Slashdot Mirror


Self-Driving Cars In California: 4 Out of 48 Have Accidents, None Their Fault

An anonymous reader writes: The Associated Press reports that 48 self-driving cars have been navigating the roads of California since the state began issuing permits last year. Of those, only four have been in accidents, and none of the accidents were the fault of the autonomous driving technology. Seven different companies have tested autonomous cars on California's roads, but Google, which is responsible for almost half of them, was involved in three of the four accidents — the other one happened to a car from Delphi Automotive. All four of the accidents happened at speeds of under 10 mph, and human drivers were in control during two of them. The Delphi accident happened when another car broadsided it while its human driver was waiting to make a left turn.

The AP pieced together its report from the DMV and people who saw the accident reports. But critics note that there aren't direct channels to find this information. Since one of the chief selling points of autonomous cars is their relative safety over cars piloted by humans, the lack of official transparency is troubling. "Google, which has 23 Lexus SUVs, would not discuss its three accidents in detail." Instead, the company affirmed its cars' accidents were "a handful of minor fender-benders, light damage, no injuries, so far caused by human error and inattention."

17 of 408 comments (clear)

  1. Not yet statistically significant by bradley13 · · Score: 4, Insightful

    I expect the number haven't been publicized, because they are still to limited to have any significance, and also because the cars have been running under fairly tightly controlled conditions.

    When there are a few hundred cars, running in all kinds of weather and traffic conditions, with millions of miles - if the numbers are still good, you can bet that they will be plastered all over the internet

    --
    Enjoy life! This is not a dress rehearsal.
    1. Re:Not yet statistically significant by Andy+Dodd · · Score: 2, Insightful

      There's also: "Since one of the chief selling points of autonomous cars is their relative safety over cars piloted by humans, the lack of official transparency is troubling."

      No it is NOT a selling point, because NO ONE is selling these cars yet. It is EXPECTED to be a selling point once development is complete - WHICH IT IS NOT.

      That said, it would be interesting to hear the details of Google's two autonomous accidents.

      Also, the headline is misleading... While a car may be capable of self-driving, if a human is in control when an accident occurs, then the car was not a self-driving one as far as the accident goes.

      As far as the national statistics (0.3 accidents per 100,000 miles) - those are national statistics, averaged across the entire country. Google's accidents all occurred with mileage racked up in the Bay Area, which is probably one of the worst places in the country to drive as far as hitting other vehicles.

      --
      retrorocket.o not found, launch anyway?
    2. Re:Not yet statistically significant by Archangel+Michael · · Score: 2, Insightful

      I suspect that "none are at fault" is probably true. But what is often left unsaid, the cars, while being legal, were doing something unexpected.

      My Great Aunt, had four car accidents in two years. None were her fault, yet they all kind of were. She was doing things in unexpected ways, that were completely legal, but not ordinary. People expect certain patterns, and when someone is outside of those patterns, it causes accidents. Not the fault, but rather the cause.

      --
      Agent K: A *person* is smart. People are dumb, stupid, panicky animals, and you know it.
    3. Re:Not yet statistically significant by HangingChad · · Score: 4, Insightful

      Yes, the data set is way to small to draw any conclusions,

      Not necessarily. Pick a pool of 48 cars at random and compare the accident rates. You also have to compare them by the accident rate per hour behind the wheel.

      This gets at the whole idea that self-driving cars have to meet some lofty standard of perfection to become the optimum choice. To replace people behind the wheel self-driving cars only need to be +1 better than human drivers.

      Self driving cars can't drive in the rain. Oh, really? Take a drive around Seattle in the rain, you'll discover human drivers suck in the rain, too. And that's in the rain capitol of the world where you'd expect people to be used to driving in the rain and they still suck (I lived there for 10 years so don't bother trying to deny it).

      The biggest obstacle to self-driving cars isn't rain or snow, it's something called Illusory Superiority. The vanity of humans who think they're better drivers than they really are.

      --
      That's our life, the big wheel of shit. - The Fat Man, Blue Tango Salvage
  2. Editorializing... by Junta · · Score: 2, Insightful

    '48 self-driving cars have been navigating the roads... Of those, only four have been in accidents'

    I know that the bigger point is that zero (known) incidents can be traced to the software making a 'mistake' (though even if the other driver is 'at fault', hard to say if a human would have done better at avoidance). The thing that strikes me though is the editorial bias here. *Only* 4 out of 48.. that's nearly 10%. That's far far above the percentage for the general population. It's perfectly likely that is simply a fluke of the small sample size, but implying that 4 out of 48 is a very promising rate of incident is pretty silly.

    --
    XML is like violence. If it doesn't solve the problem, use more.
  3. magenta line by fche · · Score: 3, Insightful

    "a handful of minor fender-benders, light damage, no injuries, so far caused by human error and inattention"

    In case any of those were done by human co-drivers in automated vehicles, this does not exonerate the automation from some share of responsibility. For example, if the presence or habitual use of the automation makes it more likely for the co-driver to become inattentive, it's partly to blame.

  4. Re:that's fine by Chrisq · · Score: 3, Insightful

    Self-driving cars need to be banned.

    Why - it wasn't their fault?

  5. Re:Not convinced by tibit · · Score: 2, Insightful

    The giveaway with autonomous cars is the need for centimeter precision in their navigation maps.

    So, you're making shit up. At least we know who you're paid by.

    --
    A successful API design takes a mixture of software design and pedagogy.
  6. Re:questions by ArcadeMan · · Score: 3, Insightful

    Has it been tested on drivethru fastfood?

    Not important. In the future, you'll order food online and a drone will deliver it to your moving car. After all, if you're not driving you'll be able to eat while you're in the car.

  7. Re:Fault may not be the right measure. by tsqr · · Score: 4, Insightful

    One can be "in the right" and still not have done the right thing.

    Pretty much what I was thinking. Back when the Earth was a molten mass and I was taking Driver Education in high school, there was a lot of emphasis on "defensive driving"; in other words, expect the other guy to do the wrong thing, and be ready for it. When you have a mix of self-driving and human-operated cars on the road, the self-driving ones better have some extremely conservative defensive driving skills.

  8. Re:that's fine by delt0r · · Score: 3, Insightful

    So who is a fault when breaks fail. It has happened. Who is a fault when tires blow out? Or fuel tanks catch fire, or airbags improperly deploy?

    Liability is nothing new and ToS cannot waive rights that are not waive able.....

    --
    If information wants to be free, why does my internet connection cost so much?
  9. Re:that's fine by hummassa · · Score: 5, Insightful

    NON-Self-driving cars need to be banned. There, I fixed it for you. :D

    --
    It's better to be the foot on the boot than the face on the pavement. ~~ tkx Kadin2048
  10. Re:AI is not predictable to humans by Pumpkin+Tuna · · Score: 3, Insightful

    I'm pretty sure that an AI car is never going to "freak out." The worst it might do is slow the heck down, which you, the person following, should be ready for anyway.

  11. It *IS* their fault by Anonymous Coward · · Score: 2, Insightful

    From the story:
    2 out of 48 have crashed by computer.
    1 out of 48 have crashed with human driver.
    1 out of 48 had someone crash into them at a junction

    Denial doesn't fix bugs here.

    It's likely they just miss a lot of the subtlety of driving, the "I think person X will do Y so I'm going to adjust my driving by Z" that goes on in peoples minds.

  12. Re:that's fine by beelsebob · · Score: 5, Insightful

    Because someone being blamable for accidents is much more important than having fewer accidents in the first place, right?

  13. Re:that's fine by knightghost · · Score: 4, Insightful

    The major part of safely driving is anticipating another driver breaking the law. Doesn't look like computers can do that yet.

  14. Re:Fault may not be the right measure. by admin7665 · · Score: 2, Insightful

    When you have a mix of self-driving and human-operated cars on the road, the self-driving ones better have some extremely conservative defensive driving skills.

    Also, expect humans to try and game the self-driving car to their benefit in trafic situations. Can't wait to see the behavior of people once they read that "changing lanes with a self-driving car parallel to you is as simple as trying to ram it - it will apply emergency brake and let you pass."