SHOCK! Robot cars do CRASH. Because other cars have human drivers Self-driving cars have put in hundreds of thousands of miles on California's roads and, according to the state's Department of Motor Vehicles, they have had four bumps and scrapes along the way. In September 2014, California's DMV issued new guidelines that allow autonomous vehicles on state roads, as long as there's a human …
By the Google car under manual override, or by other road users?
When do we get an independent analysis to see if they were really unavoidable, or if the software messed up in some way that a typical human would not have?
I doubt I am the only one, insurers will want to know and I bet people considering such a car will want the equivalent of the NCAP ratings for 'droid drivers.
Yes, I sound negative, but the burden of proof has to be one the suppliers that they are better than the average human in all reasonable situations for most people to be willing to accept them. And that includes in dealing with the other human drivers that will be around for decades to come even after commercial availability.
Yes.. Even though technically right you are pretty negative about it.
But I have very little doubt the cars are pretty good at driving themselves. The simple fact that they've so far managed to maneuver through what passes for "driving a car" in urban California without any major accidents *at all* speaks for itself. The cars already do better than the average local.
Now.. If those cars manage to do a tour through Paris and actually survive, I'd be really impressed.
How do you know they are doing so well?
Yes they have managed OK on a pretty regular US system, but how much do they depend upon GPS/maps being completely correct? how do they cope with partially closed roads? What about twisting country roads with passing places? Temporary traffic lights? Polices flagging them down due to an accident or similar? Dumb meat-bags doing stuff that another meat-bag would see the warning signals of high stupidity and/or intoxication and keep well away?
Though Google are pop-pooing it, the accident rate seems to be about 5 times more than average, so its hardly a stunning display of everything being just right.
And Google have a vested interest in playing up the success and not talking about any known problems, do you really want to end up buying the high-tech Ford Pinto?
THAT is why there needs to be an independent analysis of what has actually been tested, and when failures have occurred, what should have been learned.
how much do they depend upon GPS/maps being completely correct?
Good question. I have experience of a road doing a 1 km deviation up a ravine and the road looking like Ω while the GPS said go straight ahead like _. The ravine was several hundred feet deep.
While I'm sure the google cars would recognise the road stopping without falling into the ravine, I'm not sure they would have any idea where actually to go instead. I can imagine them emergency stopped at right angles to the oncoming traffic in the wrong lane and the nav unit melting down in consternation while making meep meep noises.
afaik the cars only use satnav to determine their global position and what they can expect road-wise. They augment that info with what they actually "see" ( hence the almost military-grade sensor/camara package on them.) If A doesn't fit B the driver is urged to grab the wheel and do things himself, so that's at least one of the more obvious "safety" measures.
Given that this whole thing is watched like a hawk by just about everyone having remotely to do with transportation and legislation, I pretty much doubt that the several teratonnes of data that's been collected about those cars can easily be fudged or tut-tutted, especially those fenderbenders. On the contrary, given the way I'm told US insurance works, one of those cars is the last things you want to hit, given that you can't just write of the dent and be done with it, as most people do. Even ignoring all the (utterly non-google) other camera-type things possibly aimed at you at the time, your little scrape has just been Borged, and you cannot avoid Insurance Interference if only just because Google insuror will want to know what happened there...
While technically possible, it's simply too risky and ultimately counterproductive to fudge the data on accidents when it comes to this project, by Google of any of the other hopefuls. There's simply too much riding on it, and even with my paranoïdar deployed I , at least, can't see how any of the companies involved could get away with it for long, if at all.
But I'll one-up my previous challenge: Having successfully circumnavigated the Arc de Triomphe and the vagaries of Parisian coureurs, we shift the scene to Italy. Let's be fair there and do it classic Top Gear: The challenge is to pass through Rome , including the Colosseum **, followed by a leisurely drive to Napoli, and a pass through the city centre there, ending at an appropriate pub with a functioning car, and a maximum of 5 human interventions.
** locals may know better...testing grounds.. But that's one route I prefer never to have to do again..
"But I'll one-up my previous challenge: Having successfully circumnavigated the Arc de Triomphe and the vagaries of Parisian coureurs, we shift the scene to Italy. Let's be fair there and do it classic Top Gear: The challenge is to pass through Rome , including the Colosseum **, followed by a leisurely drive to Napoli, and a pass through the city centre there, ending at an appropriate pub with a functioning car, and a maximum of 5 human interventions."
I'll pull you one better, based on first hand experience. Try crossing from the south part of Metro Manila in the Philippines to the north side starting around 4PM in the afternoon, starting from the Airport at the south end, onto and along the bulk of Epifanio de los Santos Avenue to the North Luzon Expressway at the north end.
HINT: 4PM is the start of the busiest part of Metro Manila's day, and the airport's busy in the best of times. NOW, however, they're building a skyway, so there's construction everywhere. By the time you get out, the traffic will have spread to the rest of the city, and drivers there can probably give any European a heart attack with their reckless aggression, not to mention the wanton disregard for signs and lane markings. About the only things they obey are the odd traffic light.
Get through that back and forth a few hundred times without significant incident, then follow up with a few spins around the outskirts at night (where headlights are not a given), THEN I'll be impressed.
"Dumb meat-bags doing stuff that another meat-bag would see the warning signals of high stupidity and/or intoxication and keep well away?"
That's always been my concern of mixing meat-bag operated cars with autonomous cars. The robot cars are likely to drive "perfectly", ie to the letter of the law while meat-bag drivers will bend the rules to suit the conditions and operate on the assumption that other drivers will do the same. Once you have a number of "perfect" drivers following the letter of the law no matter what, there's likely to be a few more bumps. Obviously, legally, it will be the meat-bags at fault since they will be the ones bending the rules.
Doesn't work quite that way, the meat bags make mistakes in programming things. Yes, the Google cars are probably quite a bit better at it, but local experience with automated systems shows it is no panacea. In my case that would be the DC Metro (light rail). In theory, with a protected travel lane, no real cross traffic to speak of, and rails instead of wheels it ought to be easier to automate these things than cars are. And for a long time Metro ran them in automated mode because the meat bags tend to brake harder than the automated systems do. Still the meat bags managed to screw it up by mucking up the sensors, resulting in a crash that killed 8 people and injured hundreds. Immediately after the incident they switched back to having the meat bags drive, and haven't reverted to the automated system since.
http://en.wikipedia.org/wiki/June_2009_Washington_Metro_train_collision
"When do we get an independent analysis to see if they were really unavoidable, or if the software messed up in some way that a typical human would not have?"
In 7 of the 11 google cases across the USA, the car was rear-ended whilst stationary.
In 1 case it was rearended on the freeway.
2 sideswipes, 1 whilst stationary
1 caused by a fleshie running a stop sign in front of the car. (Odds are good the car reacted faster and better than a meatbag too. I've had it happen and barely had time to hit the brake before T-boning the idiot)
The Goo fleet drives more miles in a week than most people drive in a year.
Very true, unless it is withheld for "commercial reasons" or trade secrets, etc..
We really need the equivalent of the air crash investigation board to deal with such events in a way that the manufacturers cannot legally get out of, or withhold evidence from.
OK, maybe not as rigorous in minor cases, but to trust something as new and potentially dangerous like this demands an independent analysis.
You can see down votes given to the most uncontentious, factual explanations on this forum. I've no idea why.
It doesn't matter how good the content, people will downvote it if it's not what they want to hear / read.
But does anyone over the age of 12 really care about up/downvotes?
I almost downvoted -- because I don't think *air traffic* style of investigation will scale. We investigate airplane accidents so intensely because any particular incident may very well be fatal to a large number of people. Applying the same investigation standards to lower loss crashes such fender benders won't be sensible over the long term, because it will become clear that robot drivers have improved the situation.
Since we're in the early phases now of learning how to make self-driving cars, it will make sense to start with deep analysis -- and then trim back to a sensible level of cost/benefit.
Because here at El Reg, we don't expect to find emotional anti-technology idiots like you.
If anything what we need is the reverse, those kinds of investigation whenever meat bags are involved in accidents. But that ain't happening any time soon. Until it does, the rules should be pretty much the same for automated as meat bag.
I commented on another article weeks ago, mentioning something that just doesn't seem to have, um, registered yet.
I have to believe that all these cars have a cameras trained on many different angles to capture not only the telemetry by which they navigate, but also to capture the "what happened" for review. That that review would be as interesting to the law and insurance companies and others seems to have been skipped over.
I use as my jumping off point all the brouhaha over Google's StreetView and everybody screaming about invasion of privacy. They were even sued when mistakenly driving down someone's driveway.
Now how many cars do you think StreetView had active at any one time? And how many times do they come back and re-do views? (They've redone my street once in 5+ years)
Now what about when half the cars in the West are instrumented to continuously record everything going on around them? All the time and everywhere? There will be no privacy on streets. (Whether I'm for or against the concept of "privacy in public" is moot, but some feel quite strongly about this)
Why haven't I seen any mention of this?
It all depends on who is in control of the images, doesn't it. If they're taken with a camera which is an integral part of a car owned by an individual, I can't see any reason why the images are not the property of that individual, and the only way one could be made to cough them up is by a requirement of a court to produce evidence. I will not be buying or driving a car which uploads imagery to any second or third party without both my permission and my positive action to make it happen.
As far as the privacy on the street notion goes, one has no expectation of privacy on the street. Many cars already have dash cams, although maybe fewer in the UK than elsewhere. All in all, I think they're a Good ThingTM, because the imagery they capture is not, for the time being at least, being borged into some Big Brother database. Hell. I really hope that the Home Secretary doesn't read these comments.
These cars have a lot of high resolution camera. The data storage for just single 1hr trip will be really quite large (let's say 4x1080p30 cameras, that's multiple GB per hour). Now multiply that by every car, and the numbers of hours driven.
Not enough storage in the world for that lot.
Lets take a look at the United States and crashes of modern vehicles. By modern I mean 5 or less years old depending on who manufactured the vehicle, BMW a little older base gm's probably about 5 years. Now lets talk bout OBDII and CAN. For the uninitiated this the the little port in under the drivers side dash and stands for on board diagnostics 2nd revision or edition. CAN is controller area network or all the little computers in your car talking to one another. So you crash you 2015 Dodge Viper and someone dies, maybe you maybe the other driver it does not matter. So now your state highway patrol shows up at the crash scene and they have the cars towed to impound. Next, depending on which state you live in, they get a warrant to inspect your car (not all states require this step). The reconstructionist does a physical inspection of your car to see if anything looks like it contributed to your crash. Next, this is the fun part, they hook their machine (computer) into the little obd2 port and download the data in the various controllers on the can. Then they go back to the office. The computer now has up to the last 15 minutes of vehicle operational parameters (most vehicles only store 30 to 60 seconds which is plenty :high end cars store more to make the more complex car problem easier to capture) this nice data tells them at a minimum: how fast you were going, throttle angle (were you driving like a saint or with the peddle on the floor), yaw rate, steering angle, brake application (or lack there of) had you been driving like this for long, any problems with the car, how much fuel was in the tank, engine rpm, did you have the ac or heater on, was the radio on and how loud, if it was night were both your headlights working, etc., etc. So pretty much anything that has a controller dumps data to the main ecu and is available for all with the proper equipment to see. So the only think this adds is movies.
Of interest is the recent move by the airline industry to involve pilots more in the flying of aircraft. It seems that if you don't give humans something to do they tend to find something else to do (distracted driving) and can't respond quickly enough in an emergency if the computer fails.
The other question people have to ask. All the laws so far say that the driver of a self driving car is still the responsible party if it crashes. So you give up control but retain responsibility. Not a big sales grabber. I know that with laws like that I will drive, yes even if I am a worse driver. If i screw up its my fault, if the software screws up its my fault. I take option 1.
"If they're taken with a camera which is an integral part of a car owned by an individual, I can't see any reason why the images are not the property of that individual"
the moment you hand the car over to the insurer, it becomes the property of the insurer. Cars already have rudimentary black boxes on board (the airbag computer is continuously recording conditions in deployment and near deployment) and those have already been used to deny insurance payouts (one example being the mercedes which was recorded at 100+mph a few seconds before it crashed on an urban road with 30mph speed limit.
Insurers have handed telematics data to police for prosecution too.
"Why haven't I seen any mention of this?"
Because the same thing happens when a photographer takes a picture of the street. Unless you specifically were the focus, the courts have previously ruled you are under no expectation of privacy on a public street.
"Because the same thing happens when a photographer takes a picture of the street. Unless you specifically were the focus, the courts have previously ruled you are under no expectation of privacy on a public street."
And that's a perfectly reasonable stance to take when there are millions of people walking the streets of various towns and cities and maybe a few people talking photos which may or may not happen to include you. But when it's every car driving past you, possibly uploading the imagery to the Borg cloud in full HD colour it might be time to revisit the law. After all, you may not have an expectation of privacy in public but you also don't have an expectation of being filmed everywhere you go as soon as you step out of the front door. "Expectation" implies a balance of some sort.
"And that's a perfectly reasonable stance to take when there are millions of people walking the streets of various towns and cities and maybe a few people talking photos which may or may not happen to include you. But when it's every car driving past you, possibly uploading the imagery to the Borg cloud in full HD colour it might be time to revisit the law. After all, you may not have an expectation of privacy in public but you also don't have an expectation of being filmed everywhere you go as soon as you step out of the front door. "Expectation" implies a balance of some sort."
No implication. It's simply you can't expect to control the part of the world that isn't yours. How do you think things worked back in the village days when everyone knew what everyone else was doing?
Google would say that.
"Google pointed out that those figures might be off, since a huge number of minor crashes aren't reported to the authorities. "
Our Cloud loving possibly future Overlord would say that.
The cars are 100% programmed by humans. When the many flaws in the software appear it may be mildly annoying or spectacularly fatal.
Well the fact that the software is written by humans is not that big issue... because the software for driving a car is rather simple. You'd be surprised, but there is not much in there on the algorithmic part.
The point why autonomous vehicles are better than humans is not that there is some clever software inside - one major reason is that they have better senses...eh sensors... and that their calculation is more accurate. Driving itself is basically keeping the steering wheel so it follows the road + avoid collisions with others.
/Zane
Driving is more than yaw computations. Sorry, was that a packet of crisps that can be safely run over or a rock that must be avoided by an aggressive manoeuvre. No time to get a response from Watson in this crappy 4G zone.
It stands to reason that a mesh of autonomous cars can process more information and not do the stupid things is humanoids do from time to time. But! What would happen if you were overtaking this car at the moment it decided that the abovementioned crisp packet was to be aggressively avoided? This could easily create accident scenarios that are not so today.
"Driving is more than yaw computations. Sorry, was that a packet of crisps that can be safely run over or a rock that must be avoided by an aggressive manoeuvre. No time to get a response from Watson in this crappy 4G zone."
A packet of crisps would probably return a different infrared signature than a rock, Plus there's the matter of motion (a packets of crisps will react to the wind differently than a rock due to weight and aerodynamics). And if it's a rock IN a packet of crisps, that's pretty much sabotage at this point.
Put it this way. A LOT of thought has gone into the various scenarios that the average driver faces as well as how we as drivers identify and react to these. The bulk of that knowledge is probably in the prototype cars, already at hand no Internet necessary. Same for the maps.
Would you like to be in a Google car going round the Colosseum in Rome, The Arc de Triomphe in Paris or, my favourite, the motorway to the airport in Delhi, at night? I would take a local taxi every time (and tranquilizers).
Somewhere there's a government regulation that specifies how frequently you can find a rock in a bag of crisps. I know. I once had to do the typesetting work on an official method that was used to determine the amount of rat feces in grain. Yes, the method was used to determine compliance with Dept of Agriculture regulations.
"Self-driving cars have put in hundreds of thousands of miles on California's roads and, according to the state's Department of Motor Vehicles, they have had four crashes along the way."
I'm just going to wait until these self-driving cars become more common; at that point it will be very interesting to see how they react to teenage drivers playing chicken with them.
In fact, knowing the novel uses which human ingenuity can find for any kind of technology, I would expect "RoboCar Chicken" to be the next big motor sport - with the tracks and arenas conveniently located on any public road you want, with participation open to all! (And expect it to be a big spectator sport on Youtube.)
"In fact, knowing the novel uses which human ingenuity can find for any kind of technology, I would expect "RoboCar Chicken" to be the next big motor sport - with the tracks and arenas conveniently located on any public road you want, with participation open to all!"
Put me down for a grandstand ticket ;-)
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
