Nissan says self driving cars are impossible today
wired.com
wired.com
On the road, an actor's actions do not merely affect itself, but all the others on that road. If you slow down at a point, that can cause the driver arriving at that point five, ten minutes after to stop. On the road, everything creates a butterfly effect. And just like you can't run around naked or listen to very loud music in residential areas, you can't just get on the road with your new toy which if fails, does not only put your life at danger, but also many others'.
It's not hard to find places where a 40min commute in a car turns into 2-3 hours using public transport.
Public transport is also not as reliable in places with extreme weather especially cold.
Have fun waiting for a bus in an open bus stop at -10c...
Certainly in rural areas and in wider peripheral suburban areas public transport is less useful. But in actual cities (London, Paris, Istanbul, etc.) with a high density of people and their residences, public transport is the way to go. I don't know much about American cities, but if we got a comprehensive and functional system of metros, buses and boats here in Istanbul, where some of you would classify as a part of the third world, I guess you are capable of doing it in the US too.
London is probably a "bad" example, I live in London and sure the public transport is probably in the top 5 in the world, so is Tokyo and Paris. But what many cities with good public transport have in common is that they had public transportation for a long time, were designed / rebuilt with that in mind when it was manageable and are relatively small for their population density.
I would not want to be the one that has to design a public transport system for a sprawling city like LA where the greater LA area covers nearly 90,000 square kilometers for a size comparison the entirety of England is 130,000 sq/km.....
https://www.bloomberg.com/news/articles/2016-08-15/uber-and-...
- Walk to the next village
- Take a bus to the next city (bus comes by every 2 hours)
- Walk to the train station
- Take the train to the next city
- Take another train to the next city
- Walk to the outskirts of the city to where I work
It would also take me several hours every day.
My car costs me maybe a maximum of 90€ every month and takes me to work in about 15 minutes.
I know what you will say. "Just live in the city and use a bike to get to work, faster and costs nothing! (apart from the bike)". But that would not only reduce my quality of living (let's face it, the countryside is nicer) but also let my rent costs skyrocket (average price per m² more than doubles).
Well I'd actually not say that, because I'm not really saying "kill all the cars". What I'm saying is that "in most cases most probably public transport is equivalent to if not better than autonomous vehicles, so the effort and the risks are not worth them". I live in Istanbul and my school and my home are 30-20 kilometers apart on a straight line. When there's no traffic it'd take ~30 minutes to drive there, and at my usual times it'd take 1-2 hours with traffic. Via public transport it's 30min fixed + 5-10min walking. And in that time I can have my morning coffee from my thermos flask and eat a simit, and I don't need to be alert at all (I'd probably be dead by now if I drove in the mornings).
So TL;DR: for me, I'm perfectly fine w/ your commute, I just say that self-driving is not really a necessary and fruitful burden, however cool it may be, in solving traffic jams and traffic safety problems.
I don't disagree with that sentiment but there were all sorts of things that "PCs" couldn't do that real computers could and they were still commercially successful (and eventually caught up to "real" computers for the most part). I could imagine that someone could come up with signage which self driving cars would understand and act on. Basically put up a QR code that links to the government approved way self driving cars should navigate the hazard, and of course a programming language that could communicate that to the car.
Of course you would not want just anyone to set up overrides or you would have pranksters sending cars off into corn fields or worse.
The part that Nissan appeared to miss is that if you can make them work in 99.9% of the cases, then you can hack around that last .1% and the cars will still be a net win in terms of better road behavior.
I'm not trying to be hyperbolic here, those really are the stakes with self driving cars. The problem is that when consumers think they are getting a self driving car they will very quickly resort to bad behavior (see the terribly sad Tesla fatality from last year.)
We really need to be careful about how these features are marketed and implemented.
The edge cases are not going to be adverse driving conditions, which machines actually handle better with superior reaction times and finer control.
People still feel that "it won't happen to them" because they have agency and can drive safer than others, it isn't neccrary true but you give up on this agency when you let a computer drive.
You can drive in a way which will reduce you likelihood of being in a fatal accident if all autonomous cars have say a 2.5% fail rate on decisions it becomes just a matter of time.
Autonomous cars aren't programmed, they're trained, like humans. So I don't think this reasoning applies. If they behave poorly in some circumstances, they'll just create new training simulations covering those cases with better outcomes and push an update out. The cars will learn better and faster than humans.
It will never be able to handle unique and novel cases because it cannot think. "AI" isn't intelligent, it is a dumb bureaucrat executing instructions.
I am in the apparently unusual position of believing that autonomy is both feasible and desirable, while thinking that the capabilities of current implementations are being oversold.
I hear this argument often, but I see no evidence for it. It just sounds like yet another flawed attempt to make humans more special than they really are.
I guess we'll see over the next 5 years as autonomous vehicles become deployed.
The rules of the road define who has the right of way. Of course, sometimes other drivers don't see you, but you literally have no way of knowing whether or not they do. The only thing you can do is begin your attempt cautiously in accordance with road rules and abort at the first sign of potential collision.
Notice how this decision process is completely algorithmic. Proper understanding of another driver's mind is a red herring. You use the almost-collision to infer what the other driver knows, not the other way around like you seem to think.
Now, I do not doubt that if all (or most) vehicles were autonomous, things could work as you describe (though traffic might flow more smoothly or efficiently if vehicles could signal 'intent'), but that is not how it works at the moment, and the transition is probably the hardest part of introducing autonomous vehicles.
Sure, after they do something seemingly stupid that you don't understand. Inference follows observation, not the other way around, and autonomous vehicles are observing all the time with reaction times much faster than a human's. Even if an autonomous vehicle weren't able to understand high-level goals, it understands and detects vectors much better than humans do, which is all that's really needed to create and adapt a plan that avoids collisions.
> I am very confident that people weigh the presumed motives and purpose of other drivers in their predictions of future outcomes
And I'm very confident that even if this were true, it's largely irrelevant to being a good driver.
So just like most cars are to be driven on paved roads, reliable autonomous cars will need to be driven on roads with good visibility of lane markings. And road construction or road hazard cones will need to be electronic, so they can give the car computer unambiguous instructions (such as "shut down until driver manual control is confirmed").
If the computer is confused I want to be the one to make the decision. I am perfectly happy having a system right now that works 95% of the time in good weather.
I think it will be at least 2025 before it can handle a Michigan winter, maybe longer. Still want a limited system ASAP.
Sounds like a Tesla Model 3 reservation is in your future! ;)
1. Thanks to the car dealer lobby here in Michigan you can't buy one in the state
2. No local charging options. Nearest supercharger is 70 miles away in either Ann Arbor or Grand Rapids
I'm not very far from my local Ford dealer and the rumor is they're getting a charging station so that is probably how I will go.
Can you not charge at home with a 240V outlet? You might also check out PlugShare.com for local charging options.
If you're in a condo or apartment, several states have legislation now mandating reasonable accommodation (at the EV owner's cost) of EV charging stations. I do not know if Michigan has such legislation though.
This makes a lot more sense for a fleet of self driving vehicles. Not for personal transportation in your own car.
That being said, the "self driving" part of my car (cruise control) turns itself off when the wheels slip already, so I agree that I could handle a car that drove itself in good weather but punted back to me in the snow. Vision-imparing snowfall is actually a more rare event than people might think. Normally the roads are quickly cleared.
A lot of drivers have problems driving in winter conditions. But most? Over 50%? That's a pretty strong claim.
That's not to say sometimes people have unreasonable assumptions as well. That's why I say it's not clear to me.
Complaining about people driving in the snow is what gets us northerners through the winter months. It keeps us warm.
This is not the kind of system they're talking about at all. From the article:
> Nissan’s cubicle-based drivers aren’t emergency backups. If the car hits black ice, it’s in charge of staying on the road. There’s no feasible way to get the human into the loop in time to act. [When encountering an unusual scenario, a] human operator would look around using the car’s cameras and other sensors and issue new instructions—direct control would pose latency issues. Like: When it’s safe, cross the double yellow and get back to the right side after 20 yards.
There is a gap between Level 2 and Level 4 (arguably Level 5 with the new standards, but I'm going to assume Level 4 means what most people would call a true self-driving car) autonomy which is dangerous, because it lulls humans into a false sense of safety. And then when the car can no longer drive itself, the user is expected to take over with a delayed reaction time and little to no context in what could be a very dangerous situation. It is a deadly valley. There are numerous studies on this topic from the aviation industry, and drivers don't have nearly as much training as pilots.
I think real self-driving cars are awesome, and the first time I sat in one years ago when I worked on it as a PhD student was truly an incredible experience. But the hype around the technology is too much, and when investors push for a product in a few years when it's not ready at best companies will fail and at worst people will die. Misinforming the public on this is a very serious problem which nobody seems to care about, even after at least 1 death which is still under investigation.
I am all for working on this technology and seeing incremental improvements over the years. But I am frustrated with the media's portrayal of how it's just around the corner (I am frustrated with the media in general), and I believe the overpromise of the technology is largely responsible for the death of the Tesla owner in Florida last year (and there are other accidents that are reported in China which get much less coverage).
People will say self-driving cars save lives and are safer. In theory, yes - but the problem is we aren't there yet. There just isn't enough data to make that claim. When Tesla said it was safer, they ignored the fact that all of their driving was on highways (which are statistically safer) and that their owners are more affluent and older (at least I assume the average Tesla owner is 25+). The data isn't available when you break down these demographics, but I suspect if it were that 1 death would actually make autopilot deadlier than manual driving. It's a moot point though - anybody trying to extrapolate meaningful conclusions from a single data point is making a ridiculous claim.
Autobraking for accidents only, is a "self driving" feature that you can buy right NOW on most 2017 cars.
Auto braking alone could save 10s of thousands of lives every year, and it is absolutely production ready.
As a thought experiment, just imagine how every daily activity is improved by self driving cars; not just relaxing while driving to work, but automatic transportation of children to school, removal of the annoying of finding a parking spot at a grocery store. Social activities are made much more convenient; having drinks with a friend, picking up parents from the airport, lending your car to the neighbor. And there is the potential of a world in which car ownership is itself redundant, and we are only speaking about lag time and comfort for transportation.
The technology has been around, but it isn't deployed to the majority of cars yet.
What I am saying is that self driving technology doesn't have to be perfect for it to change the world.
It doesn't matter if we are a ways away from getting to the "nobody owns a car, and everyone just uses self driving Ubers" future. Because the current, shit accident prevention technology can ALREADY change the world, even if it is "just" by cutting accidents in half and saving "only" 40,000 lives every year.
Incremental improvements that save lives and make driving better are fantastic. Overselling a feature that can kill people is not.
Also, the posted title is clickbait bullshit and not the actual title of the article.
The thesis of the article is that the promises made by these companies -- and the promises are being made -- are not quite feasible in the short term.
So what kind of car would not have some sort of manual override? The car that all these companies are promising!
There is something deeply unpleasant about underestimating people and underestimating the problem is generally not a great way to solve it.
In the real world hundreds of millions of drivers have been driving on extremely varied roads and traffic conditions all over the world for decades. Given the sheer scale and volume of drivers, conditions and cars involved it appears self driving proponents are massively overplaying the safety card and worse with little to no data for the other side.
Any ai vehicle that requires new kinds of roads or contraints is a completely different ball game and cannot be compared to human drivers. Trains are far safer and in many cases faster than cars. Yet the train story in the US has remained dire for decades making ai safety proponents look self serving.
Technology does not exist in a vaccum, it is to benefit us and things like cars are tecnological as much as social and political. A car gives you basic transportation but also the freedom to go wherever you want. Having ai cars, constraints for them and control that could limit mobility and freedom then become more than techology issues in which the voice of society becomes far more important than the people creating the technology.
We don't even want autonomous robots in our house, pitter pattering about like toddlers, at 3 miles an hour. We don't even trust our smart TV sets.
So let's make 2 ton versions, and throw them into a frying pan with people? Yeah, great idea. I bet they'll act just like TCP packets. Good job.
Why not segregated roadways? If we're so desperate to automate trucking and shipping, how about big, thick K-rails, and dedicated roadway?
We all know that early adopters suffer the growing pains of buggy first versions. No reason for all of society to share this burden when It hasn't even been put to a vote.
Keep these things the hell away from ordinary people, and their kids for at least another 50 years.
No one voted on if they wanted cars in the first place. I'm not sure if you're aware of how incredibly dangerous cars are, but I can tell you without looking up the numbers that if we were to vote on it today with an unbiased viewpoint, there's no way anyone would be allowed to drive a car.
I'm sorry, but I feel like you're going to be in the minority on this one. You seem really scared. I'm not trying to change your mind, but I would like for you to take a look at traffic fatality statistics and and think hard about whether or not it's worthwhile pursuit to try to bring that number down.
I would like those people who take it as an article of faith that self driving cars will cause fewer fatalities to watch closely where the liability falls.
If executives go to jail when the software crashes the car and kills or maims somebody and they let it out on the roads then I'm happy.
If instead those companies get slapped with token fines, I'm pretty sure that they're going to push out faulty software before it's ready and kill a bunch of people.
Politically it's still up in the air but currently it's looking like there won't be any criminal liability.
A correction: they're the top-selling robotic vacuum cleaners. Traditional 'dumb' cleaners still outsell them by magnitudes.
But since I can't back up that claim, I'll give it to you.
Yeah, I think this is the only real option, with the eventual goal of slowly phasing out non-automated driving altogether.
But the answer is, we already have segregated roadways for self-driving vehicles. They're called rails, and trains drive on them. People still drive cars, surprisingly. Because if that segregated roadway doesn't go someplace that a normal road does, people will still drive their cars.
Autonomy is an ambitious project. Large heavy machinery is an unnecessarily risky entry into an immature field.
Building reliable technology in a mild environment will permit the opportunity to prove trusted patterns of operation, under the safest conditions. People will be able to gain a feel for normal dispositions of autonomous motile devices, and build good collective practices for safe operating environments.
Corporations should not be trusted as accountanle entities. Volkwagen's deceptive green emissions scandal shows how far a sub-department will go, to meet the bottom line.
Competitive business practices in a hazardous and unforgiving operating environment are not what we want birthing this sort of technology.
Next up, is the surveillance of the public, imagine all those cameras and microphones driving around, goodbye privacy for good.
The best bit is it's being sold to us under the banner of safety. I don't like road deaths either, but mas-scale surveillance is just the same problem.
I share your sentiments exactly, welcome to 1984.
Driving as a whole is safe now because everyone knows the stakes. If you screw up, you die. Computers don't care if you die, nor if they themselves are destroyed.
I don't look forward to a world where one Bluetooth RCE worm can kill tens of millions of people in 24 hours.
Humans, when aware, rested, and trained, are far better input devices for cars, I agree.
Where my sympathy for self-driving cars enters is that humans can be distracted by in-car entertainment (phones, screens, other humans or animals), tired/drunk/on drugs, or inexperienced. At least in the US (and several other countries I've visited, but not lived more than a handful of months in), I don't think driving safely is treated as a serious responsibility.
Beyond that, cars are already relatively automated -- steering, braking, traction, transmissions, and acceleration is computer controlled, and my car even tracks its maintenance schedule and various subsystem status, so I think it's somewhat logical that some degree of automation occurs to assist the driver more properly (/safely) operate the machine. At some point, this means normalizing "auto-pilot" functionality.
> Why not segregated roadways? If we're so desperate to automate trucking and shipping, how about big, thick K-rails, and dedicated roadway?
Because massive structural changes would be required. Similar to why buses are the go-to for public transit in developing countries -- four wheels and a relatively flat surface and you're done. Add in the challenge of retrofitting existing routes and our tendency to build large structures along roads and this is hilariously non-trivial.
Further, how would you design the segregated roadway? Do you take an existing lane from a highway? Is this railed-lane protected by barriers to prevent vehicular debris or illegal use? How do the shipping-vehicles exit the highway without blocking/crossing lanes?
And probably a Robert Moses character might emerge and cause just as many problems re-engineering roadways in broad, callous strokes.
But there's a huge ugly blind spot we're going to unleash upon the world with cars that decide what happens and when, all on their own.
People don't even like when the cops know how many emails are filled with retarded emojis on their phones. People go batshit insane about full disk encryption.
You're all mad if you think this car business going to work out in your favor.
Point 229 in Hammurabi's code: [229] If a builder build a house for some one, and does not construct it properly, and the house which he built fall in and kill its owner, then that builder shall be put to death.
or to put it in your words. CEOs (or their kids) must drive around in driverless cars before patrons pay for them.
That's beyond uncharitable. Please don't post inflammatory, unsubstantive comments to HN.