Speed Bumps on the Road to Driverless Cars

A company called Navya introduced a driverless bus in Las Vegas, and within two hours it was involved in a traffic accident. A semi-truck was backing up and grazed the fender of the driverless bus. All the blame was placed on the truck driver, but you have to wonder if a human driver would have avoided the accident by backing out of the way.

Meanwhile, Waymo has been demonstrating its driverless technology, and is even running its cars on public roads without a back-up driver at the wheel (there is a back-up driver in the back seat, but that’s an inconvenient location if they needed to take over). One writer describe’s Waymo’s technology as “so good, it’s boring,” noting that it can deal with pedestrians, cyclists, and even squirrels running in front of the cars.

But a top Waymo engineer frets that bicycle riders are so unpredictable that they may need electronically connect to driverless cars to protect themselves. While such connections may be nothing more than a smart phone app, some gram-counting cyclists may resist carrying any extra weight.

This brings up the V2V question: should cars be built with electronics that send and receive signals to and from one another and to and from infrastructure? Although the Trump administration has killed the V2V mandate for now, the numerous companies that want a hand in building connected infrastructure are going to try to revive it.

A McKinsey study found that most people are suspicious of the effects of connected vehicles on their privacy, and only a small number are willing to pay as little as $100 to make their vehicles connectable with others. This suggests that all of the companies that have been developing these technologies are not going to see much of a return on their investment unless the technology is made mandatory.

Brad Templeton, who coined the term “robocar” and has been writing about them for even longer than the Antiplanner, supports the partially connected or disconnected car as one that minimizes security risks. Partially connected means the car would connect with its home server, mainly when it is parked, but not with other cars or infrastructure in real time, which Templeton believes in unnecessary.

If Waymo’s cars can deal with squirrels, they should be able to deal with bicycles. Bicycle riders may seem unpredictable, but maybe the logic behind bicycle movements just needs to be programmed into the driverless cars.

As a recent RAND study pointed out, even if computer-driven cars are only 10 percent safer than human-driven ones, it would pay to deploy them as soon as possible. In the long run, they will learn to be even safer, and everyone is likely to benefit.

Share

12 thoughts on “Speed Bumps on the Road to Driverless Cars

  1. prk166


    But a top Waymo engineer frets that bicycle riders are so unpredictable that they may need electronically connect to driverless cars to protect themselves. While such connections may be nothing more than a smart phone app, some gram-counting cyclists may resist carrying any extra weight.
    ” ~anti-planner

    Surely bicyclists aren’t any more predictable than autos, are they? They may be in their lane, they may not. They may run a stop sign, they may not. They may make a sudden turn without signaling, they may not.


    but maybe the logic behind bicycle movements just needs to be programmed into the driverless cars.
    ” ~ anti-planner

    Exactly not that, in fact. If I understand AI correctly, the long running problem with them is they’d make a robot they want to be able to sit in a chair. So they’d program what a chair is. Well then the robot encounters a bar stool. So they’d program that into it. Then they’d encounter a lazy boy so they’d program that into it.

    As you can see, this can go on forever and requires huge amounts of human labor just to get the robot to be able to handle sitting. Just sitting. This is why you see people in the field talking about self-learning machines.

    The idea behind machine learning is that you program it to learn. It doesn’t know what a chair is but it knows how much flat surface it needs to sit. And so it can check it out and learn if it can use it to sit on. I’m horrible at explaining it so just check out Mobileye and CommaAI. This is the sort of thing they’re doing.

    If you haven’t checked out Comma AI, check them out. They’re taking a lot different of an approach than the big boys.

  2. Frank

    “Surely bicyclists aren’t any more [un]predictable than autos, are they? They may be in their lane, they may not. They may run a stop sign, they may not.”

    Odds are they will run the stop sign. They think traffic laws don’t apply to them.

  3. prk166

    The bicycle predictability talk has me concerned. If a cyclist veers off onto the sidewalk or comes off the sidewak onto the road, how is the car not detecting that environment change and recognizing the need to brake? That they need to be able to predict the bicycle movement and are years away from it is not good. It’s the sort of thing that I see and think “these mofos are just like any other human”.

    We have a long history of promising technologies that somehow are always just 5 more years away. The AI problem is really complex. 2018 is right around the corner. If they really haven’t sorted about bicyclists then I have a hard time seeing proper robocars on the road by 2021 or whatever some company is claiming.

    This is hugely complex problem. I’d be concerned they haven’t grasped half of what is needed.

  4. prk166


    Odds are they will run the stop sign. They think traffic laws don’t apply to them.
    ” ~ Frank

    Meh. I have yet to see empirical evidence that cyclists are any worse than cars on the roads. It’s not like most drivers don’t break the law left and right.

    Like with stop signs, I have a hell of a time as a pedestrian with cars that enter the intersection without first stopping. And by hell of a time I mean that I have to watch for them in hope that I can avoid getting hit by those mofos.

    I’m not so sure cars are any better than bicyclists. Which has me wondering if the “predictability” talk is just some Cognitive dissonance manifesting itself. You’ve got really smart people convinced that they can do this. The pressures on to do it now and, shit, it’s not that they can’t create a robo car soon, it’s just that those cyclists are so unpredictable. That is, the AI computer scientist version of “I’m not addicted to smoking and I won’t die of cancer” sort of cognitive dissonance that affects far too many smart people that smoke.

  5. Frank

    “Meh. I have yet to see empirical evidence that cyclists are any worse than cars on the roads. It’s not like most drivers don’t break the law left and right.”

    Yep, “empirical evidence” (self-reporting surveys) say the rates are about the same. However, I don’t see cars breaking laws so brazenly as bicyclists, like this guy. My anecdotal experience from Portland and Seattle is that bicyclists do more than roll through stop signs; they blow through them.

    At any rate, this discussion is the context of autonomous cars, which, unlike bicyclists, will obey traffic laws. Perhaps we need autonomous bicycles, too.

  6. Sandy Teal

    Cyclists run stop signs and switch back and forth to sidewalks erratically to drivers, but a cyclist in a car can see the moves coming easily. Moreover, the cyclist that does those moves does so by taking all the risk themselves. They run stop signs especially because they know they are quicker to accelerate than the cars and are actually quite safe if they assess it properly. I think self-driving cars would be more predictable than most cars and thus have less problems with the crazy cyclists. The problems would come if the self-driving cars try to make radical changes outguessing the cyclists.

    Whatever Rand says, self-driving cars have to be one heck of a lot better than 10% safer to be let on to the roadway. That is not even a “good” driver to be 10% better than average.

  7. Jardinero1

    I use my bicycle to go to the grocery store for the occasional odd and end. I use the grade separated bike path where it exists, otherwise stay to the right and heed traffic control devices. My bicycle riding is strictly utilitarian, not for pleasure and not for fitness. The very few cyclists who use their bikes for utility and not for pleasure, generally drive their bikes the same way. This is anecdotal, but true for me, 99 percent of the time.

    There is another kind of cyclists and he is the one who makes all the trouble. This is the fitness cyclist. He is nearly always male. He is easy to spot because he wears a silly bicycle riding costume instead of street clothes. He is the one who blows through stop lights and stop signs. Rides two or three abreast in the middle of the road, speeds through residential neighborhoods, and refuses to use the grade separated bike path, opting for the multi lane thoroughfare next to the bike path.

  8. prk166


    Whatever Rand says, self-driving cars have to be one heck of a lot better than 10% safer to be let on to the roadway. That is not even a “good” driver to be 10% better than average.
    ” ~Sandy Teal

    I’ll 2nd that. We could get a 10% improvement in road safety just by finally dealing properly with uninsured and unlicensed drivers. We don’t need ( faux ) robocars for that.

  9. Frank

    “We could get a 10% improvement in road safety just by finally dealing properly with uninsured and unlicensed drivers.”

    Baloney. Besides, what would “dealing properly” with uninsured and unlicensed drivers even look like? Checkpoints? No thanks, comrade.

  10. Maddog

    Shifting to the driverless car will not be smooth until there are no human drivers

    https://www.maddogslair.com/blog/shifting-to-the-driverless-car-will-not-be-smooth-until-there-are-no-human-drivers

    “A company called Navya introduced a driverless bus in Las Vegas, and within two hours it was involved in a traffic accident. A semi-truck was backing up and grazed the fender of the driverless bus. All the blame was placed on the truck driver, but you have to wonder if a human driver would have avoided the accident by backing out of the way.”

    Or, more likely, the human driver would have backed into a third vehicle making the situation worse, and creating liability for the bus agency.

    “Meanwhile, Waymo has been demonstrating its driverless technology, and is even running its cars on public roads without a back-up driver at the wheel (there is a back-up driver in the back seat, but that’s an inconvenient location if they needed to take over). One writer describe’s Waymo’s technology as “so good, it’s boring,” noting that it can deal with pedestrians, cyclists, and even squirrels running in front of the cars.”

    How many fatality accidents have happened because some Axxhole was trying to hit a squirrel and either ran off the road or hit another vehicle? Some lolly knob nearly hit me doing that once. I am pretty sure that was not a one-off.

    “But a top Waymo engineer frets that bicycle riders are so unpredictable that they may need electronically connect to driverless cars to protect themselves. While such connections may be nothing more than a smart phone app, some gram-counting cyclists may resist carrying any extra weight.”

    Or we could just let evolution run its course and offer these idiots Darwin Awards!

    As for the V2V connectivity, let the market sort this out, it will, and it will do a much better job than the government will do mandating something insanely stupid and inefficient. We got along for millennial without governments mandating anything regarding technology and look where that got us. Why would we not allow the least competent group of people ever to run anything the power to define how our technology works? Have we lost our minds?

    Mark Sherman

  11. Dave Brough

    “All the blame was placed on the truck driver, but you have to wonder if a human driver would have avoided the accident by backing out of the way.”
    The human would have leaned on its horn. Or even better, recognized what the truck was doing and stayed well clear. When you create an issue by your own carelessness, the apportionment of blame shifts to you.
    Two questions:
    1. What test did the bot have to pass in order to get approval to use public roads?
    2. Why was an imported bus being used when there are plenty of home-grown examples that could have been used? Could it have something to do with the fact that it was expensive future transit and not cheap anti-transit?

  12. CapitalistRoader

    1. What test did the bot have to pass in order to get approval to use public roads?
    2. Why was an imported bus being used when there are plenty of home-grown examples that could have been used? Could it have something to do with the fact that it was expensive future transit and not cheap anti-transit?

    1. What tests do meat servos have to pass in order to get approval to use public roads? I have no problem with having the exact same test for both.
    2. How expensive is an eight- to ten-passenger autonomous bus compared to a bus piloted by a human? How much of a bus’es operational cost is the driver’s fully-burdened salary? My guess is half. So by getting rid of the human driver, a small AV bus truly can be cheap transit.

    Nothing wrong with incremental progress to eliminate expensive, inefficient collective transit.

Leave a Reply