TMC is an independent, primarily volunteer organization that relies on ad revenue to cover its operating costs. Please consider whitelisting TMC on your ad blocker or making a Paypal contribution here: paypal.me/SupportTMC

Autopilot reduces accident rate 50% - Musk claims - valid?

Discussion in 'Model S' started by calisnow, May 25, 2016.

  1. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,135
    Location:
    Los Angeles
    #1 calisnow, May 25, 2016
    Last edited: May 25, 2016
    This news is over a month old but I have seen no threads on it and keep forgetting to make one. In an interview in Norway Musk made an off-hand remark that the early data, based on about 50 million miles of autopilot driving (now there's double that - about 100 million miles) shows that airbag deployments are 50% lower with autopilot on. He said that autopilot is about twice as good as a human driver.

    "“The probability of having an accident is 50% lower if you have Autopilot on. Even with our first version. So we can see basically what’s the average number of kilometers to an accident – accident defined by airbag deployment. Even with this early version, it’s almost twice as good as a person.” - Musk's direct quote while speaking to Norway's transport minister in mid April 2016.

    This claim of his got repeated in at least a dozen different articles, none of which bothered to attempt to get Musk to clarify his claim. The obvious questions to ask are - is the autopilot safety data at least based on an attempt to equalize conditions? IE same roads, same time of day, etc? Or is it just a blind autopilot-on vs autopilot-off comparison - which would be pretty meaningless given that autopilot use is concentrated on the highways.

    It's disappointing that not a single journalist tried to get Tesla to clarify Musk's comments. If autopilot really does reduce accidents 50% in equalized conditions - that's huge. I have noticed that Musk has never made the claim again, nor is it on Tesla's website. You'd think that Tesla would be shouting this from the hilltops if the data is robust.

    What are your thoughts?

    Here is the original post by Elektrek. There are many more articles available with a Google search in different publications, but none that I've found dive into the data even a slight bit.

    Tesla’s Autopilot lowers probability of having an accident by 50% based on early data, says Musk

    Yesterday, Tesla's autopilot head Sterling Anderson said at the Emtech conference that Tesla now has 100 million autopilot miles of data - but there are no reports yet that he discussed accident rates:

    Tesla Provides Autopilot Update: 100 Million Miles Driven
     
  2. deonb

    deonb Active Member

    Joined:
    Mar 4, 2013
    Messages:
    3,020
    Location:
    Redmond, WA
    Based on previous times Elon touted Model S safety stats, I think this is purely accidents per miles driven.

    And thus, not a meaningful comparison.

    It may still be safer, but that's not a meaningful way to measure it.
     
  3. Muhammad

    Muhammad Member

    Joined:
    Apr 1, 2016
    Messages:
    15
    Location:
    Bay Area
    I think it is because one would probably take control of autopilot right before a crash therefore removing autopilot of any liability since autopilot wasn't technically "on" when the crash occurred. Personally, autopilot has probably tried to kill me 4-5 times in the past 6 months or so and I definitely don't think it is in any way safer than humans behind the wheel who have the ability to perceive and navigate the drivers and environment around them, not just 100 feet in front of them.
     
  4. cantdecide

    cantdecide Member

    Joined:
    Dec 21, 2012
    Messages:
    271
    Location:
    Portola Valley, ca
    I agree with everyone here... I don't it is highway to highway comparison.

    I also expect that most autopilot accidents are where it wanders outside the lane (as it does to me about once every 5 minutes) and those don't result in airbag deployment.

    I expect most airbag deployment is at intersections, and that isn't relevant to autopilot.

    My experience from 2000 miles of autopilot is that it wanders out of land every 5 minutes and I have to make an emergency takeover about every 30-60 minutes to prevent a certain crash. That is about as well as I would expect a six year old child to drive.
     
    • Like x 2
  5. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,135
    Location:
    Los Angeles
    That's the opposite of my experience - mine is rock solid except in very, very difficult conditions. Are you talking about highway driving? Every 5 minutes it wanders? Have you taken it to a service center to have the system checked out?
     
  6. kirkbauer

    kirkbauer Member

    Joined:
    May 31, 2015
    Messages:
    490
    Location:
    Atlanta, GA
    A quick Google Search tells me the average is 185 accidents per 100 million miles. Given that I have only heard of two accidents that were possibly related to AP and given that we have 100 million miles it seems that he may be understating the benefit. In my experience the vehicle does great especially in stop-and-go traffic (where I probably would be much more distracted than it is).

    If there had been anywhere near 185 accidents in 100 million miles on AP I think we'd be hearing a lot more about it.
     
  7. electracity

    electracity Active Member

    Joined:
    Jun 8, 2015
    Messages:
    1,797
    Location:
    60606
    I agree. The important point today is that current autopilot does not appear to be less safe than conventional driving.

    I think it is encouraging that the most excitable portion of Tesla supporters didn't jump all over the quote as certain proof of autopilots superior safety.
     
  8. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,135
    Location:
    Los Angeles
    My take based on a few months of ownership is that autopilot is certainly safer than a distracted driver - which is many drivers texting, doing makeup, etc. But it's probably not safer than a vigilant driver who keeps their eyes on the road. Me, I know I'm not perfect in my vigilance so I use autopilot every time I get on the freeway - plus it makes driving much more pleasant and less fatiguing.
     
  9. RogerHScott

    RogerHScott Active Member

    Joined:
    Mar 2, 2016
    Messages:
    1,347
    Location:
    Lafayette, IN
    I have heard that driver-assistance technology, in general, trades more frequent minor accidents for significantly fewer major accidents, particularly those involving fatalities. At least as a starting point, that seems like a good trade. Then we can work on reducing those
    minor accidents, too. You can (legitimately) complain all you want about weaknesses in AP being annoying, but until you can cite some
    actual data showing that AP has caused actual accidents (not just scary near-misses) you can't use that mere annoyance to undermine any claims about improved safety due to AP.
     
    • Like x 2
  10. jerry33

    jerry33 S85 - VIN:P05130 - 3/2/13

    Joined:
    Mar 8, 2012
    Messages:
    12,752
    Location:
    Texas
    I'm pretty sure the majority of accidents happen during a few seconds of inattention. It's hard to see how Autopilot wouldn't help this. Whether or not it reduces it by 50% likely depends on how you cook the statistics.
     
  11. mikeash

    mikeash Active Member

    Joined:
    Oct 26, 2014
    Messages:
    1,107
    Location:
    Fairfax, VA, USA
    By far the most accidents I see are rear endings due to inattention. Around here, they seem to outnumber all others by probably 9:1. Autopilot is essentially immune to this particular kind of accident, so I find it totally believable.
     
  12. RogerHScott

    RogerHScott Active Member

    Joined:
    Mar 2, 2016
    Messages:
    1,347
    Location:
    Lafayette, IN
    Well... if you have collision avoidance you don't actually need AP to prevent read-endings. I think there's a tendency for manufactures
    (Tesla definitely included) to lump Level 1 automation features (e.g. collision avoidance) together with Level 2 features (e.g., AP) when touting the benefits of their Level 2 gadgetry. Though, to be fair, Tesla themselves hasn't actually been making much noise about this (as noted above).
     
  13. Cosmacelf

    Cosmacelf Active Member

    Joined:
    Mar 6, 2013
    Messages:
    3,399
    Location:
    San Diego
    Answering the OP question, I don't believe Elon is comparing highway to highway driving, and thus his safety statistic is meaningless.

    For me, AP leaves me far less tired after a long drive than I would have been without AP. Yes, I do still have to pay attention, and if I'm driving fast, like > 82 mph, I have to take over from time to time. But AP is still a nicer way to drive.

    Is it safer? You'd have to look at proper statistics comparing apples to apples to answer that one. I would argue it is still a worthwhile feature even if the increased safety is minor.
     
  14. Villa-Lobos

    Villa-Lobos Member

    Joined:
    Apr 4, 2016
    Messages:
    61
    Location:
    Whitefish
    After 10,000 miles driving from Seattle to Wash D.C. as well as trips on I 90,80,70,76,71 I am a big believer in AP. Would not own a Tesla without it. So much nicer way to go on log trips. Rarely use it on two lane roads as not reliable there yet, plus the car is so fun to drive on those roads anyway!
     
  15. Lump

    Lump Active Member

    Joined:
    Mar 31, 2013
    Messages:
    2,101
    Location:
    So. Cal.
  16. Cosmacelf

    Cosmacelf Active Member

    Joined:
    Mar 6, 2013
    Messages:
    3,399
    Location:
    San Diego
    Yes indeed. Edge case. That's why you must pay attention. Tesla has never ever said that the car will change lanes by itself with no user input. It is lane keeping only. Given these limitations, the best thing the car could have done is stop, but even that isn't ideal in this traffic. The driver should not have let AP continue to drive as soon as he realized there was a bus in the way.
     
  17. Cosmacelf

    Cosmacelf Active Member

    Joined:
    Mar 6, 2013
    Messages:
    3,399
    Location:
    San Diego
    By the way, the other reason Elon's stats are wrong is that some accidents have been known to happen when AP is turn off, yet the driver thought it was on. In Elon's world that would count as a non AP accident when it clearly should be counted as an AP accident.

    AP also has a habit of turning itself off when it gets into a tough situation. If the driver doesn't immediately take over, it can lead to a bad accident which technically would be classed as a non AP accident, but clearly was caused by relying on AP.
     
    • Like x 1
  18. whitex

    whitex Member

    Joined:
    Sep 30, 2015
    Messages:
    567
    Location:
    Seattle area, WA
    Actually, most reported accidents with AP seem to be AP rear-ending a stopped car. I think there is some issue with AP detecting stopped cars . I think when the radar sees something moving towards you at a speed lower than your car is going, it can safely assume it's a car traveling along with you, but when it senses something coming at you at the same speed as your car, it can't always figure out whether it's another car in your lane or did the radar happen to bounce off a stationary object on the side of the road, which I'm sure happens all the time, especially on curves. Just my theory btw.

    Then of course there is an issue that if you do hit the brakes before your car rear end a car (as in recently discussed here crash), Tesla officially doesn't count that as accident while on AP because hitting the brakes disengaged AP just before the crash. That skewes the stats. Personally I do not believe for a second that AP reduces accidents by 50%, except if you present the stats just right.
     
  19. Cosmacelf

    Cosmacelf Active Member

    Joined:
    Mar 6, 2013
    Messages:
    3,399
    Location:
    San Diego
    Seems that Tesla needs to work on better integration of the camera system as processed by Mobileye and the radar. I suspect the limiting issue is their processor speed. I suspect Tesla has some underpowered CPU vainly trying to keep up the demands of AP. I've certainly noticed that as you drive faster, the system is less reliable.
     
  20. jerry33

    jerry33 S85 - VIN:P05130 - 3/2/13

    Joined:
    Mar 8, 2012
    Messages:
    12,752
    Location:
    Texas
    I believe the manual says that AP won't see stopped cars. If the car ahead of you is moving and then stops and you have AP engaged it will see that. If it's already stopped and you're just coming up on it, it will think it's part of the road. Additional sensors would been required to see a stopped car.
     

Share This Page