By on May 31, 2013

I’m driven

Everybody seems to be on the bandwagon for self-driving cars, everybody except the NHTSA. In  new guidelines,  the NHTSA urges states to allow use of self-driving cars “only for testing and requiring safeguards to ensure they can be taken over by a driver in the case of malfunction,” the Detroit News writes.

“We believe there are a number of technological issues as well as human performance issues that must be addressed before self-driving vehicles can be made widely available,” NHTSA said. “Self-driving vehicle technology is not yet at the stage of sophistication or demonstrated safety capability that it should be authorized for use by members of the public for general driving purposes.”

NHTSA wants test drivers to get special licenses. If a state was to allow use of self-driving vehicles by the public, the agency urged them to require a special license and to mandate that person sit in the driver’s seat, ready to take over.

Get the latest TTAC e-Newsletter!

53 Comments on “NHTSA Does Not Want Self-Driving Cars To Drive By Themselves...”


  • avatar
    CoastieLenn

    I’m pretty sure that the impulse to want to jump in and take over “in case of malfunction” will be an innate instinct to any/all users of SDC’s for the first decade or so. NHTSA doesn’t have to mandate it.

  • avatar
    Detroit-Iron

    If a self driving car can’t let me sleep while driving across Nebraska or get me home from the bar after last call then what is the point?

  • avatar

    Self driving cars rely 100% on sensors, as does a human driver with his 5 senses. Thing is, we have INTELLIGENCE, INTUITION and PREDICTION. If one our senses fails, we can readily make up for it. We can also compensate for other drivers if their senses fail if our reflexes are fast enough.

    Until artificial intelligence is a reality, I don’t trust self driving cars, nor should the legislature.

    And considering ticketing and arresting drivers is a money making industry, I doubt they’ll ever allow self driving cars to become the norm.

    If My woman and I get drunk at a club, can we hop in the backseat and trust the car to get us home on a rainy night?

    If I’m late for a meeting and there’s no parking, can the car circle the block till it finds one? And then come pick me up when I press the key fob?

    Can I trust a car to swerve to avoid a kid who runs out into the street? What if there’s ice on the ground?

    What if I feel like texting or SLEEPING instead of driving? Can the car handle a long trip?

    Can the car understand road work and avoid potholes?

    What if prostitutes rush my car because its been a slow night? Will the sensors know which is a legitimate woman and which is a tranny?

    These are the questions that need to be answered.

    In the film iRobot, they never did explain how humans could coexist with robots able to do everything they could do: better, faster and without tiring. If that kind of AI existed cabbies, truckers, pilots, etc would be out of a job.

    Imagine buying a car with NO STEERING WHEEL OR TRANSMISSION SWITCH.

    • 0 avatar
      NMGOM

      Good comment, BigTrucks…

      It gave me my morning laugh. And I agree heartily.

      SDC’s may first have to diagnose weather, traffic, congestion, and lighting conditions to determine if they are suitable for SDC mode. It could be that the only really safe SDC’s will be as “Detroit-Iron” says: driving across Nebraska (or Kansas, or Arizona, …you pick) in the middle a sunny day on well marked roads with no one else around!

      ——————–

      • 0 avatar
        gmichaelj

        These comments remind me of an English prof I had (1986) who was sure a computer could never fix all my grammar and spelling mistakes. The videos I have seen show that Google has done an amazing job getting their cars to handle all kinds of situations. At some point I see the self driving car being tested against average drivers in various situations and out-scouring them: rainy, icy, kid running out in front of, etc. I’ll bet in the not too distant future it’ll even be able out drive even the above-average TTAC drivers too. And get you back to Lincoln refreshed.

        • 0 avatar
          mcs

          @gmichaelj

          I’m actually involved in robotics research – including autonomous vehicles. I have a background in aviation collision avoidance systems – including one ground collision avoidance system.

          The Google cars are rudimentary and there are a tremendous number of situations where they will fail. All Google has is a sophisticated cruise control, they are not even close to being autonomous vehicles.

          And bye the whey, they’re are a phew limitations to spill Czech systems.

          • 0 avatar
            28-Cars-Later

            I envy/admire you, friend. I used to work with the Mckesson Robot-RX and regret not getting an EE or ME degree instead of the generic IS one I have. Granted I’m still pretty happy with the way things turned out but I find robotics to be fascinating. Saw Ironman 3 last night and aside from the overall ridiculousness of the film my geek side was very satisfied.

          • 0 avatar
            jhefner

            My whole problem with “automated control until it gets in trouble, then manual intervention” scenerio is what happened to Air France Flight 447:

            http://en.wikipedia.org/wiki/Air_France_Flight_447

            It was the middle of the night; the pilots were probably half asleep when the Airbus flew itself into a thunderstorm. It appears the air speed indicators were not reading correctly probably due to ice; so the pilots had to take over a plane in major turburlance, in the dark, without a good airspeed readout; and most likely did a high speed stall into the ocean as a result.

            Let’s take the previous poster’s example of driving across Nebraska in the middle of the night; when suddenly a not-uncommon frontal system brings high winds and freezing rain. The system decides it cannot handle it, alerts the driver, and shuts down; the half-asleep or asleep driver now has minutes or even seconds to figure out the situation unfolding at speed, in the dark, and take corrective action. Good luck….

          • 0 avatar
            Landcrusher

            Hef,
            First, your scenario is hyperbolic. A car system would not be designed such that it gave up at speed. It would stop the car first.

            The airbus is designed to be run by systems engineers more than pilots and many US pilots are not fans. European pilots generally have more systems training than stick and rudder, so airbus went that way with their fly by wire. NIMBYs and statists and elitists and crooked pols and the FAA and Luddites are in an unplanned conspiracy to kill general aviation in this country and lead us down the path to create more bad airline pilots every day.

            The airbus pilots needed to push the stick forward, just like the pilots of the Colgan air flight. Human error in both cases, though the latter was aggravated by bad system design and lack of training.

          • 0 avatar
            jhefner

            Landcrusher;

            Perhaps, and I agree with everything else you say. But your points make the arguement for me:

            Airbus’ system of taking control away from pilots and putting it in computer controls and engineers far away is causing pilots to both lose their situational awareness, as well as some of their piloting skills. In the Cologne instance; bad software design was causing the system to override the pilot’s correct actions, and caused the plane to crash. In the case of the Air France crash; pilots did not have situational awareness because they were not hand flying the plane; and with a lack of correct input due to the airspeed indicators not reading correctly and not being able to have the benefit of instruments nor see outside what was going on, caused a high speed stall and a crash.

            You seem to imply that the Airbus way of doing things is creating bad pilots. Well, the SDC is really the same thing: a combination of autopilot mixed with centeral remote control. The final result will also be the same- a generation of bad drivers who will not know what to do when the system lets them down.

            The “miracle on the Hudson” proves that an excellent pilot (who flew gliders when not flying Airbuses) can still do the right thing when everything falls apart. But how many regular drivers will continue to hone their driving skills once the SDC becomes available?

          • 0 avatar
            Landcrusher

            The reason they are bad pilots is that they never really were qualified pilots. This is important because they likely never really developed proper stick and rudder, and had they done so, they may still have been burned by the terrible airbus systems. Air France hires non pilots because the european states destroyed general aviation in the name of egalite’ and bureaucracy. I get nervous on an airbus and have to rationalize that I am still safer than in a car. Most drivers never really develop the proper equivalent to good stick and rudder – ever.
            One more thing, you can stop a car, but it’s not a good thing to do with a plane. Potentially, an unlicensed user only needs to be able to get a car off the road using the controls or wait with hazards on for help. A licensed driver with no practice isn’t necessarily far worse than average, but this isn’t so with pilots. There has been a backlash in aviation with some companies demanding more hand flying by crews.
            I would suggest that drivers licenses are too easy to get now, and we should not let up until the robots are known to be far more reliable than average drivers.

        • 0 avatar

          “These comments remind me of an English prof I had (1986) who was sure a computer could never fix all my grammar and spelling mistakes. The videos I have SEEM TO show that Google has done an amazing job”

        • 0 avatar
          Lynchenstein

          We’ve had ABS for years. This was the first of the electronically augmented inputs I’m aware of and the tech has allowed drivers to avoid countless crashes. We have lane detection, automatic parking, and so on.

          It’s just a matter of time, and I think it’ll be great.

        • 0 avatar
          wmba

          “These comments remind me of an English prof I had (1986) who was sure a computer could never fix all my grammar and spelling mistakes.”

          And he was right.

          “At some point I see the self driving car being tested against average drivers in various situations and out-scouring them: …”

          Out-scouring them? Yes indeed. Ajax to the rescue but SOS Pads have their place! Sorry, couldn’t resist.

          I like Big Truck Series’ points which are quite germane to a truly functional system. Road repair and pothole-avoidance smarts? That will be Version 31.2. Also, mcs agrees below, and he knows his stuff.

    • 0 avatar
      mcs

      Good comment BigTruck!

      @Can I trust a car to swerve to avoid a kid who runs out into the street? What if there’s ice on the ground?

      It’s not so much the kid running out into the street that’s the issue, the ability to infer that a ball rolling out into the street could be followed by a kid while simultaneously analyzing thousands of other objects and potential problems that is the real artificial intelligence challenge.

      The problem with autos on streets is that sensors alone aren’t enough. By the time a cars sensors detect a potential collision, it may not leave enough time to avert the collision. That’s why humans are needed, because we have the ability to anticipate potential problems and preemptively react to a potential situation. In other words, humans know when to be cautious. Robots still don’t have that ability yet – but we’re working on it.

      • 0 avatar
        Type57SC

        I’ve got more confidence in a V2X car knowing the street conditions than me, and being able to take avoidance countermeasures much faster and more reliable, but I worry about how to program in things like the instinct to ram a parked car to avoid hitting a pedestrian or swering so severely to avoid a life threatening truck even though you are then doomed to spinning uncontrollably across another lane and hoping for the best.

      • 0 avatar

        MCS

        That “ball rolls into a street – will be followed by kid” rule has happened to me 3 times since I took driver’s ed. 2/3 were followed by kids. Is a computer ready to look between cars and detect a child? I think not!

        • 0 avatar
          krhodes1

          On the other hand, the computer actually WILL BE LOOKING ALL THE TIME. It won’t be yapping on the phone, applying makeup, fiddling with the stereo, yelling at the kids in the back seat, be drunk or tired, or any of the other myriad things we do while nominally “driving”. Should the driver be the ultimate bearer of responsibility for what happens? You bet! But I have no doubt that the computer will be a better driver most of the time.

          I also will say it is not all or nothing. Just like you don’t generally use cruise control while traveling down heavily trafficked city streets, you won’t use self-driving mode there either. On the wide open highway? Sure, and I can’t wait, as there is nothing more boring than cruising at 75mph on the interstate hour after hour.

          • 0 avatar
            Brian P

            A good many human drivers don’t think or predict very much while driving, either – and a good many of them are being sidetracked by their smartphone or passengers.

            I tend to suspect that the self-driving cars will do better in the real world than poor human drivers but not as well as the best human drivers, and it’s probably a crap-shoot whether they’ll match the average human driver.

            I can see the self-driving cars being extremely diligent rule-followers, never exceeding (too-low) posted speed limits and never running through a red light or stop sign, thus avoiding collisions that are their own fault rather well, but not doing so well in avoiding somebody else’s collision. A human driver who is paying attention will know when they ought to run that stale yellow or early red rather than getting creamed by the unable-to-stop 18-wheeler behind them …

          • 0 avatar
            mcs

            This is some of the imaging tech we’re looking at:

            http://web.media.mit.edu/~raskar/trillionfps/

            This is something we might get as a driver’s aid prior to autonomous vehicles and one of the pieces of imaging tech we’ll need for real autonomous vehicles. When we get this working it will be amazing.

      • 0 avatar
        rollingScienceProject

        Since this was exactly my Masters subject 10 years ago, I think I can contribute some insights. I admit I may be fuzxy on thenrecent developments, but here goes.

        1) humans are very ill equipped for this task. The usual processing and acting lag (reaction time) is 2 whole seconds for the average alert person. Worse:most of the time, people are not on their best alertness level so this number goes up. Moreover, they have no direct sense of speed or inertia, making most emergency measures a “rule of thumb” affair, learned during emergency situations that are not repeatedly encountered.

        Enough wetware bashing, on to the cool stuff.

        2) the theory splits this problem in 3 levels of control: a)the car controls it’s speed to keep rolling with the traffic. This is currently sold under the Distronic Plus and other names, quite cheaply I might add.
        b) car-following: just like in the movies, you tell the car to follow some other vehicle and relax. Still quite desirable on the highway and in the traffic, and depends on having a sane driver to tag along with and do the grunt work of finding a safe path. This may come quite soon, since the necessary algorithms and sensors are avaliable.
        c) completely autonomous control. This means plotting a course (thank you Garmin), staying on the road, avoiding obstacles… the works. This may take a few years yet, especially since ambulance-chasers will hold those to such a high standard that it would have to pretty much be omniscient to forestall lawsuits after accidents which would not be preventable by any human (ex: avoiding a car coming out of a blind corner in the wrong lane at a ludicrous speed)

        10 years ago, doing the first level required expensive radar systems and custom computers. Radars are now so cheap you can buy them aftermarket, and the tablet I am currently using probably possesses anough raw power to do it, and anyway I just have to get a midlevel Benz or a 2007 Chrysler 300c to get one ready-made.
        The third level is the object of a DARPA contest called the Urban Challenge. In 2007, the winners navigated a road course which included pedestrians and the other contestants at 14mph. I can assure you that 2007 is a long way gone, even if I am not currently involved in the field. Unfortunately, the contest was not repeated after 2007 so public details are pretty scarce.

    • 0 avatar
      Landcrusher

      Truck,
      I am confused. Are you backing off from your prediction this will never work? Is it now that it will never meet your standards? It’s just a matter of time.

      • 0 avatar
        NMGOM

        Landcrusher….

        Many of us feel some ambivalence here, so don’t be surprised if apparently conflicting views emerge.

        Let me try to clarify:
        1) As I stated in another post on this topic here at TTAC, there are 5 reasons why SDC’s would be a good thing, IDEALLY.
        2) There are several serious “flies in the ointment” along the way that don’t make “ideal” easily obtainable, and we are only in the error-prone inception phase;
        3) The Google effort is primitive at best, and subject to near perfect conditions, but it’s a good start;
        4) My profession (beyond chemistry) was computerized image analysis, involving, among other things, pattern recognition. I can guarantee that this WILL NOT be any easy task, and it will involve many visual and thermal (IR) cameras, GPS locators, acoustical sensors, radar systems, and one massive super computer onboard that can provide a VERY good simulation of human experience AND anticipation in its memory banks (example: Big Truck saying ball-in-street = kid-to-follow with 67% probability).

        The problem is cost. Will such a car be cheap? No. I won’t put a real number on such a creation, the size of a Ford Fusion, but a SWAG might be $100K (with current technology). All for the privilege of sleeping, not driving, during a boring morning commute.

        —————-

        • 0 avatar
          Landcrusher

          NMGOM,
          I am happy to have a serious discussion, and there is plenty of room for thoughtful and educated doubts. Most of conversation here is neither though. Much of the benefit can be had from only having self driving cruise for where we use cruise now. I drive a car with adaptive cruise and its spectacular. Tests I have read about have also been very promising. Today’s aviation auto pilots are much more advanced and inexpensive than they used to be. Mostly because of technology and economy of scale borrowed from cars. The scale of the car business is huge. If you make a 100,000 dollar auto drive system today, it wil be under 10,000 in no time. A highway only system like the BMW or Volvo systems that can go 75 should be very doable for a few thousand soon. I would buy it. Improvement is just a matter of time. I believe Volvo already has tech to ID pedestrians and single them out for attention. Perhaps we come up with a marking system for boulevards that makes it easier. Whatever. I will join with the matter of time crowd over the never crowd on this one and I am a big skeptic usually.

    • 0 avatar
      dude500

      The problem I have with lists like these, is that most average drivers wouldn’t be able to meet this criteria either. I don’t think most drivers would be able to avoid a kid running into the street, so yes I do think a computer is better. Most drivers suck at following simple road rules, will still drive when sleepy, drive when drunk, etc. I don’t trust most drivers in the rain, in the snow, during rush hour, and at night.

      So yes, the computer *won’t do any worse* than than the average driver today. It will never be perfect, but it will be better.

      • 0 avatar
        mcs

        When the average inattentive driver screws up, the driver gets sued. When an autonomous vehicle screws up, every member of the Trial Lawyers Association has a simultaneous orgasm and immediately order up tv ads with the “have you been hit recently by an autonomous vehicle” line.

    • 0 avatar
      porschespeed

      “I know nothing about real bigtruckseriesreviw” you are from the 1950 time warp, right? Or are you just being a poseur for the 99th time I’ve seen…

      CAT actually sells and deploys fully automated big trucks to mine owners. ‘Natch, you know sweet FA about trucks. Let alone ‘big’ ones.

      Do enlighten us all about what the issues are with those driverless vehicles and why the customers aren’t clamoring for them, despite the fact they are because they do work.

      Please, graduate middle-school before you start commenting on state-of-the-shelf tech. You just make yourself look even more illiterate.

  • avatar
    Volt 230

    These robocars proponents must be the offspring of those who predicted we’d have moon bases by now! sad thing is that some so-called “car guys” are all giddy about this sci-fi BS

  • avatar
    Landcrusher

    First off, it’s foolish to say it will never happen. Second, it’s ridiculous to sit here with no standards and nit pick. Third, it’s useless to make the standards higher than the average driver (at least at first) or to require the initial systems be totally autonomous and trustworthy when even if they were you all know we would not allow them to be used that way for years.

    Flying cars and moon bases are all doable. They just don’t have a good ROI. Part of the reason for the poor ROI is not the tech, it’s the temperament. If every downtown had a STOL port, commuting by flying car would work for lots of people, but cities always rip up their airports. We don’t have flying cars because not enough of us really want one, and many of us don’t want anyone to have one.

    OTOH, there are great potential benefits to self driving cars and the potential size of the market is huge. Getting rid of the mass transit mafia would be worth it alone. Short hop airline flights can be eliminated making more space for longer flights. All good, and it won’t happen over night.

  • avatar
    gmichaelj

    Sure, we’ll never be able to come up with a system that can anticipate danger or changing conditions as well as you guys can. Computers being able to recognize patterns and respond is just sci-fi.

    I suppose all you guys can shift faster than a dual clutch transmission and brake better than an ABS system as well.

    • 0 avatar
      mcs

      “Sure, we’ll never be able to come up with a system that can anticipate danger or changing conditions as well as you guys can.”

      I’m busting my butt right now to build one. We just have a ways to go and it seems that people are a bit over optimistic about the effort and the scale of technology to do it. We’re also looking for a sort of silver bullet to speed things along.

      The idea is to evolve the product gradually. Start out with household chores, then move up to an industrial robot. Not like in the past, one that is able to replace a human directly on an assembly line using the same process and tools as the human.

      Next, maybe the domestic robot could walk to a store, shop for ingredients, then return with the purchases. Finally it would have the ability to climb in a car and drive it. This isn’t really my idea – the credit goes to DARPA that for a challenge last year wanted a robot that could climb into an unmodified vehicle and drive it. Makes sense – why tie up some hyper-expensive technology into a vehicle when it could be utilized elsewhere.

      • 0 avatar
        Brian P

        We’re seeing the beginnings of collision-avoidance systems in production today. These will become more and more advanced – thus helping the driver out by stopping the driver from doing certain things that would blatantly result in a collision (e.g. driving into an intersection disregarding other traffic that is coming, etc).

        What worries me is that they won’t impose a sufficient penalty on the driver for making a grievous error. The blind-spot detection systems now just give a meek flash or beep. If the driver makes a grievous error, like trying to pull ahead into traffic and thus requiring the computer to intervene, that driver needs to be penalized for it. Screw up once … beep (it happens). Screw up twice … BUZZZZ and the radio goes off and bluetooth goes off and the driver needs to press an “acknowledge grievous error” button. Screw up three times, car shuts down (safely) and requests a new driver. Or something like that.

      • 0 avatar
        gmichaelj

        Thanks, that seems reasonable. If I “read” you correctly it is far off – 10+ years???

    • 0 avatar
      NMGOM

      gmichaelj….

      You mentioned: “Computers being able to recognize patterns and respond is just sci-fi.”

      Respectfully, this has not been the case for at least 25 years. Pattern recognition, object identification, and landscape analysis by military systems would astound you. I know. It was my profession.

      The issue with SDC’s is the % success we would demand ultimately, and the cost of such systems commercially.

      ————–

      • 0 avatar
        gmichaelj

        Ok, so do you think such a system will have to be near perfect before it is in use?

        Given all the bad/marginal drivers I see, I wonder how good self driving cars will have to be to give us a significant improvement in traffic flow and fewer accidents.

  • avatar
    PeteRR

    Interim step: Outfit cars with sensors and cameras so as to allow operating it by remote control like the Predator drone. Then sign up millions of un/underemployed 3rd world types to drive our cars from a control center in Bangladesh.

    • 0 avatar
      dude500

      +1 this is actually a great idea. Wouldn’t be any worse than riding a taxi.

    • 0 avatar
      wmba

      Great idea! But how are they going to scrape the ice and road salt off the sensors on a cold morning? You mean they expect ME to do it? Moi?

      Surely you jest? Why do think I laid out the big bucks for this car, anyway? Harumph.

  • avatar
    raph

    Heh, I can’t wait for the Nissan Mr. Roboto GT-R 9000 and its preternatural lap times and the great forum debate over which is better in much the same way people argue over human vs. self shifting transmissions.

  • avatar
    brid1970

    Big Brother go ahead and audit me if you like….but please don’t take away the sheer fun of driving myself.

    • 0 avatar
      Lie2me

      +1 Thank you for voicing exactly what I was thinking as I read through all these comments. What about those who really enjoy the complete experience of driving? Will we be forced into these SDCs or will there be the option of not “driving” an SDC? Would it be similar to people who prefer a stick shift over an automatic? Some, including me, think that our cars are already making too many decisions for us as it is. It has been a slow gradual progression towards SDCs since the first automatic transmission was introduced and it has continued ever since to the point where a technologically well optioned car practically reduces the driving experience to passive input. Do we want cars to do it all, do we even have a choice?


Back to TopLeave a Reply

You must be logged in to post a comment.

Subscribe without commenting

Recent Comments

New Car Research

Get a Free Dealer Quote

Staff

  • Authors

  • Brendan McAleer, Canada
  • Marcelo De Vasconcellos, Brazil
  • Matthias Gasnier, Australia
  • Tycho de Feyter, China
  • W. Christian 'Mental' Ward, Abu Dhabi
  • Mark Stevenson, Canada
  • Faisal Ali Khan, India