I am 100 percent certain that if riding my bicycle in that situation there is no way I would have hit that woman. And if I had, there is almost no chance that I would have killed her.

Leave a Reply

Your email address will not be published. Required fields are marked *

ALSO ON STREETSBLOG