Slashdot Mirror
How Google's Autonomous Vehicles Work
An anonymous reader writes "IEEE reports that Google's autonomous cars have logged more than 190,000 miles driving in all kinds of traffic, and the company is also testing a fleet of self-driving golf carts on its campus. In a recent talk, Sebastian Thrun and Chris Urmson of Google gave details of the project and showed videos of the robot cars driving themselves and even doing some stunts. The goal is that the technology will help reduce congestion, fuel waste, and accidents."
To be honest, I don't know if I'm ready letting go the controls of my car. Given Stuxnet and other stuff, I am seriously worried that this automated driving stuff will break down ad 75mph on the highway. Heck, I'm still driving a manual shift car!
We desperately need this tech. Aging populations drive less well after some kinds of age-related degradations, where automatic driving mechanisms could maintain mobility for affected people. Cars could also take kids to school, and park themselves in the parking lot, making the kids much less vulnerable to child molesters and bullies, as well as freeing Mom and Dad from transportation duties.
For everyone else, automation in driving will do a better job than people that are blabbing on cell phones, eating, drinking, fiddling with the radio, changing CD's, etc. etc. If drivers don't normally do these things, then automatic driving will enable them to do so, as well as a lot of other things that might be productive, like office work or simply enjoyable things like reading the newspaper or surfing the internet.
I'm all for autonomous vehicles but I think before it goes forward (ha ha) shouldn't there be some sort of legal framework in place? I mean the first fatal accident that can be even possibly attributed to an autonomous vehicle could very well kill the industry for a while at least. (I'm reminded of how the fatal crash of one of Buckminster Fuller's super efficient teardrop shaped 3 wheeled vehicles killed that concept).
How about for all "certified" (through rigorous federal testing) vehicles, there be "no-fault" collision insurance (or limits on damages). Unfortunately I'm neither a transportation expert nor lawyer so I'm just guessing.
first post?
No.
Write boring code, not shiny code!
When I was learning to drive, my teacher told me before you dodge (or brake for) any deer on the road, always check rear view mirror to make sure a semi trailer isn't following you to brake and kill 15 people behind that. It's easy to say deer.
But what if it's a kid. Gets harder right?
I am glad this robot car takes that decision off my hands yeah? :D
I'm sure Google's thinking of ways to monetize your travels by selling your private data.
"Don't be evil"? Ha. Sounds like an empty marketing slogan dreamed up by a giant ad agency. Oh, wait...
Like:
Can it, like I do, notice that the baseball rolling down a driveway may be followed by a child who is currently invisible behind a parked SUV?
Can it, like I do, notice that the driver *behind* me is distracted by her cell phone, has started late at the last three lights, so I should give myself more than average room between me and the car in front of me, so in case it stops suddenly, SHE won't have to stop as suddenly and will be less likely to rear-end me?
Can it, like I do, notice that even though the road has been clear of ice and snow, the next curve up ahead is deeply shaded and is likely to be slick?
Can it, like I do, notice that the baby deer is one one side of the road and his mother on the other, and even though he isn't charging across, it looks like he's about to do so, so I better slow down? Because this happened to me not one week ago, and it DID charge across.
Can it react to the highly dynamic and unpredictable world in ways that require human intelligence? And before you claim it's going to be a better driver than I am, note that I got my license in 1973, and have YET to have any sort of accident. Because I can provide human intelligence to the task of driving, and unless you want to claim the machine is as smart as a person, I am not so sure I believe I want to leave these choices up to it. I'll gladly fly on a computer flown aircraft, because that's a very, very different kind of task. No children appear in the sky from behind parked vehicles, and the pilot stands by the entire time to take over if something doesn't go right. Car drivers won't be that alert: they'll just doze off or play with their phones and not even glance out the window the whole time.
If automatic drivers initially reduce congestion (due to smoother driving) that effect will soon be diminished by increasing amounts of traffic.
Judging by my friends, most of them seem to prefer driving their own car if it will take them no more than 1.5 times as long as using the subway. If they had automatic drivers, they would probably accept even longer delays, so they might take their car instead of the subway even in rush hour when it would take more than 3 times as long.
This will also reduce incentives to move closer to your place of work or taking other measures to reduce your dependence on cars.
Don't get me wrong. I'm all for this technology. But make sure to combine with substantially increased gasoline taxes or you will be spending a large part of your life trapped inside a car.
this doesn't seem safe
I love powerlifting
i love the summary. the technology will help reduce congestion, will help fuel waste, and will help accidents. whoops :)
Until it gets tested by an independent group, it's still just a company's claims about their own product.
I wouldn't claim the machine is smarter then the smartest humans, but I would be willing to bet it is smarter then the average human (Just look around walmart, what you see represents the majority of the people on the road).
the impressive math to prove that airline software operates correctly, obviating the familiar testing found in most software development efforts, is also not popularly recognized.
Is it a rule, that there's an exception to every rule?
... and the company is also testing a fleet of self-driving golf carts on its campus.
Highschool coaches everywhere are jumping for joy.
Seriously. And it's more than enough reason to get it legal on the streets.
Just have a way to lock out the manual override until you can pass the built in breathalyzer. It can transmit encrypted times and duration of computer control to local police vehicles in addition to the fact that you locked yourself out so they won't bother pulling you over.
Of course you will probably have to pre train it to park. Or have a parking assist mode that you can engage even while intoxicated that won't let you get into an accident. And of course a button to pull over just in case you need to throw up.
No more drunk driving or risking the fine and you have your car to get to work in the morning. And no $50 one way taxi ride or having trouble getting a taxi at last call. Besides most people would probably prefer to nap on their way home anyway.
This is interesting, but I have to ask, why is Google doing this? What is in it for them?
It's nice that they are doing it, but as a business, they can't be sinking money into it (and risking being sued or damaging their reputation) without the expectation of a reasonable return.
So, how are they making money on this? Perhaps while you are driven around you have more time to make use of Google ad-supported services?
-- Braden's law of data: All data spends some of its lifetime in an excel spreadsheet.
A new program has unknown reliability, But a production program gains refinement and confidence with age. While it can still fail it becomes less likely. Just look at all the Fortran code still running.
What scares me most about Walmart et al is that all those mouth-breathing 'tards who can't even steer a shopping cart/trolley have driven to get there.
I say we take off and nuke it from orbit. It's the only way to be sure...
Just about everyone outside of US drive manual as well
I think this is an acceptable alternative to taking drivers licenses away from the elderly.
"I assumed blithely that there were no elves out there in the darkness"
The goal is that the technology will help reduce congestion, fuel waste, and accidents."
A much bigger goal for me is that it will increase mobility for the elderly and disabled. I saw my parents lives become dramatically circumscribed when they lost the ability to drive. I am within long view of the same thing myself, and I hope that these cars will be available before I get there.
And, to answer some other points, it doesn't have to be as good as the best driver, it only has to be as good as the average driver. In fact, if these cars existed, we could be more draconian about banning the very worst drivers. How would road safety change if we replaced the worst 10% of drivers with a automated car equal to the (raised) average?
And how many accidents are caused by drink, tiredness, texting/phoning etc? I am not saying they would all disappear, but if you could hand over to an automated system while you sent a text, the roads would be much safer. (I know you are an idiot if you test while driving, but empirically idiots exist).
In fact, the system could observe the driver and offer/demand to take over if the driver was below standard. (And call the police or refuse to run if way below standard?).
Consciousness is an illusion caused by an excess of self consciousness.
"The goal is that the technology will help reduce congestion, fuel waste, and accidents."
The goal is to make Google a lot of money.
right now I put my life in the hands of the other people on the street. the ones already playing with their phones or yelling at their kids in the back (commute is past 3 schools, 1 a HS)
If the drivers aren't watching the road now, at least put in a computer to watch for them. Hell, just have it apply the brakes if something gets in front of the car and it'd be an improvement
I wouldn't claim the machine is smarter then the smartest humans, but I would be willing to bet it is smarter then the average human (Just look around walmart, what you see represents the majority of the people on the road).
I would bet the machine is dumber than sh*t. It might be able to react quicker, but it isn't smart. All it will do is analyze a bunch of algorithms and choose what it is programmed to choose as the best one. That doesn't make it smart, just efficient.
Besides, Mr. Spock was definitely smarter than Captain Kirk, but Kirk seemed to excel at making the better choice.
...just wait for the Windows version...
Excuse me, but please get off my Pennisetum Clandestinum, eh!
Only Robots will be allowed to drive cars
Although all valid points, these are relatively rare circumstances. What about simpler everyday things, like yielding to a fire truck running on siren/lights?
40K for the car by itself. The LIDAR alone will set you back a quarter of a million. No idea what radars are involved, not to mention the processing power to do all the mapping. Then there is the value of all the software and other IP...
I would love to take my commute and shove it. That and be able to go out, get hammered and then have a computer safely drive me home. And I do believe I will see self driving cars within my lifetime, its a no-brainer with all the technology we have at our disposal.
Then again I do enjoy driving from time to time. I went down to Street, MD the other week and then drove up through Lancaster, PA to Litiz for a truck show. I had a great time driving through the country, such beautiful scenery. I even had fun navigating the narrow, winding country roads at night. It was a challenge.
Hell I own a 1961 Mack B61 tractor that has a two stick 9 speed duplex transmission and a 205HP Mack Thermodyne turbo diesel. I have a blast driving around old manual trucks. You just cant beat that feeling and the incredible simplicity of the machine, no electronics, not even to operate the engine. Electric is only needed to start it and you shut it down by pulling an engine stop cable. Heck, if the battery goes dead you can just pull start the truck. The generator only runs the lights and heater motor, and maybe a radio, if you had one. Even the wipers are air operated from the air brake system.
190,000 miles isn't allot. Just proving a new engine design can run reliably takes around a million combined miles of testing. I am glad they worked out traffic navigation but how many of those miles were in severe weather conditions? Snow, sleet, high winds, heavy rain, fog etc. Weather is never mentioned and the cars primary means of 3d mapping is done with a laser.
And here is the better question. If an automated car gets into an accident and someone is severely injured or killed, who is to blame? The occupant in the driver seat, vehicle owner, robot driver manufacturer or car manufacturer? Who gets sued? Seriously, as long as ambulance chasing lawyers are out there, car makers and their lawyers will have to figure out how not to get sued into oblivion and divert blame or responsibility for such cases. No matter how good the engineering is, the liability is a huge issue. Its easy now as you just blame the driver, a human being.
And as the number of automated cars increases, how do they cope with human piloted vehicles? Do they have the ability to see a blinker, slow down and let someone merge into their lane? If an automated car isn't "courteous" to other human drivers things could get hostile on the roads. If the robot cars are perceived as slow and lack courtesy, people will be hostile to robot cars creating a dangerous driving situation. With all the talk of robot cars packing themselves tightly together, how do these packs of cars react to human drivers. How do human drivers merge onto a highway with a conga line of robot cars driving along nut to butt. Around packed cities you also have allot of variables, people just walking into the street either knowingly or they are oblivious. A human driver can see a person walking toward the street and slow down in anticipating that the person will not stop. What about driving down a street and you see children playing? I always slow down when I see kids. I once almost hit a kid who ran into the street chasing her dog. I literally stopped two feet from her. That scared the shit out of me and since then I am cautious when I see children playing near a road. Can a computer sense that?
And lastly lets not get into stuxnet type worms causing massive pileups which result in injuries and deaths by attacking robot cars. I keep hearing about wireless cars connecting together forming their own networks to communicate. To me all I hear is an attack vector. Imagine a worm that commands all of the robot "cars on demand" to swarm a city or part of town. Imagine a million compromised cars blindly headed to the capitol building in Washington as protest of a war, policy or elected official. Clogging streets, running out of fuel on the roads blocking them etc. All sorts of mayhem. Its actually sounds funny now that I think of it.
Computers can drive flawlessly but the amount of v
I'm all for this! As a lead-footed driver I need something to help me reduce my speeding fines!
The "totally locked down search with a solid product" Google? Or the "randomly starts projects, fiddles with them for awhile then leaves them to mold or randomly kills them" Google? Not trying to flame, honest, but this company has some seriously spotty reliability issues when you take their whole portfolio into consideration. We're Google, we do search. And maps. And email. And social networks (not really). And....driverless cars?
Human and computer driving both have costs and benefits:
Human costs:
1) Inattentiveness
2) Emotional excitability
3) Lack of ability due to age, intoxication, other
Human benefits:
1) Ability to respond to extremely unusual situations
Computer costs:
1) Electro/Mechanical failures
Computer benefits:
1) Perfect attention to the road.
2) No emotional excitability.
3) Highest level of driving ability attainable.
I'll take the computer. With a manual override just so I can flatter myself.
No, congestion is caused by having too many cars for the the amount of road. If you are on an 8 lane highway, a there is some dumbass swerving back and forth across the lanes, and you and him are the only people on the road, there will be no congestion. Conversely, if you are driving on a two lane highway that is at full capacity. Adding hundreds of extra cars will cause congestion even if everyone driving perfectly.
Yes, people doing dumb things can be the final straw, but the idea that roads can take an infinate amount of traffic as long as people don't do dumb things is a myth.
as soon as people see the word "beta" on the side of the cars.
While it may not be able to make higher-order decisions, it does have sensory organs that humans lack. With radar and infrared detectors, this could detect the invisible deer to the side of the road and make nanosecond decisions, unlike the 100 microseconds it takes a human to react to anything.
I think we are likely to see this adopted and put into wide use before we see people start really warming up to voice activated text messaging. Few things outrage me more than a driver staring at their lap while they are "driving." At least with autonomous vehicles someone is paying attention to the road.
How will this affect the automotive insurance industry? If I have a car that does all of my driving then I should never have an at-fault collision. So it only makes sense to me that my rates should be a fraction of what they are now. If the insurance industry sees this as a potential threat to their profits then I think they will lobby hard to keep it from advancing.
I completely agree with your arguments, I think it is the unseen, anticipated events that will make this system less safe than having a skilled human driver.
That said, no human driver is likely as consistent as such a system could possibly be. But I agree, the 'bugs' that would become evident if such a system were immediately deployed would become cable tv fodder from day one.
For these reasons, I think carpool/HOV lanes should be replaced with automated driving lanes. Start with long distance travel using road trains/grouped motorcades, with human intervention on entrance and exit, and I think this would be a good place to start rather than going straight to automated in-town commutes. It would do much to alleviate public concerns little by little. I want significantly more testing than 190,000 miles by a few cars before I entrust my life to an automated system.
My $0.02.
I work on a volunteer project at my company. The actually bought us the hardware similar to that on the Google car.We are no where close to the Google car but it is fun side project in my free time.
My biggest issue with this coming to the general masses is the cost of the sensors on the vehicle.
The velodyne last I check was 75k. http://www.hizook.com/blog/2009/01/04/velodyne-hdl-64e-laser-rangefinder-lidar-pseudo-disassembled
If they are using the same GPS used in the urban challenges those cost 130k
I would love to see more work on using cheaper hardware and improving the software to work with the cheaper COTs hardware.
Sebastian Thrun of this course Instructor and Google's Eric Schmidt's reputations are built upon crimes, conspiracies and censorship on my public challenges.
Sebastian Thrun of this course Instructor and Google's Eric Schmidt and are part of these plotted murders. And that innocent Stanford girl May Zhou was murdred by people on their side for their sake during their fight with Stanford Authorities over a criminal case which happened on Stanford campuse in 2004, in which an irrational and brutal lady named Gabriele Scheler bitterly assaulted me in the lab and then falsely accused me for sexual assault trying to escape facing the legal consequences. Eric Schmidt and Sebastian Thrun joined with this criminal suspect Gabriele Scheler to fight against ruling from Stanford Authorities. They had plotted murder on May Zhou to threaten me and to terrorize Stanford; and later, when they found I would not compromise with them but actively pursue the case further, they had plotted a murder on me as well. They didn't make it a reality in my case because they are closely watched by police and they fear leaving evidence (not because they have any mercy on me as a human being). Proof of real names, dates, photos, detials of events along with a police case number are listed in my blog link [ http://tysurl.com/BsEnQ4 ]. Eric Schmidt and Sebastian Thrun had not paid for their crimes and they would have to.
--- See if Sebastian Thrun and Eric Schmidt dare publicly deny anything I said here.
The video talks about how they want to drastically increase the number of cars on the roads since the computers can drive very close to each other.
Sorry but we don't need more cars on the roads. What we need is to stop our crazy unsustainable growth in traffic.
They mentioned that the traffic is growing at a rate of 2.9% per year. Doesn't seem like much right? Well that's an exponential rate and it means that traffic will double every 23 years. So first they want to fill the entire surface of roads from 8% to 16% full of cars, then 32%. By then people who want to drive themselves probably no longer can and what have you gained? 40 years? then what? 4 more decades and you've got 100% of road surface area on the planet occupied by cars at all times anyway. I enjoy driving, don't start making plans to make that enjoyment go away.
3% growth in any aspect of society becomes unsustainable very quickly. That's the nature of exponential growth.
Liberty.
This will help my commute to work, I walk, knowing that cars will stop if I walk out in front of them!!
I like how whichever idiot made this has his spambot pass of CAPTCHA challenges, made it set a homepage and sig, but never set it to use HTML instead of BBcode.
"When information is power, privacy is freedom" - Jah-Wren Ryel
"The goal is that the technology will help reduce congestion, fuel waste, and accidents" ... and still allow people to sit in their own private metal cages without having to deal with anything icky like other human beings.
No, but it still has a brake pedal. You should probably be paying attention to the road while in one of these things, and intervene when you deem necessary.
You sound like you have exceptional (compared to most of the population) situational awareness skills.
And this is precisely why we need autonomous cars.
nb: Your laws may vary.
Just make sure you don't run stop signs or give way signs, without actually stopping and giving way to other traffic. Don't overtake someone on the side they're turning. Don't run your bike into an intersection AFTER traffic has entered, leap on it after coming out of a turning lane while running straight through, and then expect them to give way to you. Don't ride through red lights and expect all traffic to give way to you. Don't cut off oncoming vehicles because you don't want to wait - they have the right of way, and you'll probably die. Don't ride through crossings with pedestrians on them - they have the right of way. Don't cut around parked cars, then cut in front of them expecting them to give way to you even though you've not signalled your intention to merge. Don't ride against traffic, even for a little bit. Don't zig-zag across an entire lane because you're going up a hill, causing all the traffic to back up behind you while not making allowances for them to pass. Don't ride on footpaths, and especially don't cut corners because you might hit pedestrians who can't see you. Don't cut around heavy vehicles that are backing in a controlled intersection. Don't ride in the blind spot of cars, under one metre from the vehicle itself - if the driver has to take evasive action, you will> hit him and it will be your fault. Don't cut through controlled railway crossings when a train is coming. Don't cut through railway crossings when road control staff have ordered you to stop. Don't cut through driveways and expect people to give way to you so you can get back on the road. Don't ride in high speed zones unless you're capable of travelling at the high speed. Don't walk your bike forward through traffic and then be surprised that drivers are getting pissed at you for preventing them from moving because you're obstructing them. Holding the handlebars is not an excuse to skip signalling that you want to turn. Having a walkman on is not an excuse to not develop situational awareness - don't turn unless you know your way is clear.
I'm sure I'll remember dozens more later.
You are an awesome driver. I say we kill all humans that cannot do what you say you can do.
Or you are a liar. And a coward.
I haven't heard anything about how these driving systems might behave around each other. They've spent all their time building them up to work in environments with lots of human drivers. What if these systems pick up and there are eventually very few human drivers left?
A.C.: All very valid and thoughtful. It could turn out that it reduces the NUMBER of accidents while also changing the DISTRIBUTION OF CAUSES. If that happens, it may be difficult for people to wrap their minds around it. For example, the number of accidents caused by distracted driving (cell phones, attractive people on the street) could drop essentially to zero while the number caused by missing cues that seem obvious to humans, could rise. If the overall number dropped, but with the horror/tragedy factor increased (e.g. a rise in ball chasing children struck as in your first example), would societies accept such a turn of events?
Like:
Can it, like I do, notice that the baseball rolling down a driveway may be followed by a child who is currently invisible behind a parked SUV?
Can it, like I do, notice that the driver *behind* me is distracted by her cell phone, has started late at the last three lights, so I should give myself more than average room between me and the car in front of me, so in case it stops suddenly, SHE won't have to stop as suddenly and will be less likely to rear-end me?
Can it, like I do, notice that even though the road has been clear of ice and snow, the next curve up ahead is deeply shaded and is likely to be slick?
Can it, like I do, notice that the baby deer is one one side of the road and his mother on the other, and even though he isn't charging across, it looks like he's about to do so, so I better slow down? Because this happened to me not one week ago, and it DID charge across.
Can it react to the highly dynamic and unpredictable world in ways that require human intelligence? And before you claim it's going to be a better driver than I am, note that I got my license in 1973, and have YET to have any sort of accident. Because I can provide human intelligence to the task of driving, and unless you want to claim the machine is as smart as a person, I am not so sure I believe I want to leave these choices up to it. I'll gladly fly on a computer flown aircraft, because that's a very, very different kind of task. No children appear in the sky from behind parked vehicles, and the pilot stands by the entire time to take over if something doesn't go right. Car drivers won't be that alert: they'll just doze off or play with their phones and not even glance out the window the whole time.
To all of your questions, I would say 'Yes'. I can see no reason why an autonomous car could not be programmed to handle those situations. A major difference, though, is that as you gain more experience you become a better driver. As the autonomous car algorithms are improved, all autonomous cars become better drivers. New autonomous cars, unlike new drivers do not require experience before they become good drivers.
These are all very important points. I think there are some examples of them in the video.
For example at 9:00 the speaker talks about how proud they are that their car didn't run down some pedestrians crossing the street. The car stopped just meters from them while making a left hand turn, blocking an oncoming car. The higher level decision that I'm sure the parent would make in practice is to judge the intentions of the pedestrians on the street and not commit to the turn. Although it's not clear in the video, it looks like the pedestrians didn't even run the light. The speaker doesn't acknowledge that their car blocked the oncoming lane resulting in a fairly dangerous situation. As you can see in the video, the oncoming car eventually swerves around the Google car.
Later on, at 12:00 he jokes about how a car nearly sideswipes him while merging onto the freeway. The speaker doesn't acknowledge the Google car's role in that dangerous situation. It's a simple lack of defensive driving on google's part. Yeah, they are not at fault for driving down the freeway, but it's not hard for a human to anticipate the fact that a car in a merge lane tail-gating behind a big semi-truck is going to change lanes at any moment. The Google car had an empty lane to the other side that it could have changed into at no cost.
I think these driverless cars are an insanely great innovation for our society, and I was really impressed with their TED talk too. But I just wish that now that they have the nerdy computer technical issues resolved, they can work on higher level algorithms, perhaps consulting with defensive driving professionals. I think defensive driving algorithms would be pretty interesting too, involving game theory and optimization. They seem to get it right with their four-way stop video, where they say they had to drive forward a bit to show their intention to proceed.
"Follow me" the wise man said, but he walked behind.
It doesn't have to be better than the best driver. All it has to do is be better than the AVERAGE driver to save tons of lives per year.