Self-Driving and Driver-Assist Technology Linked to Hundreds of Crashes, U.S. Data Shows
Source: New York Times
Over the course of 10 months, nearly 400 car crashes in the United States involved advanced driver-assistance technologies, the federal governments top auto-safety regulator disclosed Wednesday, in its first-ever release of large-scale data about these burgeoning systems.
In 392 incidents cataloged by the National Highway Traffic Safety Administration from July 1 of last year through May 15, six people died and five were seriously injured.
Teslas operating with Autopilot, the more ambitious Full Self Driving mode or any of their associated component features were in 273 crashes. The disclosures are part of a sweeping effort by the federal agency to determine the safety of advanced driving systems as they become increasingly commonplace. Beyond the futuristic allure of self-driving cars, scores of car manufacturers have rolled out automated components in recent years, including features that allow you to take your hands off the steering wheel under certain conditions and that help you parallel park.
In Wednesdays release, NHTSA disclosed that Honda vehicles were involved in 90 incidents and Subarus in 10. Ford Motor, General Motors, BMW, Volkswagen, Toyota, Hyundai and Porsche each reported five or fewer. These technologies hold great promise to improve safety, but we need to understand how these vehicles are performing in real-world situations, said Steven Cliff, the agencys administrator. This will help our investigators quickly identify potential defect trends that emerge.
Speaking with reporters ahead of Wednesdays release, Dr. Cliff also cautioned against drawing conclusions from the data collected so far, noting that it does not take into account factors like the number of cars from each manufacturer that are on the road and equipped with these types of technologies. The data may raise more questions than they answer, he said. About 830,000 Tesla cars in the United States are equipped with Autopilot or the companys other driver-assistance technologies offering one explanation why Tesla vehicles accounted for nearly 70 percent of the reported crashes.
Read more: https://www.nytimes.com/2022/06/15/business/self-driving-car-nhtsa-crash-data.html
hlthe2b
(102,376 posts)(supposed driver) was totally unengaged. What fury would emerge...
JustABozoOnThisBus
(23,367 posts)... where the driver was totally unengaged? Also fury?
Self-driving tech has a way to go. So does human driver training and ability. Both could use improvement.
hlthe2b
(102,376 posts)responsibility. Not sure how you can not get that.
JustABozoOnThisBus
(23,367 posts)Being more expensive, the odds are better that it's insured. Getting a computer to take responsibility, vs getting a texting driver to take responsibility? What are the odds?
Response to hlthe2b (Reply #1)
JustABozoOnThisBus This message was self-deleted by its author.
hlthe2b
(102,376 posts)JustABozoOnThisBus
(23,367 posts)hlthe2b
(102,376 posts)Bernardo de La Paz
(49,043 posts)About 100 per day. Averaging more than one person dead per crash.
Of 392 tech vehicle crashes, only six deaths.
BumRushDaShow
(129,491 posts)it's "preventable" due to the completely artificial means.
But I think more obviously, you would have to look at accidents per total number of those types of vehicles on the road vs "all".
Bernardo de La Paz
(49,043 posts)BumRushDaShow
(129,491 posts)A "self-driving car" is not impacted by "distracted driving" like a human-controlled vehicle, unless there is a software flaw that causes it to "task switch" to do something non-critical at the wrong time.
Bernardo de La Paz
(49,043 posts)A 98.5 percent drop in death rate is "not ready for prime time"?
(Yes, I know, it is not true "per cent", but it is a convenient hook to use.)
BumRushDaShow
(129,491 posts)so trying to do a true "stats analysis" is meaningless unless you compare apples to apples.
Bernardo de La Paz
(49,043 posts)When doing testing and trials, a sample of 30 is sufficient in many statistical situations to get 95% confidence.
390 crashes is not the sample size. The sample size (metaphorically, because it is 100% sampling) is the total number of techy vehicles which is well north of a million. That is a large sample.
I take it to mean that most techy vehicle crashes are much lower speeds, probably due to considerable braking even if it was initiated too late. Many deadly human driven crashes don't have any braking.
Comparing crashes to crashes IS THE comparison. It IS apples to apples. A crash is a crash according to whatever criterion is used to tabulate the figures (probably when there are police reports plus a minimum say $500 damage, but I don't know).
BumRushDaShow
(129,491 posts)~826,000 cars out of a total of over 3/4 billion (~278,000,000) cars in the U.S.
That's 0.2%.
And worldwide, it's estimated that there are ~43 million such autonomous or semi-autonomous vehicles out of a worldwide total of upwards of 1.4 billion vehicles, which is about 3%.
The report was summarized here in a good article here-
By Andrew J. Hawkins@andyjayhawk Jun 15, 2022, 9:00am EDT
(snip)
ADAS crashes
Teslas numbers were much higher than other companies, most likely due to the fact that it sells more vehicles equipped with Level 2 systems than its rivals. (One estimate puts the number of Tesla vehicles with Autopilot or Full Self-Driving at 825,970.) Tesla collects real-time telematics data from its customers, giving it a much faster reporting process. Other automakers typically have to wait for reports to arrive from the field and sometimes dont receive them for months. From July 20th, 2021, to May 21st, 2022, there were 273 crashes involving Tesla vehicles using Autopilot, according to the report.
The EV companys crashes represent the bulk of the total 392 crashes reported during that period. Other automakers didnt come close to Teslas number of reported crashes. Honda, which sells its ADAS features under the brand Honda Sensing, disclosed 90 crashes. Subaru, which packages its ADAS under EyeSight, reported 10 crashes. Ford disclosed five crashes, Toyota reported four crashes, BMW reported three crashes, and General Motors, maker of Super Cruise, only disclosed two crashes. Aptiv, Hyundai, Lucid, Porsche, and Volkswagen each reported one crash.
Of the 392 crash reports, only 98 included information about severity. There were six crashes that resulted in serious injuries and five that resulted in fatalities during the nine-month reporting period. NHTSA officials did not disclose the manufacturers that reported the fatalities but said that information would be included in the raw data released Wednesday.
Throughout the briefing, officials declined to address specific questions about Tesla, cautioning against drawing conclusions about any one company. But theres no question that Tesla is an outlier when it comes to driver-assist technology.
(snip)
https://www.theverge.com/2022/6/15/23168088/nhtsa-adas-self-driving-crash-data-tesla
Because you are trying to compare different "levels" of autonomous functionality against other levels, then it completely skews the statistics, let alone try to compare to "all" (autonomous, no matter how much, plus non-autonomous).
The graph in the above excerpt gives a good idea of attempting to compare "apples to apples" and what the result was (notably when it comes to Tesla).
Bernardo de La Paz
(49,043 posts)The statistics given are for ALL accidents of non-autonomous, where the death rate is above 1.0 per accident,
... and ...
for ALL accidents of autonomous or "assisted", where the death rate is less than 0.02 per accident.
Comparing ALL to ALL is apples to apples.
BumRushDaShow
(129,491 posts)and "software", which is where this argument started and then devolved.
People can and do get "distracted" and that along with certain road conditions and possible faults in the vehicle itself, can cause an accident.
However the point of the "automated vehicle" was to remove the "distraction" part out of the equation and strictly react to the road conditions and/or conditions of the vehicle, and they are not there yet.
If the "software" is engaged at night and misses a deer darting out into the roadway that a human can "see" moving in the woods from outside of the car's camera/sensory periphery, then the human could "react" faster while the car does so too late.
So no, it's still not "apples" to "apples".
The graph does a "software to software" (company to company) comparison between types of vehicles with similar functionality and it suggests that improvements need to made for that one company within the meaning of "autonomous". But then throwing that into looking at accidents out of total vehicles is meaningless except maybe for minimizing the importance of improvements.
Bernardo de La Paz
(49,043 posts)I'm sorry I took about 20 minutes to compose a detailed reply since you raised some good issues, but then the DU software obliterated it because I (distractedly) press the "go back a page" button. A "preventable" catastrophe if the software were improved. I'm sorry I don't feel like spending that amount of time again.
Happy Hoosier
(7,392 posts)Because it's pretty much impossible to quantify the number of accidents the technology potentially prevented.
I work in automation (in aviation), and the trust factor is huge. People are generally emotional and reactionary. In our studies, humans tend to not trust automated technology unless it is about 10 times safer than human control. Sometimes that number is even higher, depending upon the perception of how dangerous a particular operation is.
So generally speaking, people are not great judges of the actual risk, because risk tolerance is not necessarily rational.
BumRushDaShow
(129,491 posts)And there you have the big difference.
Although you might have multiple aircraft in the same airspace, unless you have a formation of fighters (jets or choppers), the aircraft are much further apart by regulation.
This is why whenever I see sci-fi films continually showing "jet cars", I always chuckle that we can't even drive in "1 dimension" (forward and back) let alone "3 dimensions" (forward/back - up/down - side to side)... and then I often throw this example in to illustrate -
Happy Hoosier
(7,392 posts)... pretty much anything which requires a genuine "see and avoid" ability.
IMHO, I would not trust a car autopilot without close supervision. I'd treat it more like a "pilot relief" mode instead of a set it and forget it mode.
But my point was really about peoples' terrible risk assessment skills. People are awful at intuitive assessment of risk. Their subjective perceptions are wrong a huge percentage of the time. Flying commercial air in the USA is 200(!) times safer than driving a car on both an hourly AND a per cycle basis. And yet some people are TERRIFIED of flying and simply won't do it.
In my line of business, we have a motto "At least 10 times better or forget it" and sometimes THAT is not enough.
BumRushDaShow
(129,491 posts)This type of automation IS going to happen... eventually. We are in the embryonic stages of it. Aircraft have it, watercraft have it, and spacecraft have it. The one big difference is that the former 3 are being operated in a less congested space than a car, and that is where it will be tricky to improve to that level of granularity, but it will happen one of these days.
The closest to it is continually being updated in trains (that do travel on a track) and things like a PTC, as we know how "automated" and "distracted" can equal what happened here in Philly 7 years ago.
Deminpenn
(15,290 posts)realistic simulators by pilots. Not sure that's the case with car/truck/suv autopilot software.
Happy Hoosier
(7,392 posts)You're right... in aviation, our stuff is tested out the wazoo and has to be certified. There isn't any consistent standard for ground transport.
Lars39
(26,116 posts)ananda
(28,876 posts)But I think they're good and helpful.
I can't see how they'd cause crashes, but
the only time the brakes come on automatically
is during cruise control.
I like the screen flash and beeps that come on
when I'm backing out of a parking space, for
sure.
BumRushDaShow
(129,491 posts)and also includes the "self-park" feaature that I have never used.
Happy Hoosier
(7,392 posts)I have an automatic braking feature that does not require cruise control that saved my butt at least once.
And I do REALLY like the adaptive cruise control.
I like the Lane Keeping Assist just fine, but I'd never depend upon it to actually drive the car (and the car won't keep it engaged if the driver isn't providing at least occasional steering input.
I DO like that the car wiggles the steering wheel if it thinks I'm leaving my current lane. That hasn't saved me yet, but I can imagine it would help prevent me from falling asleep and drifting off the road.
ChazII
(6,206 posts)day making deliveries.
Miguelito Loveless
(4,474 posts)Speaking with reporters ahead of Wednesdays release, Dr. Cliff also cautioned against drawing conclusions from the data collected so far, noting that it does not take into account factors like the number of cars from each manufacturer that are on the road and equipped with these types of technologies.
ripcord
(5,537 posts)I think I can safely drive my own car.
Lil Liberal Laura
(228 posts)Yummy.
NullTuples
(6,017 posts)"We'll do all your maintenance...but you have to agree not to report accidents to the authorities" for years made it look like Teslas were far safer and more reliable than they turned out to be.