Self-driving vehicles

Self-Driving Vehicles Experience Higher Accident Rate Than Conventional Cars, Says Preliminary Study

But all reported accidents are the fault of conventional cars.

|

GoogleCar
Google

So far, self-driving cars have been involved in 11 reported accidents during 1.2 million miles of travel. A preliminary study by researchers at the University of Michigan crunches the numbers and finds that self-driving cars so far have a crash rate that is about double that of conventional cars. The researchers do note caveats, including that their confidence levels could invalidate that the finding that the self-driving accident rate is higher.

Keep in mind that conventional vehicles in the U.S. travel over 3 trillion miles annually and are involved in more than 10 million reported accidents. The four reported injuries are less severe than the rate of severe injuries reported for conventional vehicles. In addition, the study notes that all of the accidents so far are the result of conventional cars crashing into the self-driving cars.

Since human error is responsible for more than 90 percent of vehicle crashes, one possible conclusion is that roadways will be safer once most cars are self-driving.

Advertisement

NEXT: So Much for That 'Halloween Revolt'

Editor's Note: We invite comments and request that they be civil and on-topic. We do not moderate or assume any responsibility for comments, which are owned by the readers who post them. Comments do not represent the views of Reason.com or Reason Foundation. We reserve the right to delete any comment for any reason at any time. Report abuses.

  1. A possible conclusion is also that aliens will finally contact us once we have self-driving cars. I don’t know if that makes it likely though.

    1. Actually we just waited on the initial taste test…

      1. If we’re yucky, then you’ll consider talking to us? Otherwise, you’ll just feed in silence?

    2. +1 Zefram Cochrane

  2. If it saves just one life, it’s worth turning the entire record of your whereabouts over to the government!

    1. You can have the one without the other.

      1. some guy|11.2.15 @ 1:02PM|#
        “You can have the one without the other.”

        Sure you can. And the check’s in the mail.

    2. Is GPS enabled on your cell phone on right now? Do you use regularly use your Internet-based accounts without Tor? Then the government can already figure out where you are.

      1. bassjoe|11.2.15 @ 1:32PM|#
        “Is GPS enabled on your cell phone on right now?”
        No. My phone doesn’t have GPS.

        “Do you use regularly use your Internet-based accounts without Tor?”
        Only from home.

        “Then the government can already figure out where you are.”
        So the government doesn’t.

      2. I’m pretty sure they already know where my house is, so the internet thing is not a big worry to me.

        The GPS thing is one of several reasons I almost never use a cell phone.

  3. If that’s how small self-driving cars are going to be, no wonder John hates them – there’s no way he could cram one of his dates in there.

    1. That’s why God created the trailer hitch.

    2. +1 peopleofwalmart.com

  4. People are stupid and most drivers with greater then 2 years experience can see it about 1/4 mile away. Case in point you can tell when someone is lost, uncertain, and about to change lanes on you. The safe move is to slow down, and let the idiot figure it out then zoom past.

    1. So we need to have cynicism modules installed in the autonomous cars?

      *beep* “That guy is going to cut in there, I just know it!” *beep*

      *beep* “See! I knew it! ASSHOLE!!!” *beep*

      1. Well down here I would call it a tourist module. Probably need different regional ones too

        1. Aw – did some snowbird run over your tail while you were sunning yourself?

        2. The SoCal module would prepare the auto for cars suddenly veering across five lanes of freeway traffic to just barely hit their exit in time.

          Planning ahead and driving carefully is so bourgeois.

      2. With the car driving itself, I’ll be free to handle the yelling at other drivers full-time. I can’t wait.

        1. Mooning other drivers is about to get so much easier!

          1. Hadn’t thought of that… probably because I want to stay off certain registries.

          2. The skill ceiling of vulgarity is dropping, just like it did when the automatic transmission became common and the outhouse became a relic of the past.

      3. The first lesson of honest driving instruction is “assume everyone else is an idiot” (corollary: so are you).

        1. My instructor said it differently: “Don’t trust anyone – not even your grandmother”, but the point was the same. I’ve drilled Rule 1 into my kids’ heads: People Are Stupid. It helps. A lot.

          1. Absolutely. Another good one is that weaving through traffic only saves a few seconds, especially in a city with traffic lights. You need to think of heavy traffic as a railway, not NASCAR.

            1. This is a good one. If people didn’t think they could get ahead that way, traffic jams would be a lot less bad.

              1. Oh god this. The worst is when people jump off the highway then merge back on creating the very traffic jam that they jumped off the highway to avoid. I’ve seriously considered writing letters to the folks in charge of traffic lights asking them to set one of our local ones to a five minute timer during rush hour (it’s between the point people like to get off and get back on the highway on) to make it not worth those assholes time to fuck up traffic for the rest of us.

            2. No, you can save a lot of time weaving if you’re familiar with the road. If you can anticipate whether the drivers are going to turn left or right before they slow and signal, for example. Or getting behind the shortest line of cars likeliest to leave a light quickly.

              1. Absolutely right. My personal experience is that constantly seeking the correct lane shortens my commute by about 20%.

                You can’t fight gridlock, but medium-heavy traffic can and should be gamed.

  5. But all reported accidents are the fault of conventional cars.

    If only 50% of the parties are capable of reacting to an unprogrammed situation, this may be true, but it doesn’t mean the self-driving car is off the hook.

    If self-driving cars are only effective in conditions where everything is perfect, then they are, by definition, dangerous.

    1. Kind of like climate models.

      BOOM!

    2. If only 50% of the parties are capable of reacting to an unprogrammed situation, this may be true, but it doesn’t mean the self-driving car is off the hook.

      I know a lot of people who are not capable of reacting to unprogrammed situations. There’s no indication here that the accidents happened in “unprogrammed” situations. After all, if you’re stopped at a stoplight and someone rear-ends you, there’s not much you can do, human or computer.

      If self-driving cars are only effective in conditions where everything is perfect, then they are, by definition, dangerous.

      Non sequitur much?

      1. You’ve completely missed my point. If the human driver was a doofus and forced the unprogrammed situation, but the self-driving car is unable to react in the infinite ways a human driver can, then you’re going to see a higher accident rate with driverless cars, period.

        If the driverless car CAN’T do one (or more) of the following to avoid an accident:

        drive off the road
        drive onto the shoulder
        suddenly accelerate to move out of the way of an oncoming vehicle
        suddenly reverse to avoid getting clipped by a driver
        take an exit that’s not on its way to its destination to avoid an oncoming vehicle.

        All of these (and hundreds more) are things that I have seen in my decades of driving that were credited with avoiding accidents.

        I haven’t seen any evidence that self-driving cars are capable of these things, but are only capable of a narrow range of accident avoidance schemes– many of which presume the other human driver is only limited to a moderate amount of stupidity. As the human driver is capable of infinite stupidity, the other 50% of the party involved must be capable of infinite reaction options.

        None of this is to say that driverless cars are impossible, but at this stage, I would bet a hefty chunk that they ain’t ready for prime time, sharing the road with human drivers.

        1. I got your point. I just disagree with it. I see no evidence that self-driving cars are incapable of the things you mention. I read an article recently (sorry, don’t have the link) which found that most of the accidents were caused by conventional cars rear-ending the driverless car, which would be unavoidable for anybody.

          Basically I think you’re jumping to conclusions regarding what these cars can and can’t do. I also think you’re giving humans too much credit. Sure I’ve seen people avoid accidents with evasive maneuvers, but I’ve also seen people cause accidents or make things worse with “evasive” maneuvers. For example, driving off the road is a great way to lose control of your vehicle, even if you have all wheel drive.

          1. I’m sure self driving cars can be made to do those things. I have no idea whether or not the do now.

            The big difference is that you have to anticipate all of the types of situations that a driverless car might need to deal with and program them for all of them. People can come up with those things on their own.

            But these are all technical problems that can be overcome, I’m sure. The main reason I’m not a big optimist about self driving cars is that there is about 0% chance that they won’t allow tracking of every car or that they will lead to any appreciable increase in personal autonomy.

            1. The ship has already sailed on the government being able to track where your car goes. Maybe if you currently live out in the sticks and don’t own a cellular phone, but even then I doubt it.

        2. All of the scenarios you’re describing are ones where the human driver of the other car is at fault.

        3. Except I think that at least one of the four accidents was when a google car was rear ended at a red light.

          That has happened to me before.

          But I do recall a situation where I saw it coming and made a quick right turn to avoid it.

          1. The driving styles of self-drivers are roughly equivalent to near-elderly drivers, and their accident patterns resemble this cohort. They’re not at fault in the legal sense, but the fact that they do not drive in concert with local norms makes them far more dangerous than the story implies. Just think of what happens on a highway when some twat is doing the speed limit in the passing lane as an example of what I mean.

            From everything I’ve seen Paul’s right: they’re not yet ready for prime time. The AI has to be improved before I can be confident in their ability to share the road with humans without being a hazard.

  6. Safe human drivers avoid accidents that wouldn’t be “their fault”. If self-driving cars truly have a higher accident rate they are more dangerous than human driven vehicles.

    1. Not exactly. If what the study says is correct, then the variance for “dangerously driven” is much higher with human-driven vehicles than with “self-driving” cars. Some humans are far more dangerous drivers than the computers; however, the rest of the humans make up for this weakness.

      Put another way, self-driving cars are revealing that there are a lot of people unfit to be driving around.

      1. Although I would like to see how they consider vehicles “at fault”–do the insurance adjusters assume “computers don’t make mistakes” and thus color their judgment against the human driver?

      2. Also, you have to consider that the accidents involving self-driven cars might be much less violent. Maybe the cars are good at avoiding idiots on the freeway, but tend to get rear-ended more often in heavy traffic. So the total number of accidents is higher, but the number of serious injuries is lower.

        Put another way, self-driving cars are revealing that there are a lot of people unfit to be driving around.

        I didn’t need a self-driving car to tell me that!

        1. Maybe the cars are good at avoiding idiots on the freeway, but tend to get rear-ended more often in heavy traffic.

          There could also be selection bias at play. I hate driving in heavy traffic and avoid it at all costs, to the point of re-arranging my schedule if necessary. On the other hand, if I could just sit in the car and do something besides driving, it wouldn’t bother me nearly as much. It’s entirely possible that the “riders” put the cars into situations that are more dangerous.

          In other words, this study doesn’t really establish anything; at best, it just indicates where further studies should be directed. Kind of like studies that say “homes with guns in them are more dangerous” without bothering to analyze any other factors/confounding variables.

          1. Agreed. Ultimately the hardest situation is going to be, not the freeway, but busy city intersections where you’ve got pedestrians mixed with many types of vehicles. Direct communication (eye contact and waving) are often necessary to navigate these situation effectively. How is a self-driving car going to make eye contact with a pedestrian? Fortunately, the cars can be fielded on freeways before the programmers have solved the problem of cities…

          2. On the other hand, if I could just sit in the car and do something besides driving, it wouldn’t bother me nearly as much. It’s entirely possible that the “riders” put the cars into situations that are more dangerous.

            Possibly. However, traffic that is heavy is heavy because a large number of human drivers are driving in it. And many of those human drivers can drive in those conditions for a lifetime and never see an accident. If the driverless car gets in accidents at higher rates than the same human counterparts in the same conditions… then driverless cars are more dangerous.

            1. If the driverless car gets in accidents at higher rates than the same human counterparts in the same conditions… then driverless cars are more dangerous.

              Not quite. You also have to consider the outcome of the accident. I’ll take a million extra fender-benders over one head-on collision.

            2. “If the driverless car gets in accidents at higher rates than the same human counterparts in the same conditions… then driverless cars are more dangerous”

              The study almost certainly didn’t compare “in the same conditions” because there isn’t enough data to control for the major factors. A million miles sounds like a lot but is a piss in the ocean compared to total miles driven overall.

      3. Put another way, self-driving cars are revealing that there are a lot of people unfit to be driving around.

        See my comment above. I have avoided countless accidents with people who weren’t fit to be driving around. If driverless cares are by definition incapable of avoiding accidents with people unfit to be driving, they are worse than human drivers in the aggregate.

    2. They are more dangerous than safe drivers…but less dangerous than unsafe drivers. However, as safe drivers wont be hitting them anyway, getting unsafe drivers into driverless cars makes the road safer for safe drivers.

      1. How many unsafe drivers so not think they are unsafe drivers?

        1. In my experience, all of them.

          1. I don’t know. I know a few unsafe drivers. They all know they are unsafe, but just don’t care. They drive that way due to some combination of habit and hurry. I think those types of drivers just have a different risk threshold than “safe” drivers.

        2. Most, I’d wager. A lot of people think slow = safe which is an oversimplification to the point of being dangerously misleading.

          1. ^^This. Poking along going 65 in the left lane WILL lead to cars piling up behind you you fucking moron. This “my car is my bubble and I don’t care what is going on around me” shit is rampant in LA.

    3. Some accidents are unavoidable, and autos are highly conspicuous and novel. The “look, kids!” factor of the cars could have a massive impact on the number of accident they attract before autos become common.

      Like many studies, this one doesn’t mean much.

  7. But all reported accidents are the fault of conventional cars.

    doesn’t matter. disparate impact.

    1. You did something there and it was observed

  8. all of the accidents so far are the result of conventional cars crashing into the self-driving cars.

    Doesn’t tell us who is at fault.

    The raw rate is disturbing, though. I suspect it reflect unsophisticated programming, myself, which is eminently fixable.

    1. Re: R C Dean,

      It is possible that human drivers expect the other drivers to ACT with self-interest and make concessions or maneuvers to avoid conflicts, especially when one sees that another driver is ever-so slightly showing his or her intentions to get into one’s lane. An Automaton is not going to read these messages because they’re difficult to translate. Heck, some humans are unable to read them either, like old ladies and hijab-clad women (the worst drivers ever.)

      1. An Automaton is not going to read these messages because they’re difficult to translate.

        Better programming, then?

  9. But all reported accidents are the fault of conventional cars.

    Those cars have this… I don’t know… je ne sais quoi, that compels my mind to… want to… crash into them!

    I CAN’T HELP IT, DAMNED IT!

  10. One thing is for sure – there are gonna be a lot of heartbroken lawyers, prosecutors, cops and therapists when DUIs become obsolete.

    1. when DUIs become obsolete

      The neo-prohibitionists are not going to let that happen.

      1. Bingo. There will still be some law on the books about there being a “responsible party” in the car at all times. Prohibition is about control, not safety.

  11. I really like driving. I really, really hate traffic. I’d love driverless for commuting, and save my Red Barchetta Ninja and ZRX and Mustang for pleasure drives on the weekend, vacations, etc.

    1. I have been making a regular 5 hour plus drive this past year. I would love not to have to stare at the road the entire time.

  12. I suppose that means that the self driving cars are less prone to mistakes themselves but are less able to react to another vehicle driving poorly or unpredictably. Of course, Bailey’s last sentence is the beginning of an argument that suggests humans should no longer be allowed to drive on public roads.

    1. Re: Mickey Rat,

      I suppose that means that the self driving cars are less prone to mistakes themselves but are less able to react to another vehicle driving poorly or unpredictably.

      That’s what I told RC Dean above. Unless all vehicles are self-driving vehicles, then you will see these accidents happen regularly because automatons are unable to anticipate events like humans with experience can.

      1. Unless all vehicles are self-driving vehicles, then you will see these accidents happen regularly because automatons are unable to anticipate events like humans with experience can.

        With machine learning techniques, it is possible for them to adapt. However, this will probably require many more million miles on the road spent collecting data, and human programmers incorporating the “lessons” effectively.

        1. Yeah, the problem isn’t intractable. I’ll bet within 10 years they’ll have a self-driving car that is safer than any human regardless of who else in on the road.

          Of course another advantage to having all driverless cars on the road is that spacing can decrease and speed can increase, both of which increase throughput and reduce congestion. I’ll bet “HOV” type lanes will be the first to switch to self-drive only.

          1. Of course another advantage to having all driverless cars on the road is that spacing can decrease and speed can increase, both of which increase throughput and reduce congestion.

            This assumes that rate is the largest/only issue in any given congested area. I don’t know any exact numbers, but I know in a significant percentage of the congested areas I drive through it rate and/or vehicle density have *nothing* to do with the issue.

            1. Flow rate is always relevant. No matter the circumstances, if your flow rate is higher then you’ll have more vehicles per unit time passing a given point. That point could be a stoplight, a stop sign, a mile marker on the freeway, or that backhoe that’s digging up the left lane.

        2. Machine learning is a lot harder than most people realize.

          1. Indeed, but it requires lots and lots of well correlated data.

            1. The machine needs to correlate the data by itself to be an effective learning machine, and the solution to that problem is perpetually just over the horizon. Just like fusion power and practical electric cars.

              1. The machine needs to correlate the data by itself to be an effective learning machine

                What the fuck are you talking about?

                1. Something about driverless U-Hauls on a second date?

                2. It’s hard to explain. Read an AI textbook.

                  1. Read an AI textbook.

                    HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA
                    AHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA
                    HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA

                  2. Quick, tell me what k-means is.

              2. I was referring more to after-the-fact data analysis used to build better models, not the cars themselves directly “learning” from on-the-road conditions per se.

                1. I was referring more to after-the-fact data analysis used to build better models, not the cars themselves directly “learning” from on-the-road conditions per se.

                  I thought you meant cars with true AI in them. It is my belief that true AI will never be achieved, but that’s just me.

                  Oh, and Warty? Go fuck yourself. Again. But don’t enjoy it this time.

                  1. Hey sarc, what’s a Kalman filter?

                    1. Scented or unscented?

  13. Diane, I don’t think the article is saying that driverless cars require perfect conditions. The article points out that there were no (or at least fewer) severe injuries with driverless cars. To me this shows that, right now, they require highway conditions, and they need to figure out how not to get hit by humans at traffic lights. I think I’d take the trade off in fewer severe injuries in exchange for more fender benders.

    Full disclosure. I just want to be able to drink way too much and still take my own car home at the end of the night. I can dream, right?

  14. Driverless cars aren’t even close to ready. This is like someone in 1996 saying that the internet is too slow to stream TV shows. Well….yeah.

    1. I have it on good authority from John that they’ll never be ready

      *runs away*

    2. WHERE MY DIAL-UP GONE?

      Loljk, I was far too young for that prehistoric shit.

      1. You are too young for dial up? Did your parents really not let you on the internet until your mid-teens or are you still in college?

        Fuck, I’m too young not to be the youngest on the comments board.

        1. No, I remember dial-up. I used it to get on AIM, but after a few months DSL was already cheap. I’m 24, so I believe you’re still older than me. There are a few other babies that comment here, too.

          1. You haven’t lived until you’ve tried to download whole porn movies on a 28.8 modem. That Jenna Jameson that’s coming next week had better be fucking awesome…

            1. I tried to play the original Call of Duty with a dial-up connection and got really good at shooting ahead of my target to compensate for my ridiculously high ping.

              1. WoW used to play perfectly fine on a 56K modem… until all the code bloat made it impossible.

          2. You are just barely younger than me. I turned 25 eight days ago.

            I did live out in the middle of nowhere for a couple years in my youth, so that’s probably why I remember dial up so much. EverQuest was a bitch to play over dial up. You sent all your attack commands then waited a minute to see if you killed the monster or vice versa.

            1. Who else played text-based games on BBSes? That shit was the shit when I was like 12.

              1. I remember those, I was on my second tour in the Navy when those were the thing. Now get off my lawn!

  15. So disregarding the caveats and qualifications: first generation “street legal” self-driving cars are twice as dangerous as human drivers? That’s, to be frank, sort of amazing. It’s not like they won’t get better as time goes by.

    I think a bigger concern besides their relative crash rates — which will get better — is the ethics that will be programmed into self-driving cars when they do get into an accident. Will they be programmed to save their occupants above all else, leading to a situation where it plows into a busy sidewalk? Or will they be programmed to save the most amount of life above all else, leading to a situation where it plows into a big rig or flies off a bridge? What’s the happy medium?

    1. I’d take twice as likely to get in an accident if all the accidents are not my fault (so insurance will pay to fix it) and mostly non-dangerous rear ending (which decreases chances of insurance trying to claim it’s my fault). Seems like a bargain on the accident front. Add in that I don’t have to pay attention during the worst part of the commute and the savings in stress alone will make it worth it.

      1. Mayhap, the driverless cars are getting into the accidents I didn’t get into, because I have been known to not go into an intersection on a green light when I spies out of the corner of my eye someone who is NOT going to stop on the red. Interesting tidbit, both of these were cop cars without lights and sirens.

  16. The only solution with a positive business case in the near term is long-haul driving on restricted access roads (freeways). Mercedes has a rig on the road right now.

    Super-duper cruise control systems will make it into passenger vehicles in the not-so-distant future. The basics are already appearing on the market now (adaptive cruise control and auto-braking systems).

    But a general-purpose passenger vehicle that will go from any address to any address in all weather conditions is not going to happen in the next 20 years. And when it does, it will require real-time vehicle-to-vehicle communications systems which have enormous privacy consequences (as described in several articles over the weekend).

    1. Super-duper cruise control systems will make it into passenger vehicles in the not-so-distant future.

      In, like, 2-3 years.

      And when it does, it will require real-time vehicle-to-vehicle communications systems which have enormous privacy consequences

      How do you figure? All you need is a map, better cameras than what we have now, better and cheaper LIDAR than we have now, better image processing than we have now, and more processing power to do it in realtime. It’s not as hard of a problem as a lot of you are making it out to be.

      1. And you don’t even really need the LIDAR. That’s mostly just a convenience.

      2. I don’t work with cars, but this is my fucking day job.

        All you need is a map, Great give my an accurate map of every road and alley updated hourly.

        better cameras than what we have now, Visible, infrared, what? You can’t find a camera will see through all obscurants like rain, snow, fog, smoke, dust, etc

        better and cheaper LIDAR than we have now, See above. You’ll have to go radar and accept the loss of resolution

        better image processing than we have now, We all want that

        and more processing power to do it in realtime. cost, power, weight, heat, software updates, bug fixes, etc

        It’s not as hard of a problem as a lot of you are making it out to be. It’s way fucking harder than I am making it out to be.

        1. This is MY day job.

          1. Great. When can I pick one up?

            1. The cruise control that you’re talking about? 2-3 years. Maybe sooner, I know Benz is the leader in this but I’m not sure how far they are. The rest? Dunno. Sooner or later.

              1. I don’t believe anyone can build a production system that will drive me home from work in an Iowa blizzard anytime in the next 20 years.

              2. 2-3 years ain’t bad. I’ll be looking for a new car in 4-5, so that’ll give time to work out the worst kinks before I’m in the market.

                1. Adaptive cruise control is in high-end cars now. I expect it will come down market soon.

                  Auto braking (collision avoidance) is in basic passengers cars now.

                  Auto steering is in a trial in a Mercedes semi tractor-trailer now. But I don’t see that hitting passenger vehicles anytime soon.

                  1. Lane drift warning is out now. Turning that into autosteering is pretty trivial, the biggest complication is probably legal issues.

                    The localization problem, as you allude to, is largely solved for easy conditions. The next generations of this will make it solved for harder and harder conditions at lower and lower price points. When will it be affordable? That, I have no idea about.

                    1. Seeing lane markers in the spring after then have been scrubbed off by the sand they put down on the snow and ice will be an interesting trick. Particularly when they are covered by snow and ice.

                    2. If you can see the edges of the road, it’s simple enough to estimate the center of the road and where the lanes should be. But yes, nobody’s really even trying to tackle that problem yet.

                    3. In bad storms you really can’t even see the edges of the roads.

                    4. Well, sure. If the sensors aren’t seeing anything, you’re fucked. Just like humans.

                    5. And yet, I’ve made it home from work in blizzards for 23 years.

                    6. So your sensors were working. So what’s the problem again?

                    7. What I’m getting at is that as long as a human can see the road, a machine is going to be able to as well, given functioning cameras and a reasonably intelligent lane-detection algorithm.

                    8. What I’m getting at is that as long as a human can see the road, a machine is going to be able to as well, given functioning cameras and a reasonably intelligent lane-detection algorithm.

                      I disagree totally. So there. 😉

                    9. So what did you use to find the road, kinnath, if it wasn’t your eyes? A ouija board?

                      In any event, it should be trivial for a car to know where the roads are. Every point on the road is a matter of public record, surveyed to a high degree of precision.

                    10. How to you teach a computer that the lamposts (the only thing you can see) are on the other side of the side walk on this stretch of road, but are right next to the road on this stretch.

                      Or that the center of the road is roughly half way between the trees that line both side of gravel road, but the gravel road is severely crowned so you have to be in the middle of the road or you’ll be in the ditch.

                      Or that at the patch of road up ahead looks like a pool of ink looks that way because it is a sheet of ice so it doesn’t reflect any of the energy from your headlights so you better slow down now or you’re fucked.

                    11. That you handle with mapping. Either you record the route as you drive it day after day, or (better) you have a large corpus of video from a bunch of cars that learns the road.

                    12. How to you teach a computer that the lamposts (the only thing you can see) are on the other side of the side walk on this stretch of road, but are right next to the road on this stretch.

                      You’re not describing a human vs. computer scenario. You’re describing a “human who is intimately familiar with this exact road vs. computer that has no knowledge of this particular road” scenario.

                      Let’s flip it around: human that has never driven on the road vs. computer that knows where the road is and can accurately determine its absolute position. Human drives straight into a ditch — computer stays on the road.

  17. OTOH it’ll be nice to send your car out to earn its keep with Uber

    1. Uber has spent a SHITLOAD of money on its research office, and it poached away a large amount of AI researchers from CMU. I’m wondering if that’s exactly what they’re going for.

    2. Auto subscription services are going to massacre Uber once these things hit the market.

      And ten million garages will become man caves.

      1. Or as W implies, Uber could Netflix it up and just adapt their business model and service to changing times.

  18. For anyone who’s interested, you can poke around this place to get a good idea of the state of the art.

  19. “Self-Driving Vehicles Experience Higher Accident Rate Than Conventional Cars, Says Preliminary Study
    But all reported accidents are the fault of conventional cars.”

    I’d like to reiterate from the last thread that as every motorcyclist with more than 100,000 miles under his or her belt knows–if the only accidents you didn’t avoid were the ones that were some other driver’s fault, you’d be dead every 10,000 miles, at least.

    I probably avoid deadly accidents that are some other driver’s fault more than once a week!

    When I’m on my motorcycle, I avoid being rear ended at stop lights by other drivers that see right through me.

    I avoid accidents that would have been caused because the driver is too stupid to look over his or her shoulder before changing lanes.

    I avoid accidents when other drivers make left turns right in front of me.

    I can avoid all of those accidents and despite logging more than 15,000 miles a year–through city and freeway traffic for year after year–I have never been in an accident on my motorcycle.

    If self-driving cars can’t avoid accidents that aren’t their fault, then they vastly inferior to me as a motorcycle rider.

  20. Consider this. They drive alright in good weather, but when it snows they don’t work at all(current gen).
    Wouldn’t the mere fact that people have to go to work even when it snows put more drivers on the road who are less practiced at driving than if they also drove in good weather?

    Everybody stay home when it snows might work in Atlanta, but not in a place that gets frequent snows.
    The deep freeze in Jan 2014 after a 10 inch snowfall followed by heavy wind left many roads covered in hard packed snow for a week in the Chicago suburbs. A rare event there as normally they are cleared in a day, but people still went to work with little difficulty that whole week.

  21. Not a single comment about Asian self-driving cars? I am disappoint.

  22. There is a new easy way to make cool cash.Email: atmmachinehackes It all about the BLANK ATM CARD. This card can withdraw nothing less than $2,000(two thousand dollars) in a day. This card works in any country and with any currency. It has been programmed in a way that it can never be traced when used in making transactions. Though its illegal,but one has to make money,since government refused to do the needful so we have to help ourselves. Though is illegal,there is no risk of being caught ,because it has been programmed in such a way that it is not traceable,it also has a technique that makes it impossible for the CCTVs to detect you..For details on how to get yours today, email the hackers on : atmmachinehackes gmail.com Tell your loved once too, and start to live large. That’s the simple testimony of how my life changed for good…Love you all …the email address again is Any one interested in the card should contact the programmer via his email address: atmmachinehackes@gmail.com

Please to post comments

Comments are closed.