By on June 5, 2020

When the United States began passing legislation allowing automakers to begin testing self-driving vehicles on public roads, it was framed almost entirely as a safety issue. Proponents claimed that the only way to eliminate roadway fatalities was to take the human brain out of the equation and let cars drive themselves. Having enacted a similar no-thinking policy themselves, legislators agreed — pleased to have ensured a death-free future on little more than empty corporate promises.

At the time, we were still complaining about the unreliable nature of advanced driving aids, and how such systems seem custom-made to dull your reflexes behind the wheel. There was a sense that, if everything went perfectly, maybe autonomous vehicles (AVs) could reduce accidents by previously unheard of levels. That feeling didn’t last particularly long here at TTAC and, by 2018, we started noticing we weren’t alone.

The Insurance Institute for Highway Safety (IIHS) grew increasingly critical of AVs starting a couple of years ago. On Thursday, it released a report claiming the idea of a no-crash future spurred by automation is a fantasy. Instead, the IIHS says cutting-edge technology will likely struggle to stop just a third of all accidents. 

From IIHS:

Conventional thinking has it that self-driving vehicles could one day make crashes a thing of the past. The reality is not that simple. According to a national survey of police-reported crashes, driver error is the final failure in the chain of events leading to more than 9 out of 10 crashes.

But the Institute’s analysis suggests that only about a third of those crashes were the result of mistakes that automated vehicles would be expected to avoid simply because they have more accurate perception than human drivers and aren’t vulnerable to incapacitation. To avoid the other two-thirds, they would need to be specifically programmed to prioritize safety over speed and convenience.

As stupid as it is to say AVs “aren’t vulnerable to incapacitation,” when we already know advanced driving aids relying on similar hardware fail anytime a camera gets dirty or a sensor gets bumped out of position, IIHS is on the right track.

Driving is an extremely involved task; the mind makes calculations on the fly, helping you process information and enact a usable plan in milliseconds. Programming a computer to remember one thing is easy, and it will accomplish that task with such effectiveness and repeatability as to embarrass us, but they come up short when you chuck in trillions of variables, which driving does. Here, the mushy and easily distracted human brain still reigns supreme.

“Building self-driving cars that drive as well as people do is a big challenge in itself,” said IIHS Research Scientist Alexandra Mueller, lead author of the study. “But they’d actually need to be better than that to deliver on the promises we’ve all heard.”

The automotive industry may have dug its own grave on this issue. In order to speed up development, automakers needed to test vehicles in real-world scenarios. To get there, however, they first had to promise the moon in regards to safety… and regulators were willing to believe.

IIHS refuted the bold safety claims with a study (more of a thought experiment, really). Researchers imagined a future in which all vehicles on the road are self-driving and totally effective at avoiding accidents created by an inattentive or incapacitated diver. Those types of accidents account for about 34 percent of the serious accidents we see every year. The rest stem from a driver either planning incorrectly (e.g. mistakenly thinking they can make a last minute exit) or executing poorly (e.g. failing to safely make an exit when they could have) and can only be stopped if total control is stripped from the driver and the car never once breaks the rules of the road.

Many experts note that simply allowing non-autonomous vehicles to share the road with self-driving cars would be a huge problem in itself. Even if AVs are perfectly networked together, operating as one gigantic system, human drivers would serve as difficult-to-predict anomalies on the road. Meanwhile, the IIHS’ own assessment of the present-day safety systems that are supposed to foreshadow true self driving shows them to be saddled with problems.

“Our analysis shows that it will be crucial for designers to prioritize safety over rider preferences if autonomous vehicles are to live up to their promise to be safer than human drivers,” Mueller said.

And there’s the big problem — and the reason why automakers probably won’t pursue this much further. The industry wants to sell vehicular autonomy as a feature, not as standard equipment baked into every unit. We also don’t think it was ever all that interested in promoting safety. Otherwise, we wouldn’t see distracting screens creeping into every corner of a car’s cabin and advanced driving aids with glaring shortcomings.

[Image: IIHS]


Get the latest TTAC e-Newsletter!

28 Comments on “IIHS Denounces Concept of Total Safety From Autonomous Cars...”

  • avatar
    SCE to AUX

    I agree with them, but they’re effecting ruling out the possibility of SAE Level 4 or 5 autonomy, which require no driver inputs or attention.

  • avatar

    “on similar hardware fail anytime a camera gets dirty or a sensor gets bumped out of position”

    Not necessarily with newer tech. The dirtiness can be solved with a cleaning mechanism, and sensors that don’t rely on precise alignment are certainly doable. In fact, one of my vendors sent an update that makes one of their sensors self-calibrating. On my end, I have friends in the body shop industry that complain about aligning sensors and from my own experience I hate, hate, hate calibrating the things myself. In fact, I’m fighting with calibration on something right now. There are newer techniques like road mapping with ground-penetrating radar systems to let a vehicle know where it is. Flir gives good vision in bad visibility systems.

    More comments later…

  • avatar

    All you need to do is watch Westworld season 3 to see AI cars repeatedly remotely hacked, taking their passenger on a ride against their will.

  • avatar

    I now await a juvenile, petulant tweet from Sir Elon.
    AVs were an irresponsible unrealistic pipe dream from the beginning. It boggles the mind that the likes of Ford and GM are dumping billions down this rat hole.

    • 0 avatar

      “irresponsible unrealistic pipe dream from the beginning.”

      I think the timeline and the particular tech they are using now is a dead-end, but in a decade (or more) provided new AI technologies that are in the process of being developed along with new types of sensor technology also under development, we could have AV systems that are better than humans. Still won’t prevent you from dying in all situations, but they can be better than a human.

      • 0 avatar
        Art Vandelay

        The problem is he is charging people for it now…not a decade from now. Regardless of how you feel about AV’s, what he is doing is scummy at best.

        • 0 avatar

          They are getting some features, but not full fsd. You don’t have to buy it. I guarantee you I’ll skip right by it on the order form (unless I get it for competitive eval. but I doubt I will). Lots of good aftermarket upgrades for the 3 can be bought for that money.

  • avatar

    We should create a driver militia and movement against AV proponents.

  • avatar

    Imagine the one third less driving deaths utopia outlined by the IIHS…while overall deaths will be lower it will be essentially a death lottery on the road…molecules indiscriminately bounding from A to B with a small probability of a catastrophic intersection.

    Now I know all of us here are great drivers who cherish and respect the great wide open…Our death rate could conceivably be well below even this projected average.

    I think that marginal improvements in death rate while forfeiting all control of our individual fate is rather upsetting and frankly nauseating.

    • 0 avatar
      Art Vandelay

      If we want to lower the death rates, increase the training requirements for obtaining a license and move to a European model of enforcement, as they actually have lower death rates. This is a proven solution with decades of data to back it up versus some pie in the sky dreams of a car driving itself.

    • 0 avatar

      We could slash the death rate by a third, without even changing what we drive, but we don’t want it.

      Roundabouts were soundly rejected by US drivers since we don’t know how to merge, refuse to try.

      We’d rather sit hours a day at lights, hoping not to get rear ended, then pull out and get T-boned.

  • avatar

    The problem with self driving cars is that no one knows how to build one. No one knows if the technologies being pursued today are going to be good enough and no one can predict when they will reach a point of being good enough. 6 months, 6 years, 60 years, it’s really just a guess. No one can even agree on how to approach self driving or what the requirements will be. I think a car needs to be completely autonomous, not connected, and to be able to think and problem solve on its own with multiple and redundant ways of interpreting its environment, just like a person. Unfortunately there are no computers anywhere that approach those capabilities, not even close. If we had a computer able to do those things it would have far more interesting applications than merely relieving you of the tedium of driving. You’d have robots in space repairing satellites and exploring the planets. Think of the possibilities.

    • 0 avatar

      “no one knows if the technologies being pursued today are going to be good enough”: Actually, we know they (like the types of AI in common use) won’t be good enough and some of us are working on developing new technology.

      “6 months, 6 years, 60 years, it’s really just a guess.” That’s true, but I can guarantee it won’t be 6 months. 6 years isn’t likely either. We’re sitting around conference tables debating various white papers at this point.

      “just like a person.”: exactly. things like intuition are important.

      “Unfortunately there are no computers anywhere that approach those capabilities, not even close.”: Well… kind-of sort-of. We’re building with FPGAs now. The actual hardware could be built now, it’s how the hardware is arranged that is the problem and much more research is needed. A lot.

      “it would have far more interesting applications than merely relieving you of the tedium of driving”: Oh, absolutely! In fact, to get funding, some of us are having to push the medical uses. In fact, the “other uses” are a lot safer to develop since they’re not piloting around 2-ton vehicles. One approach is to develop a personal assistant that could perform household chores. Then as technology improves, give them the capability of driving. There are tv series that depicts what could be developed. The series deals with the benefits and the problems. Granted, this sort of tech is a ways out, but that is the direction some research is headed.

      This series was on Amazon at one point:

  • avatar

    How is a significantly reduced death rate a bad thing? Nothing will bring it down to zero, but IIHS is claiming this one thing will get us a third of the way there. That’s pretty impressive.

    • 0 avatar

      Not even Subaru Safety Sense can help some people from wrecking…

  • avatar

    As a bit of SME on the subject of AI, I have to say that this is painfully obvious. It has little to do with dirty sensors, etc., but rather fundamental problems of AI (or lack of thereof). Even with some dramatic breakthroughs it will not happen for 30-40 years at best.

    Most likely it will never happen at all as it will simply become unnecessary. The trends, even before the current SNAFU, have been towards reduction of freedom, including the freedom of movement. The roads will become “smart”, making fully autonomous driving practically impossible, even if not illegal.

    That’s why at home we try to avoid “smart”-anything and drive older manual cars. Enjoy it while you still can. As Kingsley Amis would say:

    So bloody good luck to you, mate,
    That you weren’t born too late
    For at least a chance of happiness,
    Before unchangeable crappiness
    Spreads all over the land.

    • 0 avatar

      @slitno: “Even with some dramatic breakthroughs it will not happen for 30-40 years at best.”

      I’d put it at less than 20 years. Mathematical AI has some great uses, but I think more direct biological modeled systems are the future. Moving from mathematical’s elemental neurons to a model that resembles the neocortexes pyramidal cells. I’ve started the process to go back to school for neuroscience. I have interests in grid cell theory and the mechanics of human spatial reasoning. I’m also very interested in artificial intuiton. A key ingredient that seems to be missing from the current AV systems. As far as advanced sensor technology, I’ve performed some interesting experments in using photogrammety to reconstruct 3D object representations from reflections in the sides of vehicles. The idea being to eventually use bits and pieces of reflections to see around corners and spot hazards and potential collisions that a human would never be able to see. Part of my background is in aviation collision collision avoidance systems and I have a couple of systems out there. Hardware background includes FPGA experience and I’m using that to develop custom circuitry.

      • 0 avatar

        I’ll remind you that Elon, as well as others, have boasted again and again that FSD is just around the corner. Thousands of gallons o fa virtual ink have been used to talk about neural networks and artificial intelligence and data collection, and other pseudo-intellectual-oid blather to no avail. They can say whatever they want. They can say anything is coming in 3 months or 6 months or next year because that buys them a little time. But what they can do is make ANY meaningful progress. Tesla is no closer today to achieving real progress toward FSD than they were 5 years ago. Their cara have NONE of the hardware that will eventually be needed, not even close. Predicting a technological breakthrough 5 years out Ian like predicting the weather 7 days out – it just crashes any bugs we done.
        The people working on this problem don’t even know what they don’t know. You can’t work on a problem if you’re not aware of it.
        Notice how Elon never talks about FSD anymore? People are tired of hearing about it, it’s just noise.

        • 0 avatar

          “The people working on this problem don’t even know what they don’t know. You can’t work on a problem if you’re not aware of it.”

          So true. So true. But, some of us smaller guys are hoping we’re headed in the right direction and will have nice patent portfolios waiting for them when they hit the end of their rat holes :^) .

          I’m not sure I’d call what we need a technological breakthrough. There are lots of neuroscience white papers being produced that we’re discussing and using that research to guide us. It’s a lot of small tasks and little bits of incremental research that will eventually get us there. It won’t be a single big breakthrough. It’s going to be a long haul that’s for sure.

          One cool side product of all of this research is that we will gain a great deal of knowledge about the human brain and more insight into various brain disorders.

  • avatar

    If the nag screen made you agree to waive your right to sue the manufacturer in case of an accident before you could enable autonomous driving, would you do it?

  • avatar

    How 2.5L engine can generates only 225 hp? 2.0L Ecoboost in my car delivers 245 hp. Same with Audi. No MT option on Japanese car? When did that happen?

  • avatar
    SCE to AUX

    Mazda is the class nerd who gets the best grades and wins the spelling bee, but all the girls hang out with the farmer guy who drives a pickup.

    One night last month I walked through some local car lots, and laughed out loud at the Mazda prices. $30k+ for a tiny CX30?
    Sorry, but a Rogue Sport can be had for thousands less, and the differences are discernable to very few customers.

  • avatar

    A non-zero amount of this too is IIHS trying to position itself as an expert / gatekeeper of autonomous safety and trying to find ways to help ensure the auto-insurance industry still exists after autonomy begins cutting down on insurance need.

    If autonomy works as promised, why would you or I have insurance? The automaker would insure the vehicle. And why would GM talk to any current mainstream insurance agency? They would have better ways to insure themselves.

    Best to ensure that they are seen as qualified judges of autonomous tech, and to ensure a narrative where autonomy will effectively never “get there”, to keep their backers in business.

  • avatar
    schmitt trigger


    Very Interesting comment of yours.

    And to make absolutely sure that it was you and not your grand kid waiving it, have it “signed” bio-metrically, with both a fingerprint and face recognition.

  • avatar

    This article just came out and illustrates how fast the AI hardware technology is progressing. This is not the old school AI used in current AV technology (afaik):

Read all comments

Back to TopLeave a Reply

You must be logged in to post a comment.

Recent Comments

  • Steve: Borla.
  • Funky D: “Ford says it’s all part of the Ford+ plan, a strategy described by the company as an effort to...
  • TDIGuy: Demand should be driven by consumers, not business or politics. Business should do what it can to create...
  • Mike Beranek: That two-tone Ramcharger in the ad looks better than ANY modern large SUV. Love the stacked lights and...
  • SaulTigh: Always dreamed of having a V12 as a kid. If you wanted smoothness and great power when I was young, a V12...

New Car Research

Get a Free Dealer Quote

Who We Are

  • Adam Tonge
  • Bozi Tatarevic
  • Corey Lewis
  • Jo Borras
  • Mark Baruth
  • Ronnie Schreiber