I'd be interested to see how it made progress somewhere like Sicily (Naples is meant to be even more exciting to drive in but I haven't been there) or Thailand or India.
[Note that I haven't visited Thailand for about 20 years, Sicily for 10 years and India ever so perceptions may be out of date or incorrect based on media/TV coverage for the India case.]
I was in Palermo not all that long ago (2009) and I did wonder how self-drivers could ever navigate the cluster there. In my trips to Seoul, I could see them doing okay on the main roads, but once you get off of those things get really tight and irregular very quick.
Heck if a self driving car can get through some of the old parts of Boston and Philly I think they'd have a chance. Hell, if they can navigate D.C. with all the crazy turns that change depending on the hour of the day they definitely have a chance.
The single biggest problem for autonomous cars is that they have to cope with rules and systems designed for idiot humans. If you exclude that driving is simply a problem of finding a reasonably continuous plane in roughly the direction you want to go.
It's an interesting problem that exposes the fundamental difference between humans and algorithms:
While some humans may be able to adopt to driving in Sicily or Thailand after the U.S., algorithms will never be able to do that. Unless, of course, specifically designed for driving in the U.S., Sicily and Thailand. (Or Russia for that matter, where driving is barely subject to any formalization.)
Humans can learn and adapt, it's in our nature, but an algorithm is just a reflection of some narrow aspect of what we, humans, have learned. Change a few external variables and your algorithm fails miserably.
I'm sorry, but somehow I don't believe in the future of self-driving cars, just like I don't believe in algorithms that would do image or voice recognition as well as we do.
I don't want to be rude but I don't think you know very much about self-driving cars. They are much more advanced than you think. Also you talk Thailand, I guess you mean the free-for-all situation ... Well, humans would adjust their behavior to the reality of self driving cars but in reality, these cars are much better at these situations than you think. You greatly underestimate the state of the technology even in its current form. We're talking the possibility of reducing 95% of accidents and fatalities. The sensor perception and trajectory calculation is much better than what humans can do. Driving is not about solving natural language riddles. You don't need "strong AI."
Oh I don't blindly believe what Google says either. Google is hardly the only one in this game. There many analysts in this field thinking the same thing. This is already going to affect toll roads being built in the next few years regarding risk propositions. The risk proposition will change. Go and talk to people at the Transpirtation Research Board next year. This is not just Google starry eyes but very serious.
I don't understand, what law of nature do you think there is that prevents algorithms from being able to emulate human intelligence? What magic sauce is there in the human brain that means it is in principle not replicable by computers? What do you think an algorithm is?
No law of nature prevents algorithms from being able to learn like we do, except (1) I don't think Google's (or BMW's or Volvo's - whatever) algorithms are even trying to do that, and (2) algorithms that learn like humans haven't been created yet.
Ah, ok. I of course agree with you about the narrow case of Google's current self driving car initiative. It's likely they don't yet have the capability to adapt to the varied conditions a human could, and won't for some time.
I suppose I was thrown off by your phrasing "fundamental difference between humans and algorithms" and "algorithms will never be able to do that" and "I don't believe in algorithms that would do image or voice recognition as well as we do".
In fact, it seemed to me the entire point of your comment was to take the specific limitations of Google's current algorithm, and use it as an example of a general principle that algorithms are fundamentally limited in ways humans are not. If that wasn't the point of your original comment, what was?
I think I'm still technically correct in that no algorithm can imitate humans, because no such algorithm exists yet. I should have said "humans and existing algorithms" probably.
Google's approach to the problem has nothing to do with the (nonexistent) "A.I." as media likes to present it, it's just incredibly complex software that's trying its best to tackle the problem of navigating the roads, recognizing signs, other vehicles, etc. I can imagine it's an enormous effort that doesn't guarantee success. See "Siri, Call Me An Ambulance" meme as an example of a comparable effort ;)
But the point of the self driving car seems to be to dehumanise the driver to make it more predictable, reliable and less distractable. Create too good a copy and you will need to do something to stop it getting bored and distracted.
It is quite possible that the learning abilities come along with downsides.
I agree but it might work if a vast majority of cars are automated ones.
I can imagine a transition with "smartcars" : you drive with help and advice from your car.
Yes but going through a city in Southern Italy how often is it going to say something other than STOP! If you don't assume that other drivers are going to change course and avoid you won't make much progress.
The skill and alertness of the drivers in Sciliy were much greater than in the UK and the US but they had to be as those were relied on more than rules and systems.
I also wonder about what human driven cars, motorbikes and bikes might start to do once they know and can identify that there are nearly uncrashable and will reliably take avoiding action - will they learn that they can just pull out and let the computers avoid them?
Ok, cool. So insurance companies should be lining up to insure autonomous cars, ya? As soon as an insurance company will insure an autonomous car then we will be able to buy them and use them. Or maybe the autonomous car makers will accept risk? This will be great! Cause the only person who wont have to have insurance is - me, the driver! Er... passenger.
Insurance companies won't care if you're driving yourself or using a self-driving car.
They price their premiums so that they always make a profit, regardless of the risk.
That's why drivers that have accidents pay more than drivers with no accidents and young drivers (that are statistically more risky) pay more than not-young drivers.
Insurance companies will simply adjust the cost of insurance for self-driving cars to account for risk based on past data. If those cars will turn out to be involved in less accidents, the insurance will cost less. More accidents => cost more.
Even better, Google and the automaker's could get into the market of managing fleets and their own insurance directly or via wholly owned subsidiaries. That's what I would do if I were in their position. Basically, this is going to completely disrupt the auto insurance industry as well and no one has more and better data to price the insurance than Google that now has hundreds of thousands of miles logged and will always be in possession of the sum total of all data on all miles ever driven. They are no longer performing statistics on a sample size, but are instead doing so based on an entire population.
Basically Google can do the same for auto insurance that Amazon has done for online retail. i.e. they can operate on razor thin margins that no one else will ever be able to match because they'll be operating on sample-based statistics and it will take them a while before they have a significantly large sample to properly price their product.
Is no one going to point out the fact that the data came from Google themselves? I mean, I'm excited about autonomous cars too, but what happened to taking everything with a pinch of salt?
This is a good point. They compared the braking and acceleration of the vehicles vs. a human driver. A human driver supplied by Google. Nevertheless it's clear this technology has immense near-term promise.
It's also interesting to think about the financial implications for localities as data shows that the average police officer issues speeding tickets for approximately 300k/yr. (http://www.statisticbrain.com/driving-citation-statistics/)
I'd like to know what those autodrivers do for failsafety. What about software crashes? bugs? sensors giving bad readings? intermittent loose connections? chips failing?
They should at least talk to the engineers who design airplane autopilots, who know how to deal with that stuff.
In that same vein but perhaps more subtle, what of the social failsafes? I keep imagining this scenario (one which I run into consistently in the Bay Area), where I am looking to merge into the lane to my left because my lane has ended, and nobody will fucking let me in. Does the automatic driver have a level of aggressiveness that it follows? What if two automated drivers have the same level of aggressiveness and one wants to merge and the other won't let it? Do they just stop, resigning themselves to an endless loop of near catastrophe because neither will yield? I'd love to know more about exactly by what driving conventions these automated drivers abide. I think it'd go a long way to help people understand how such a process could be automated at all.
I'd guess that if both cars are automated, there will likely be some type of vehicle to vehicle negotiation process between them to determine who has the right of way.
What do people do when their cars fail or there are drinks behind the wheel? I agree with what you're saying sort of but there are very serious people working this technology (it's not just Google) but there will be accidents but much less than with human drivers. It may seem fantastic but if you actually look at the current state of the technology and even conservative projections by people like the Economist ... It's quite impressive.
Exactly. This sort of thing is what embedded programming is all about. "Real Time Computing" is the relevant keyword.
Just like a frontend dev learns HTML/CSS/JSframeworks and a backend dev learns JSframeworks/Node/(no)SQL, embedded developers learn about creating real-time, fault tolerant, highly recoverable systems.
Tell an embedded engineer that he should consult with Boeing and he'll give you the same strained smile that you would give a family member who suggested consulting with Amazon/Ebay/YouTube because "they are good at making websites."
I used to write embedded firmware for factory machinery, and I can tell you that there are a few obvious nuisance bugs in my car's firmware. The kind of things that probably wouldn't be present if they had even owned or driven the car.
How does privacy factor into this? Doubtless these cars would be even easier to track than modern ones. But could they be hackable or overridden by control to be driven someplace?
The code behind this will have to be completely open sourced to allay all fears.
I don't doubt for one second someone in the US government doesn't already have plans to backdoor, take over and turn an autonomous vehicle infrastructure against people if they deem it necessary for "the greater good." Imagine one day your car decides you need to drive out into a facility out in the desert for "questioning..." or suddenly a group of them get commandeered as an ad-hoc roadblock, or any number of ridiculous but maybe too-tempting to resist possibilities, which people couldn't escape should these things become a dominant form of transportation.
Well, just wrap your whole body in aluminum foil ... Is this serious? Also what stops this from going into non autonomous cars, the tracking bits, anyway?
Oh my, really? The Feds could just track you down and grab you. Why is this worse? I agree, in some fantasy dystopia this may seem scary but ... If this ever happened, I mean, we would have much bigger problems than the self-driving car.
Come on- isn't the prospect of being in a personal transportation that could override your controls not scary at all?
And remember, it isn't just potentially the government who might do this. What about hackers working on behalf of some malicious and sinister agency? What if instead of kidnapping you, they mess with the controls and cause you to drive into incoming traffic? Or worse, an entire highway's worth of cars in a terrorist attack?
I'd be first in line to buy an autonomous automobile, but these are still real concerns. The more control you cede to programming, the more rigorous the code has to be.
Who are these sinister organizations? Chaos? E.V.I.L.? Or the FSB or what?
A terrorist attack? Wouldn't you be more worried people going crazy and driving down the wrong side of the highway?
By your thinking maybe we shouldn't even have cities because they are dense target rich environments. Even under an attack by Loki or whatever sinister organization from a comic book, you're still talking about something far safer than what we have currently.
I think it's hilarious that HN is the first place to see people saying the Boston Bombing was no big deal and more babies are killed each year by peanut buffer or something but yet here we have people more concerned about Magneto controlling cars and causing terrorist actions than reducing the very real 95% of traffic deaths. That's pretty hard to take seriously!
I never said that we shouldn't have self-driving cars. I'm saying the code that govern these cars should be open-sourced. To prevent governments (or al-Qaeda or the mob or someone else with the resources to get hackers) from having an advantage in figuring out ways to abuse these systems.
We already live in an age where pacemakers with wireless components can be hacked, and potentially kill the people who wear them.
Call me crazy, but I consider a future of self-driving cars to be a fantasy dystopia in any case.
I find it odd that people seem to be upset with governments intruding on their privacy in a purely abstract sense, but seem not to have issues with potentially giving up their physical autonomy completely to a system run by software (and potentially, by extension, a government.) Why be worried about the effect of a global internet surveillance apparatus but not be concerned about that being extended to the realm of vehicles, or the effect that that might have on personal freedom?
I did not get any new information out of this article. Not only does the title basically tell you everything that the article is going to talk about, but the statement is obvious. Of course an autonomous driver is safer than a human.
I think this was the main thing that hasn't been shown with data before: "One of those analyses showed that when a human was behind the wheel, Google’s cars accelerated and braked significantly more sharply than they did when piloting themselves. Another showed that the cars’ software was much better at maintaining a safe distance from the vehicle ahead than the human drivers were."
They were known to be very safe with a very low (almost non-existant) accident rate, but I believe this was the first presentation of some of the more detailed driving data such as acceleration and braking.
There are real legal questions on what happens when robotic cars are really ready. For instance, what happens IF the vehicle actually does have an accident? Who is liable? Is it the "driver" who wasn't actually controlling the car, or the company who wrote the driving software? There will be several hundred ambulance chasers ready to sign up victims of any accidents in order to get a class action suit against Google.
While that's true, it's not as big a deal as you'd think if these cars are dramatically safer. Auto insurance is a 180 billion dollar industry in the US (that's a billion with a B). If the cars are actually safer, they will have dramatically lower insurance premiums, which will allow the manufacturer to charge more money for the car upfront and still come out ahead even if they assume lots of liability. People will find a way to make it work because there's just too much money to be made.
Good point! It could actually be that with lower accident rates robotic cars will produce, automobile insurance companies actually help drive the change.
There are very serious people working on this. The reality is there will be fewer accidents. Some people think think the manufacturers themselves will take on the insurance liability, eventually "disrupting" the auto insurance industry completely. Sure, there will be media hay made out if any accidents in the beginning. It's like anything. It will be hard to deny the safety of this mode of transportation. Actually, the most unsafe aspect about this will be combining human drivers on the road with efficient autonomous driving systems.
> For instance, what happens IF the vehicle actually does have an accident? Who is liable? Is it the "driver" who wasn't actually controlling the car, or the company who wrote the driving software?
Why would it be exclusive? If a vehicle actually has an accident with a driver, and owner who is not a driver, and a manufacturer, then, depending on the circumstances, it is possible for all three to have some degree of liability. With a robotic vehicle, there's really not that much difference.
"And existing product liability laws make it clear that a car’s manufacturer would be at fault if the car caused a crash, he said. He also said that when the inevitable accidents do occur, the data autonomous cars collect in order to navigate will provide a powerful and accurate picture of exactly who was responsible."
Predictably many people are skeptical. But just stop and think about how many times you've caused an accident or near miss, ignored a road sign, got stuck while merging, etc and ask yourself if the computer would have done a better job. The new buzzword is "zero fatalities" and it's a far too important goal to just dismiss with typical I'm-a-better-driver-than-any-dumb-machine attitude.
Yep, it starts. I happen to really love driving, clean record, etc. In 10 years, it's going to be getting touchy for folks who still prefer driving themselves rather than letting the self-driving cars do it. The freedom will be exchanged for safety, and nobody will blink.
The cost will be prohibitive for most people. And in the USA, the car culture is too ingrained. I don't see autonomous cars and human driven cars on the same roads. I don't think the technology will be ready in ten years anyway.
No i think you're wrong about that. Even conservative estimates put it between 2018-2020. Also I don't think it's so much that people would be buying these cars. I think you will see different models, club cats, zip cars, etc. Sure people will buy them but I don't cost will deter then being a significant percentage of the total transportation in 10 years.
I have to say, you are probably 100% wrong on all counts. It's all opinion of course but once these are on the road and they work they will spread quickly costs will come down, and culture will adapt. The value prop is just too high, it's like any other computerization: sure, there are some charming aspects to past practices but the machines always win.
Why wouldn't they be on the same roads? Everyone will gain from having autonomous cars on the roads, both people riding them and people driving their own cars.
I like driving too, sometimes. Sometimes it's awfully boring and I wish I could do something else while on the road.
No it doesn't discuss other self-driving competitors unless you think of humans as competitors. Let's get real, very soon self driving cars will be tremendously safer and more efficient by many factors than human drivers.
If the word "data" in its context may be replaced with the word "information", then it is being used as a "collective noun" and may be treated a singular. If it would be better replaced with "the individual items of information that we have collected", then it should be treated as plural. Which seems right in this case, "Information shows that Google cars...", or "The individual items of information that we have collected show that Google cars..."?
I would read it more like the former, that "data we have collected" is similar to "information we have collected". Alternately, data is also often a synonym for "set of observations" or similar, a shorthand term for "data set", etymology notwithstanding.
A common exchange: "Can you send us the fracking data?" "Sure, once I get approval I'll send it over." If you instead said "send them over" in this context it would suggest, at least in typical usage I'm familiar with, that there were not only multiple data points, but multiple data sets being set over.
[Note that I haven't visited Thailand for about 20 years, Sicily for 10 years and India ever so perceptions may be out of date or incorrect based on media/TV coverage for the India case.]