Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Thursday October 21 2021, @03:11AM   Printer-friendly
from the I-don't-want-to-pay-for-systems-that-don't-work dept.

The US American Automobile Association ("AAA" — national auto club) has over 60 million members. It recently ran tests on current Advanced Driver Assistance Systems (ADAS) and concluded that heavy rain caused a high rate of errors. https://www.automotivetestingtechnologyinternational.com/news/cavs/aaa-research-reveals-that-heavy-rain-often-defeats-adas-functions.html

During closed-course testing using simulated moderate to heavy rainfall, emergency brake system tests using a variety of vehicles resulted in collisions occurring 17% of the time at 25mph (40km/h) and 33% at 35mph (56km/h). Of greater concern are the findings from the lane-assist system tests, in which vehicles veered from their lanes 69% of the time.

[...] More encouraging are the results from tests looking at the effect of a dirty windshield, in which vehicles’ screens were stamped with a controlled mixture of bugs, dirt and water. AAA reports that the overall system performance was not affected.

“Vehicle safety systems rely on sensors and cameras to see road markings, other cars, pedestrians and roadway obstacles. So naturally, they are more vulnerable to environmental factors like rain,” concluded Greg Brannon, AAA’s director of automotive engineering and industry relations. “The reality is people aren’t always driving around in perfect, sunny weather so we must expand testing and take into consideration things people actually contend with in their day-to-day driving.”

I'm an AAA member, just about to ante up my $90/year. Nice to see AAA doing some research, in addition to the more mundane services like lockout and towing (both of which I've used, over the years).


Original Submission

This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
(1)
  • (Score: 0) by Anonymous Coward on Thursday October 21 2021, @03:25AM

    by Anonymous Coward on Thursday October 21 2021, @03:25AM (#1189067)

    Road-side service is great, and their maps are terrific.

    Too bad, the navigation devices displaced maps, and most insurance companies provide road-side services.

    Still, if you like road trip, AAA is worth it. I like those big bound-paper maps, and whenever you use road-side service from the insurance, they knock off points to raise your premium.

  • (Score: 0) by Anonymous Coward on Thursday October 21 2021, @03:49AM

    by Anonymous Coward on Thursday October 21 2021, @03:49AM (#1189072)
  • (Score: 3, Interesting) by Beryllium Sphere (r) on Thursday October 21 2021, @06:19AM (6 children)

    by Beryllium Sphere (r) (5062) on Thursday October 21 2021, @06:19AM (#1189097)

    It's good that somebody did, but the manufacturers or the regulators should have made sure the systems could work safely in foreseeable weather before they ever got on the road.

    I'm even afraid we're going to lose our chance at self-driving vehicles because the market perceptions will be poisoned by unsafe ones and nobody will buy them even after they're working. "The cat who has sat on a hot stove will never sit on a cold one".

    • (Score: 3, Insightful) by mhajicek on Thursday October 21 2021, @06:36AM (1 child)

      by mhajicek (51) on Thursday October 21 2021, @06:36AM (#1189102)

      What would frighten me most, would be a system that thinks it knows better than me, and slams on the brakes or swerves at a bad time. These things are still in beta, yet being legally mandated. They'd better have an off switch.

      --
      The spacelike surfaces of time foliations can have a cusp at the surface of discontinuity. - P. Hajicek
      • (Score: 2) by JoeMerchant on Thursday October 21 2021, @01:09PM

        by JoeMerchant (3937) on Thursday October 21 2021, @01:09PM (#1189174)

        Careful what you wish for... next step will be to declare any rainstorm "too hazardous for driving" and all the self-driving cars will pull over and stop until the sun comes out and the pavement dries off.

        --
        🌻🌻 [google.com]
    • (Score: 3, Informative) by Runaway1956 on Thursday October 21 2021, @06:54AM

      by Runaway1956 (2926) Subscriber Badge on Thursday October 21 2021, @06:54AM (#1189104) Journal

      AAA has always done stuff like this. https://aaafoundation.org/research/ [aaafoundation.org] I think they're a little late weighing in on this stuff, but this is what they do. I have to respect them for financing this kind of research.

    • (Score: 1, Interesting) by Anonymous Coward on Thursday October 21 2021, @12:23PM (2 children)

      by Anonymous Coward on Thursday October 21 2021, @12:23PM (#1189161)

      > I'm even afraid we're going to lose our chance at self-driving vehicles ...

      Nothing to be afraid of, according to this recent post on the private SAE (Society of Automotive Engineers) board,

      The 90% number [% accidents caused by human error] is a misinterpretation of the tri-lateral study from the 70s. Just because 10% of the accidents were mechanical failure, that doesn't mean the remaining 90% are human error. Removing the not-at-fault and environmental causes (groups A, E & G from the study) it leaves 6% human condition (alcohol, sleep, mental state) and 40% human decision.

      The real question is whether a machine will make these decisions any differently than a human. It's easy to say of course they would, it's very hard to prove they would. Also would it make other decisions resulting in an accident that a human wouldn't make. To think a machine would be perfect in all causes is hubris.

      For example, the original Google car's most frequent accident was getting rear-ended at a stop sign. After stopping it would creep forward to get a better look-out; however it would turn off the brake light during this operation, and the cars behind thought it going to keep going, not stop again. The algorithm should have kept the brake light on when going below a certain speed threshold, like a human driving with an automatic transmission. This design decision caused created new accident scenarios.

      The stop sign creep was an easy issue to analyze. It occurred frequently with few variables. While most accidents have a very low frequency rate considering the trillions of miles driven in the US every year, with an exceedingly large number of variables. Traditional statistical analysis breaks down when there's over 50 degrees of freedom. I've found good result with less than 10 degrees of freedom. Statistical methods for commercial aircraft (one-in-a-million frequency, 3,500 degrees of freedom), I've had success applying them to similar systems, but there's not enough detailed data to do that for highway accidents. I suspect autonomous vehicles will have approximately the same accident and death rate that humans have.

      In other words, self-driving will remove the need for humans to drive, but won't decrease the accident rate--it's just that the accidents will be "different". Me--I'm waiting for a self-driving car that is statistically "safer" than drivers in my demographic--don't drive impaired, past teen age hormones, some advanced driver training, often drive a stick shift (keeps me engaged) and a few other factors statistically in my favor.

      • (Score: 3, Insightful) by HiThere on Thursday October 21 2021, @01:54PM

        by HiThere (866) Subscriber Badge on Thursday October 21 2021, @01:54PM (#1189190) Journal

        The thing is, they're already better than my demographic. I tend to lose focus on driving and think about some programming proble. (I pulled my own drivers license over this issue decades ago.)

        --
        Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
      • (Score: 2) by Immerman on Thursday October 21 2021, @08:17PM

        by Immerman (3985) on Thursday October 21 2021, @08:17PM (#1189356)

        Of course, self driving could also potentially avoid a whole lot of so-called "not-at-fault" and "environmental causes" collisions. Because let's be honest - no matter what legal precedent has established, in reality there's very, very rarely such a thing as a "no fault" or "environmental" collision. Either you weren't paying attention, or you were driving too fast for the conditions.

        Person suddenly steps out into the road from an obscured location? They are at fault. Obstacle near an intersection obscures oncoming cross-traffic? Whoever put the obstacle there is at fault, and arguably the driver as well for not slowing down despite the usually-obvious fact that the obstacle is potentially obscuring cross-traffic. Driver was keeping up with traffic instead of obeying the speed limit? The street designer is at fault for relying on speed limit signs that everyone knows will be ignored rather than designing a narrower/twistier street that increases perceived danger to the point that people people will actually drive at a safe speed. (Though that last one is admittedly beyond the scope of this discussion.)

        As for environmental? That pretty much comes down to three things, two of which I already mentioned: Poor street design, someone obscuring an intersection, or weather. And if you're driving too fast for the weather, that's again your fault.

        Meanwhile, AI drivers look to be positioned to be better at seeing around obstacles (I believe Tesla's actually have sensors positioned so they can see *under* surrounding vehicles?), as well as having the potential to dramatically increasing response time, eliminating the 1/4s to 1s it takes an average driver to start braking or swerving in response to new information. Nerve signal speed means that it takes a minimum of 1/8 second to react even for a highly alert driver waiting for a signal they know is about to come, with their foot hovering over the brake. Add in brain-procesing time and even slight distraction and that number gets a lot bigger. And that translates to a significantly longer braking or swerving distance. Even if the AI can't avoid a collision altogether, it can almost certainly dramatically reduce the impact speed.

        And as far as weather is concerned - there's already AIs capable of performing controlled power-slides in race cars in heavy rain or snow, while maintaining lap times that rival all but the best professional drivers. Unlike obscured sensors, maintaining control over the vehicle in pretty much any scenario is unlikely to be a problem for a half-decent AI (in fact, I'd argue that loss of control is any but the most unusual conditions is clear evidence of a manufacturing defect). Which brings up another point - the average driver can't actually maintain control over their car when slamming on the brakes, or in unusual weather, so even if they avoid the collision they saw coming there's a fair chance they'll lose control in the process and hit something else.

  • (Score: 3, Interesting) by PiMuNu on Thursday October 21 2021, @08:35AM (5 children)

    by PiMuNu (3823) on Thursday October 21 2021, @08:35AM (#1189128)

    > More encouraging are the results from tests looking at the effect of a dirty windshield, in which vehicles’ screens
    > were stamped with a controlled mixture of bugs, dirt and water. AAA reports that the overall system performance
    > was not affected.

    What about when the cameras/LIDAR/etc are sprayed with mud?

    • (Score: 3, Interesting) by JoeMerchant on Thursday October 21 2021, @01:13PM (2 children)

      by JoeMerchant (3937) on Thursday October 21 2021, @01:13PM (#1189176)

      We have a few PoE cameras around the outside of the house, two look through very small apertures and the third has a clear dome over it. All of them lose vision when a bird / bug flies up close, but the small apertures are much more dramatic since the obstructions can get so much closer to the lens. The dome sounds like a good idea, but really sucks at night when it's trying to project IR illumination through fogged glass.

      Binocular (or more) vision would be a huge improvement in reliability, one lens randomly covered 0.01% of the time means two lenses covered simultaneously 0.000001% of the time (if the coverings are purely random...)

      --
      🌻🌻 [google.com]
      • (Score: 2) by PiMuNu on Thursday October 21 2021, @01:42PM (1 child)

        by PiMuNu (3823) on Thursday October 21 2021, @01:42PM (#1189185)

        Yeah, I'm just imagining a car driving through a muddy puddle, in which case the blockage is 100 % correlated.

        • (Score: 2) by JoeMerchant on Thursday October 21 2021, @02:02PM

          by JoeMerchant (3937) on Thursday October 21 2021, @02:02PM (#1189196)

          a car driving through a muddy puddle, in which case the blockage is 100 % correlated.

          Depends on sensor placement. High and wide should be less than 100%, but there is still a lot of correlation.

          As cheap as cameras are these days, you could have a "camera bar" of 100 sensors & lenses across the top of the windshield, just need an algorithm to determine which ones are blocked or not, and some kind of physical cleaner to keep at least a few of them working sufficiently all the time.

          --
          🌻🌻 [google.com]
    • (Score: 0) by Anonymous Coward on Thursday October 21 2021, @02:53PM

      by Anonymous Coward on Thursday October 21 2021, @02:53PM (#1189215)

      At least you don't have to debug the windshield.

    • (Score: 0) by Anonymous Coward on Thursday October 21 2021, @04:53PM

      by Anonymous Coward on Thursday October 21 2021, @04:53PM (#1189276)

      . ... sprayed with mud?

      And here is the solution for anyone that doesn't want the ADAS turned on -- "spray the sensors with mud" permanently (I'd use some black tape). Even if there is no "ADAS off" switch, this should blind it enough that it will shut down. There are probably similar fixes for the radar units that are also part of the ADAS sensor suite.

      Around here there is an annual inspection where warning lights have to all be off, so I'd take off the tape for the inspection, then put it back on.

  • (Score: 0) by Anonymous Coward on Thursday October 21 2021, @08:14PM (1 child)

    by Anonymous Coward on Thursday October 21 2021, @08:14PM (#1189354)

    Attorneys will insist ADAS be disabled in bad weather to avoid liability.

    The cruise control on my wife's Lexus slows the car down when there is an object in front going slower that the cruise speed setting. When heavy rain, snow or dirt limits the range of the sensors, the car turns off cruise control.

    • (Score: 0) by Anonymous Coward on Thursday October 21 2021, @09:00PM

      by Anonymous Coward on Thursday October 21 2021, @09:00PM (#1189384)

      The lawyers might eventually do that...but they don't seem to have done it yet. Case in point is that AAA was able to run their experiment with heavy rain (simulated with sprayers) at the camera locations, and the ADAS systems kept working.

(1)