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."
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."
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.
4 Out of 48 Have Accidents, None Their Fault
I think that compares well to the average Californian.
Perhaps you would like to read the story somewhere other than the NYT because paywalling is stupid and offensive, even if you know how to bypass it. Thanks, Seattle Times, for just showing me the flipping article.
Here's a news flash: You can get the same AP newswire article anywhere. Yet people still link the NYT. That's poor internet etiquette given that they paywall.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
'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.
This is all well and good having self driving cars, however, (ok, so this is through reading the unfortunate things that happen to people through stupidity, and malice on reddit, liveleak, etc)....
Hopefully we'll never get to the stage where you just 'pop the infant' into the car and tell it to go to grandmas (assuming grandma is there, and nothing happens along the way). Sad, far fetched, but you can bet that this will happen somewhere and some unfortunate may be hurt as a result in something entirely preventable).
This was a waking dream this morning. Sad thing to wake up to. Hopefully never happen.
Computers are still too stupid.
One of the analogies I've seen at a speech on the subject went something like, "A computer can detect an object in the roadway, but it can't yet tell if it's a paper bag that can be safely run over or a rock that's apt to damage the car."
This was a few years ago, so I don't know if it's still true or not, but it does demonstrate the programming challenge in processing something exceedingly simple that even the most inexperienced human driver would be able.
I want peace on earth and goodwill toward man.
We are the United States Government! We don't do that sort of thing.
To be fair, 2 of the accidents happened while under human control. That suggests that yes, the computers are at least as good as the humans... That said, the sample size is tiny, and critical info like miles driven is missing, so who knows.
While I get that the autonomous system wasn't at fault, the real question is, "Were the accidents something a human driver could have avoided?" I'm a professional delivery driver and one of the things that's constantly drilled into us is to essentially watch out for the stupid people. Sometimes you have to yield right-of-way because it's clear the other driver isn't going to. Do autonomous cars know that?
"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.
The autonomous may not have been at fault, but one wonders whether some of the accidents would have been avoidable by a fleshy driver.
That the vast majority of human drivers around them were able to avoid accidents despite the presence of dangerous automated cars.
You may consider it a lot, but when you consider NONE were the fault of the driver or the car itself then it is nothing at all. We cannot always get out of the way of a driver who is not paying attention. Why do you seem to be blaming the autonomous cars and not the people who caused those accidents?
When you cant win, ad hominem.
Self-driving cars need to be banned.
Why - it wasn't their fault?
You are not considering the mileage driven. These cars are on the road for 100k miles + a year, so consider that 4 out of 720 cars were in an accident.
When you cant win, ad hominem.
American components, Russian components, they're all made in Taiwan.
the preceding comment is my own and in no way reflects the opinion of the Joint Chiefs of Staff
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.
One can be "in the right" and still not have done the right thing. For example, if the light is green I'm in the right not to slow down for the intersection. But that doesn't mean I shouldn't take precautions to check if someone is coming the other way. If I had I might have avoided the accident that was not assigned to my "fault".
On the other hand it's also possible that google cars will be better drivers than the average person. One might hope they use different CPUs for the texting and the driving.
Some drink at the fountain of knowledge. Others just gargle.
It's still better than letting computers drive. I've heard the damn things even sleep on the job. I've also heard they have a dog watching while they're asleep but that doesn't make any sense.
Get free satoshi (Bitcoin) and Dogecoins
How does one make eye contact with an autonomous vehicle at an intersection, or when merging lanes? Human drivers will have to learn a separate protocol.
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.
Get free satoshi (Bitcoin) and Dogecoins
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?
Big issue with AI controlled cars in a human-dominated traffic is that AI doesn't react the same way people do. Sure, all-AI traffic would likely be more efficient and less prone to accidents, but we are nowhere near this. Instead we have AI that to humans is hard to predict.
For example, huge puddle on the road, most humans would unwisely drive through it. What would AI do? No idea, and I wouldn't want to be driving behind it when that happens. What about a hobo at the end of the offramp begging for change? Would AI freak out about pedestrian on the road? No idea, and I wouldn't want to be driving behind it to see what happens.
Honestly, you can say it wasn't their fault, but nearly 10% of them in 6 months have been involved in accidents. Even if it wasn't the fault of the technology itself, why is the accident rate so high?
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
It also means that the cars aren't driving autonomously at all times.
To me this implies that there simply isn't any comparable data yet.
Slashdot social media options: AIM, ICQ, Yahoo, Jabber and Mobile Text. Why no MySpace?
You are not considering the speed they're going at and which roads they are going on. It's easy to avoid accidents when going sub-25 speeds on a predefines subset of roads. Whether you're human or not.
Until we see some data on how autonomous cars do on all kinds of roads and driving speeds and conditions, I don't think we should extol their safety. Going 55 mph over a hilltop on a country road, or avoiding a deer is a bit different. Or a busy bumper-to-bumper city street where no-one will let you over in the next line unless you force the issue.
I'd also ask how long it took for the car to get from A to B, and how it compares to a human driver. Time is important to people; enough so that we're willing to deal with risks to save time.
Actually theres a historical precedent for legislation concerning semi autonomous non human traffic. Horses.
It's not entirely as ridiculous an analogy as it initially sounds.
This is why "no fault" insurance was invented. Even with people, the liability in traffic accidents is often complex. Require no-fault insurance, you insurance is responsible for you, mine is responsible for me, and you are basically done - at least as far as car insurance is concerned.
What may be left is civil liability. If a manufacturer produces a genuinely faulty product, they can be sued in a completely separate action. The laws are already in place for that. The problem I see is the exact opposite of what you are worried about: People will sue, regardless of the quality of the product. Because bad things aren't supposed to happen, and if they do, someone must be to blame.
So someone without insurance t-bones my autonomous car, and I want compensated. The other driver is broke, so I sue my car manufacturer. Stupid, but entirely possible under US law. Loser pays would be the simplest solution: if you file a stupid lawsuit, you'll be paying the other side's legal costs. In the case of class action suits, the attorneys for the class should be liable for the loser's legal costs if they lose.
Before any company brings autonomous cars to market, the US tort system has got to be fixed.
Enjoy life! This is not a dress rehearsal.
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.
Because someone being blamable for accidents is much more important than having fewer accidents in the first place, right?
The major part of safely driving is anticipating another driver breaking the law. Doesn't look like computers can do that yet.
Way to not demonstrate any understanding of that axiom. If self-driving cars are highly correlated with an increased number of accidents, is that not something to be concerned about?
If you want a vision of the future, imagine a youtube comments section scrolling - forever.
Accident rate in general: 4-5%
Accident rate so far with only 48 vehicles: 8-9%
Without details I'm not going to accept "someone else's" fault as fact.
Another sticking point: A traffic cop in the street directing traffic,
Are these cars able to deal with traffic cops? I kinda doubt it.
This is an ex-parrot!
Accident rate in general: 4-5% Accident rate so far with only 48 vehicles: 8-9%
Ah, no, the meaningful rate is accidents per miles driven, not per car. These cars are driven constantly (for testing, data collection, etc), surely way more that twice the average.
Honestly, you can say it wasn't their fault, but nearly 10% of them in 6 months have been involved in accidents. Even if it wasn't the fault of the technology itself, why is the accident rate so high?
Because they are constantly on the road, for testing and data collection. I would imagine that they drove more in these 6 months than many cars do in 6 years.
if someone is shooting at car, how does the car react?
As long as it's driveable, it should keep on driving - precisely the opposite of what most people would do.
Really? I'm pretty sure, if somebody started shooting at my car, it would be pedal to the metal until I was sure they were a long way behind me.
All I want is a secure system where it's easy to do anything I want. Is that too much to ask ~~ Randall Munroe
16 is the first statistically significant sample size.
Statistics do not work that way! Goodnight!
systemd is Roko's Basilisk.
I believe the summary said that that humans were in control of the cars during at least 2 of the 4 incidents. You can't blame the self-driving car if the self-driving feature is disabled and the human takes over. That would be like blaming Google Maps for bad directions if you turn it off, take a left turn when Google had said to turn right, and wind up lost.
With a 2 out of 48 accident rate, that's 4%. Of course, that's a very small sample size. It would be interesting to see how the accident rate changes with many more autonomous cars on the road.
My sci-fi novel, Ghost Thief, is now available from Amazon.com.
That's a good start, but there's a lot more to it than that. An experienced driver can tell when a car in the next lane wants over (even though it isn't bothering to signal), and can even guage its desperation based on how bad the "body" language is getting, and proximity to things like intersections, exits, etc. An experienced driver knows when the driver behind them in traffic is an accident waiting to happen (is that dude reading a copy of Ivanhoe or something?), and get into another lane. An experienced driver knows that the left lane across from a Starbucks is to be avoided in the morning, particularly if there are multiple late-model coupes and sedans in that lane.
You could probably put all that smarts into an AI, but I doubt its there yet.
Sigh. If you took a sample of 48 US human-driven cars, you'd have a 20% chance of getting 4 accidents out of that 48.
I'm not trying to say that's huge, but that sample size of 48 is far too small to draw conclusions from statistics that contain millions of samples. The error bars are massive.