Tesla’s Elon Musk Says Autonomous Driving Not All That Hard to Achieve | Hacker News Reader

Tesla’s Elon Musk Says Autonomous Driving Not All That Hard to Achieve

wsj.com

65 pointsT-A97 comments

A better title would be

> Tesla’s Elon Musk Has No Fucking Idea What He’s Talking About

I guess going to space makes autonomous cars look simple. But in reality it's probably far easier to launch a rocket and dock with the ISS than build a autonomous car.

There are so many uncontrollable and unpredictable variables going on when you are driving. The only way that it has been done so far is by Google on a small set of roads that have been meticulously mapped and analysed. What happens if that area changes due to roadworks? It won't work.

Attempts by others are just as limited. Take a look at the TerraMax wiki page: http://en.wikipedia.org/wiki/TerraMax_(vehicle). This is basically for more off road/rougher situations in which plowing over small bushes and trees to the side of the path isn't an issue. But even still note that it got confused by tumbleweeds in a DARPA challenge.

It isn't a solved problem, and won't be for a long time. This is like people in the 80's claiming that we would have flying cars and hover boards by now. The only way right now to get autonomous cars en masse is to embed transmitters or magnetic strips in the road. Essentially we would need to put the cars on tracks.

This guy runs down a list of things that are unsolved and really fucking hard: https://www.youtube.com/watch?v=x5CZmlaMNCs#t=94

Everyone interested in autonomous cars should look at this video.

John Leonard puts tremendous passion into the subject, has done amazing things and talks about how hard this stuff is.

Thanks

Awesome video and another illustration of the critical role that DARPA government funding plays in subsidizing Silicon Valley. I think he mentioned "DARPA" about a million times in the video.

DARPA and massive state intervention in the economy should be right up there alongside private entrepreneurs like Musk who complete the last mile of bringing tech to market. It just kills me that there can even be a thing called a "Silicon Valley Libertarian" who's against "big" government. SV is a product of government investment.

I think many issues mentioned in this video could be (partly?) solved with better infrastructures. Just imagine how many infrastructures we have built for the auto cars: We built roads, traffic lights, stop signs, parking signs, we painted lanes, etc, etc. We've made a huge investment into the infrastructures simply for the cars to run safely and efficiently.

With self-driving cars, new infrastructures need to be built, not for human, but for different censors mounted on the car.

The nVidia solution is a bit different to the Google approach and seems to be quite effective? So the Google approach isn't really the only way.

I would get into any car that could safely navigate Mercer Street in Seattle for a full year.

As long as there is still a wheel and pedals, the car can just beep and say hey i can't handle this you take over. Or it can just pull over and say it's your turn now.

No, it can't just beep and tell the driver to take over. They haven't been concentrating on the road, there's no way they can keep their attention on what's going on for so long with nothing for them to actually do or think about, they have no idea what's going on and can't possibly come up to speed in the time it'd take to react to an unexpected and potentially dangerous situation.

It can pull over to the side of the road fine.

And if the driver is indisposed? Many suggest that with self-driving cars you can simply go to sleep!

IMHO, the path forward is very gradual. Dates below are production vehicles, experimental vehicles are much earlier

Add sensors and assess how they work - Lane departure warning, vehicle detect (years ago)

Next, add lane keeping and distance keeping (we're there, with caveats)

Next, V2V communication (2017)

Next, all the hard problems, one at a time, including infrastructure (2050)

This is like people in the 80's claiming that we would have flying cars and hover boards by now.

It's not a biggie but I'd say your excellent post implicitly encompass the inaccuracy of maps and so me pointing out that flying cars got their hype in the 50's NOT the 80's only reinforces your basic message.

And I would like to point out that we don't have flying cars not because we can't have them, but because there's no economic reason to. As a civilization we just chose not to make them. The same doesn't hold for autonomous vehicles - there you have a strong incentive, both economic and moral.

(Ironically, it's the same reason we don't have Moon or Mars bases yet - in 60s/70s Wernher von Braun himself had a pretty good plan for a Mars settlement using technology of the 50s. Again, we did't go there because no one wanted to bankroll such a mission (hell, it convinced the US GOV to roll back the whole space race before it goes interplanetary and starts to cost much more than it already had)).

Really, we do have flying cars, we just call them helicopters and they're too expensive for normal people to afford. People with sufficient wealth get to use flying cars, though.

The nice thing with self-driving cars is that it's mostly a computer problem, so the interval from working at all to being affordable by the common person should be short.

I'm going with Elon on this. Google, Uber, Apple and the Big 4 all competing to hit this. And that's to say nothing about the military and how they'd use self driving vehicles in a war zone. That's so much R&D horsepower it's hard ot fathom. Today we are mapping the world, have insane sensors. Cars are getting to being self driving for large periods of time of freeways due to lane change control and speed measuring via radar. 2025 is when the pressure will start to be exerted to get human drivers off the road because they'll be the only ones who don't act in a predictable reliable way and thus the single biggest threat left.

Elon Musk has built (companies and hired people that built) cars that are capable of driving autonomously on the highway. A whole fleet of them in customers' hands will be doing this in the real world in a matter of weeks or months at most.

Full autonomous driving is, of course, much harder than just highway driving. But to say that Musk "has no fucking idea what he's talking about" is completely ludicrous.

There are levels of autonomous driving. We have the basics today. We have cars that can reverse park, maintain appropriate distance between cars, avoid collisions etc.

The next level is coming in the next couple of years. Volvo is targeting 2017 as the start of its rollout:

http://www.cnet.com/au/news/volvo-autonomous-drive-me-detail...

This is autonomous without the tricky parts of driving in inclement weather.

I really don't know why people keep ignoring the large car makers when talking about autonomous cars. They have proven experience with managing complex automation systems and the money/time to pull it off successfully.

I agree with you. I've followed the advancements from automakers like Mercedes (who also put their tech in trucks). Highway driving and parking in highly static environments is at an acceptable level. But that doesn't mean the problem is anywhere near solved.

I'm floored by your hubris. Do you really think it's likely that someone who runs a car company and funds artificial intelligence companies isn't well-informed about this subject? Is that really the likeliest explanation of why you two have reached different conclusions?

I'm going to have to agree with the top post, sometimes really smart people say stupid things. I wouldn't read too much into it.

Do I believe someone who delivers cargo to the ISS and is building enough battery production capacity to satisfy the world's EV production requirements? Or do I believe commenters on Hacker News. Tough call. /s

How about the hubris of over-analyzing a sound-bite quote reported by the WSJ, and ignoring Elon's past, lengthy statements that the next step is "on ramp to off ramp" driving? Which is a lot easier than all the hard problems being mentioned in this discussion.

I disagree with you. The problem of autonomous vehicles is yes, a very difficult one, but steps are being taken towards it at a rapid rate. I work in the area, and parking and standard autonomous motorway driving, where the main required algorithms are lane departure, vehicle detection and adaptive cruise control are pretty much considered 'solved' in the research community.

I absolutely agree that there are plenty of aspects that are not solved, and possibly this is my passion for the domain shining through, but I would be much more optimistic for the future, and the near future at that.

The ISS is extremely difficult to "berth" (not dock) with an automated spacecraft due to safety rules to protect a $150bn asset. Berthing with the ISS is currently assisted by a remote control arm, Canadarm2. An actual automated docking (Dragon v2) is even more difficult.

What Elon is saying is that he knows the steps towards an operational automated driving system, and can expand its capability at each step. Most of the limitations are "regulatory" and not technical.

Difficulty and regulatory requirements are not the same thing. The difficulty of berthing with the ISS is technically quite low. There is a single object to contend with (the ISS) of which the full parameters are known. There are no other objects to get in your way. Visibility is good. All in all it's a highly predictable and very static environment.

I completely disagree. The limitations are technical not regulatory. Elon Musk might think he knows at a high level how to slowly introduce the technology. That's great and all but the technology doesn't exist yet and won't for a long time. I'm talking about 100% driving duties here, not just cruising on the highway and parking in a controlled environment.

This seems like marketing-speak to me. Or perhaps they haven't gotten far enough in their development for it to be really hard yet. Here are some examples of reasons this is harder than you'd think:

- At a complex intersection, how does the computer decide which traffic light is yours?

- You see an obstruction in the road. How do you decide if it's something like a plastic bag you can drive through, or something you need to swerve around?

A very experienced team has been working on this at Google for almost a decade. I'm not sure how Elon's team would have skipped so many steps.

Indeed,

And even the google car ignores entirely the traffic lights that aren't on the highly detailed map guiding the car [edit: that disappears when it loses connectivity. But connectivity is a "solved problem" too, right?].

But the simple parts of the problem are solvable. The very simplest parts of the problem could probably be solved with analog relays.

One theory: get enough hype and support for autonomous cars and cities will be forced to install wireless guides for autonomous cars. Then the only problem is people stepping in front of them. So you ignore that problem and away you go.

I mean, regular automobiles, when they were first introduced, caused massive casualties. Only in the eighties did we begin to get this under control with air bags, sane speed limits and so-forth.

Cars have been just rolling death for a large portion of the history of people using them. Well, autonomous cars can just leverage that fine history.

We haven't been improving on drivers, though. Twenty years ago, humans as a group were as bad as they are today and as they will be in another twenty years. Or thirty or fifty. You can add seatbelts and rear parking cameras, and you can legislate against texting while driving, but you can't legislate away tired or distracted drivers, or just plain unskilled drivers. As a group, the only way we improve is through demographic shift. But autonomous cars get better, as a whole, and consistently, and rapidly.

Maybe autonomous cars will be at the 50th percentile in five years, but they're only as good as the bottom 10% of drivers, well, we'd still be better off with those 10% of drivers switching--which would lead to infrastructure to improve it. (I think more importantly, social understanding/recognition of them. The hard problems presented in Dr. Leonard's video posted on this thread aren't peds stepping in front of them, but cops gesturing and unprotected left turns across traffic where you need to make eye contact.)

Finding the bottom 10% sounds hard, because, yes, a lot of bad drivers think they're good drivers, but I think the majority of us know we suck. For many, their insurance companies know too.

For the traffic light problem, I'd start off by programming the car with common scenarios, and when you hit one it doesn't understand, throw up a prompt that tells the driver to tap the accelerator when it's ready to go. You can then improve from there. "Driverless" doesn't have to be all or nothing. We can ease into it.

For the plastic bag, I find it odd how often people bring up scenarios like this that will supposedly stump a driverless car, yet humans are terrible at as well. "What if a massive truck comes careening down the road straight at you, and your only avenue of escape is through a playground full of children? What does the car do?" Well, somebody dies, just like now. My experience is that humans are terrible at deciding whether some crap in the road is safe to run over, so computers can hardly do worse. And I think they could easily do better. A front-facing radar (already standard equipment on many cars) will tell you whether that lump the camera sees is actually substantial or just filled with air.

"What if a massive truck comes careening down the road straight at you, and your only avenue of escape is through a playground full of children? What does the car do?"

We are going to let a computer make that decision? I don't think so.

Why not? A computer can make a much better decision than you do (doing a proper evaluation of options and probabilities) and then execute it much better than a human ever could.

I mean, not pushing for autonomous driving just feels... weird.

The thing is, not only will we let a computer make the decision, but I'm sure we already have.

Automatic collision avoidance is a real-world feature. There are real cars out there you can buy which you can drive straight at a brick wall with the accelerator to the floor and it will not let you hit. I'm sure that avoiding collisions with pedestrians (or big trucks) is part of the deal. If the correct answer isn't "brake" then the computer will get the wrong answer, and almost certainly has already somewhere in the world.

People discuss autonomous driving as basically flipping a switch. We have manually driven cars, and then one day we'll wake up and there will be an autonomous car on the road. But it won't be like that. Autonomy will be a gradual process. Car makers are steadily bundling in more and more automatic driving features. Eventually they'll fill in all the gaps, one by one.

> "I view it as a solved problem"

Wow. We've been trying to nail "simple" things like street light timing algorithms for years and all of a sudden, self-driving cars are being prototyped and promised for sale soon. I feel like this is popping out of nowhere, even though self-driving cars are the type of things fantasized about in science fiction from the 20th century. I even thought "I, Robot" was presumptuous to depict automated cars being the majority in 2030. And yet here we are...

It's really weird for this sort of parallel invention to happen independently due to corporate secrecy, rather than because of free exchange of ideas and designs. Also: all of these working VR headsets are coming out of the woodwork after decades of being insufficiently immersive.

"If human thought is a growth, like all other growths, its logic is without foundation of its own, and is only the adjusting constructiveness of all other growing things. A tree can not find out, as it were, how to blossom, until comes blossom-time. A social growth cannot find out the use of steam engines, until comes steam-engine-time."

Charles Fort

http://www.resologist.net/lo104.htm

edit - a quick game of Charles Fort Discovering Things would be great right now, though it is difficult without a frog and an umbrella.

The free exchange of ideas and designs will ultimately win out. This will be evident by 3DRobotics outpacing a company like DJI. Also, I agree with the whole VR thing. Have you heard of the Lighthouse tracking technology by Valve? Crazy stuff

My biggest puzzlement is why after decades, VR is now suddenly a viable thing. But I've heard the answer is because the rise of smartphones have presented us with cheaper high-resolution displays and motion sensors and other tech that finally brought the medium out of the '80s.

As far as self-driving cars go, I remember this article, written just over a decade ago: http://archive.wired.com/wired/archive/12.03/robot_pr.html

Now that's astounding advancement.

This reminds me on how researchers call speech synthesis a "solved problem" since decades. When asked why they don't use it for presentations, they say that it just doesn't sound very well..

Speech synthesis is substantially more of a solved problem than you think. Look at Vocaloids, for example. Here you have entirely computer-generated "voices" singing pop songs, and achieving a surprising degree of commercial success. Or look at something like the Tupac "hologram" that sang one of Tupac's unreleased songs at Coachella. Granted, that was taking samples of Tupac and combining them, but it was combining them seamlessly to perform a new work. That's pretty close to pure synthesis.

If I put your comment to any of the popular TTS systems, it will be understandable, but painful to listen to. If I create a presentation and use a TTS instead of talking myself, people will cringe and hope it's over soon. That's what I meant.

Also, the examples you mentioned were very specific. The Tupac song was a one-off, limited scope and highly optimized before release. If I'd use the same system to render any random text, it would still sound off.

So yes, speech synthesis does "work". But to find back to the original topic, it's like driving drunk..

We're not "here" though. Elon Musk saying "it's a solved problem" and it actually being a "solved problem" are two extremely different things. Musk is a smart guy and a good engineering manager, but he's far from being a supergenius, and has said stupid things before.

Google has a system that kinda works, but requires a very controlled environment, and doesn't cope with exceptions all that well. Hopefully it is "fail safe", in that cars will stop or slow down rather than run the risk of running someone over (something humans arguably fail at way too often!), but it's not yet clear that it's sufficient for a workable system.

Maybe Musk has something magic up his sleeve that will change everything, but so far there's zero evidence for that, and given Musk's history of bluster, I'm pretty skeptical.... If he really has something, he'll show it eventually.

He seems to think that these problems are just bugs in the system. We as humans know how to solve them, so it's just a case of coding that into the cars brain.

Did you see the video nvidia showed where all the cars are networked, so as one car figures out something, it uploads this to the cloud so all cars get that knowledge. ..

Yes,

Here we are, with the smaller, easiest subset of the autonomous driving solved and massive hype anticipating the rest being solved "just 'cause" but offering no details concerning the solution.

I mean, with this hype flurry, we might soon see posters of flying cars. Imagine that once again we could look forward to that and once again be disappointed. Wouldn't that be a rush?

The coast-to-coast autonomous drive: "A human driver is going to be sitting behind the wheel at all times, but won’t interfere unless the car requests help or has to turn off the highway." -- Yes, an autonomous can drive a directly forward on a freeway with other cars or up a mountain alone. ETC.

http://www.digitaltrends.com/cool-tech/the-self-driving-car-...

I imagine a lot of things look not all that hard when you're trying to build a rocket that can go to space, come back down and neatly land itself on a helipad...

To me that sounds much easier than an autonomous car that has to work any time, deal with pedestrians, things falling off other cars, rain, snow, ice and many other unpredictable factors.

I think those thing all on their own is not the hard problem - the hard problem is deciding whether to avoid the child crossing the street by driving over the cliff or slamming into the oncoming traffic...

That's just a dark scenario of parent's problem.

well - yeah it's pretty dark scenario... but it's an extreme example of why it's a harder problem IMO to solve than a rocket landing back on earth. You have multiple cases simple or complex where a decision must be made to either drive over a pot hole or drive into the oncoming traffic lane or off the road altogether. I'm certain a computer controlled car can make these decisions faster than a human driver, but much like a computer can typically not always judge the difference between my sarcastic comment and it's literal meaning it I would imagine have a difficult time solving these problems... maybe NP problems are solvable now?

Don't get me wrong as a parent - I am super excited that perhaps - just perhaps, my kids when they are ready to hit the road won't have to worry about learning how to drive :D and I can rest easy the other drivers won't be crazy too.

Everyone who wants to take the other side of the self driving car argument brings up stuff like this. Also they say what about icy roads and bad weather etc.

Guess what, humans also suck at this. What would YOU do? Do you hit the oncoming traffic, drive off the cliff, hit the kid? Likely the right choice is to slam on the breaks and hope for the best. That's likely the same choice the computer would make, only it will have a faster reaction time making it more likely for the kid to make it.

We kill 30,000 people a year driving. Humans suck at driving as a general rule. Will a self driving car be able to handle icy foggy roads? Probably not. And it will note that ahead of time and say dude you really shouldn't be going out in this, I can't handle it. If you want to try it, you'll have to drive yourself today.

I think a driverless car would just slam on the breaks, and do so much quicker than you or I ever could.

And why is it always a child? You sound like a politician trying to make a "think of the children" argument.

Except that when you have self-driving cars, those other cars are making similar decisions and opening a hole for a car that just might need to avoid that child.

This is why self-driving cars are going to be far safer than humans. Your self-driving cars will see a thousand hazardous situations as you are driving and adjust to avoid them pre-emptively and silently.

Yes, there will be some bugs on the way there. And it will kill a few people. But we kill about 30,000 people in car crashes per year and about 5,000 pedestrians per year with cars.

The programs could be quite buggy, and still beat that by I suspect a very wide margin.

The decade or so long problem is the mix of self-driving cars and those piloted by humans. And yes, the benefits will be obvious.

As if humans are any better at this decision?

Very few humans can execute split-second, emergency decisions correctly.

Most simply panic and whatever happens, happens.

I don't think that is so hard, given that it can make more common (i.e. less contrived) decisions all the time. In that case, various priorities would be weighted, and those weights can be adjusted to perform as their designers and society prefers, in various simulated scenarios.

Are you asking whether the life of the child should be given more priority than the life of the occupant? Who knows. Regardless, my bet is the car will perform better, by any measure, than a human. If only by being a lot quicker to respond and avoid the situation entirely.

I think the tough problems involve communication and negotiation with other drivers. A car might have a hard time detecting eye contact, or noticing when someone is waving you to go ahead. It might also be bad at being aggressive enough to merge. But some of these things human drivers are really bad at....especially new drivers or the elderly, both of whom are often timid. People might be able to see that it is an autonomous car and be more likely to take advantage of the fact that they know the car won't do anything rash like risk an accident to force its way in.

The nice thing is that these cars are going to keep a lot of data. Some people may not like that, but I don't think it's a bad thing that if you keep driving dangerously, the autonomous cars are going to note your license plate number and your insurance company is going to find out.

That decision is the easy problem. No sensors right now can reliably identify a cliff or a child, even in optimal conditions.

The hard problem isn't a technical one, in my opinion. Lets say we include an 'emergency stop' button on the autonomous cars. It will be early on, when the autonomous car decides wrong or doesn't detect the kid, and the lone passenger wasn't paying attention and/or was drinking.

The question will be, was the human responsible? Should he/she have hit the 'stop' button, and was their intoxication the reason why. Will they be charged with vehicular manslaughter?

Also, construction zones, traffic cops, parking garages

It's really not a hard technical problem at this point. It's mostly social and political. Think about the typical car out there:

- You need a piece of plastic and metal to unlock it. Losing that object is a serious inconvenience, especially if you're on a road trip. Compare that to unlocking your cellphone or your bank account.

- It doesn't have a "settings" functionality. For example, you can't configure a car to refuse to move if a seatbelt is unbuckled. Such a feature (perhaps a default that you can disable if you really want to) would save many lives.

- It would be easy to write an app to help you optimize your driving for safety / fuel efficiency if you could download data that most cars have, but you can't.

- Your car could warn you if you're approaching a red light or a stop sign and not slowing down. It's trivial, and it would save lives.

These are just a few examples of ways in which most cars in the market are archaic and inefficient. This is mostly because of two reasons: the incumbents have no incentive to make changes that don't increase sales, and governments don't pressure car manufacturers to include state-of-the-art life saving technology until it has existed for a LONG time and everyone knows about it:

http://www.history.com/this-day-in-history/federal-legislati...

The difficulty in making self-driven cars the norm is mostly about the politics of the transition.

> Your car could warn you if you're approaching a red light or a stop sign and not slowing down. It's trivial, and it would save lives.

It's definitely not trivial.

It's not trivial from first principles, but traffic sign recognition is more or less a solved problem at this point (production Teslas read speed limit signs as you drive by them, for example) so it would be trivial to implement given what's available now.

Teslas being able to read a single format of signs is absolutely not indicative of their ability to read and recognize the hundreds of traffic signs in the US, many of which are located in non-standard positions.

If traffic sign recognition really was a solved problem, every car manufacturer would have implemented it by now. If implementation really was trivial, we'd all be talking about which self-driving car we're going to buy this year, not when the first self-driving car might become available.

I've seen it in action in other contexts on all sorts of different signs. And for alerting a driver who's failing to stop at a stop sign, all you have to look for is one kind of sign that's even easier to recognize than speed limit signs: just look for a red octagon, and no need to OCR the text.

I'm a huge fan of Musk, but this was an intemperate thing to say. Inevitably at some point someone is going to get killed by one of his autonomous cars and at the wrongful death suit the plaintiffs lawyer will bring up video of him calling the problem not that hard and "solved" and he will have a much worse problem on his hands.

The truth is, it is hard to make any system as flawless as we're going to want these systems to be.

> Mr. Musk has said Tesla plans to be the first company to offer customers an autopilot feature. “I think we will be the leader in autonomous cars that you can actually buy,” he said.

That is interesting to read, especially with the knowledge that they brought this technology from Bosch/Daimler. Without their technology cooperation Tesla still would not even had adaptive cruise control, which every decent car of the last 10 years offers.

BTW, from the technology point of view, we are at about 90% of the self driving car. But developing the last 10% is hard. Besides that there are a lot of ethical questions and insurance questions to answer.

Ethical: what happens when the car detects an unavoidable accident: where should the crash in? Who decides that? Who is responsible for that? These days it is easy to answer that questions: it's the driver. But in a self driving car?

Insurance: what happens if an accident takes place and a person dies from it? Who is responsible? The driver? The car owner? The car manufacturer? The sub-system manufacturer? The developer?

There are a lot of non-technical issues to solve.

Re: Insurance - that one will be straight forward. Cars with Full On automation/DriverAssist will be 10x (if not more) safer than human drivers, so the accidents that do occur, will be fully covered by insurers. Your policy will probably be 1/2 of what it is today as well.

The one that gets me, is how to make decisions regarding people doing traffic direction. What happens when there is someone standing in front of you, wearing a traffic-safety vest, directing traffic down a lane. Do you go down that lane. Do you ignore them and keep driving forward, potentially running over the person if they don't get out of the way. Do you hit the brakes and accelerate backwards? Depending on the context, any of those decisions might be correct - but I doubt that any "Automated Driving" system will know how to handle it in the next 30 years.

Now, Driver Assist, in which the car keeps you in your lane, brakes before hitting the care in front of you, swerves into the emergency lane in the event someone is about to side swipe you, stops at traffic light/stop sign - these are solved problems, and I expect will be rolled out (already are in many cases) in most cars over the next 10 years.

Actually, it sounds simple. You never run over the person directing traffic because of the usual anti-running-over mechanisms that the car will have. You can teach the car to recognize the signs policemen use to wave, and it seems feasible to learn the car to recognize gestures - after all, AFAIR, Google already solved that for recognizing hand signals that cyclists give. And OTOH, if you want to direct traffic in the world of mixed driver/driverless cars, it's on you to know you might be dealing with a machine and make sure you gesture clearly.

The point I was trying to make, is depending on the situation, all three responses are entirely reasonable. If it's some idiot/kid/whackjob playing games, you keep moving forward, making it clear he needs to clear out of the way. Maybe lay on the horn. If it's East Palo Alto (or wherever the high-crime area is in the United States right now), and you are about to get jacked, you hit the brakes, reverse, and get the hell out of dodge. If it's a normal traffic situation (Construction, Police Officer, etc...) - you follow their directions.

A human can make that assessment in a few hundreds of milliseconds - it's going to be 20+ years at least before any automated driving system is going to figure out what to do in situations like these. (Please don't get overly focussed on the "Some random person pretending to direct traffic situation" - I'm sure you can come up with a hundred other scenarios which require executive, not tactical assessment.)

This is why I believe DriverAssist is going to be incredible, particularly on freeways - and already is today - It's an awesome thing to have a recent model BMW brake from 85 mph to a complete stop when the car in front of them slams on the brakes. An entire category of accidents is going to be virtually eliminated as DriverAssist is added to Vehicles.

I'm of the believe that my children will not have to learn to drive and the children they have will not be allowed to drive (under normal circumstances)

I'd like to propose finding a new street parking spot in NYC after moving out of the way of the street-sweeper as a Turing test for the autonomous car.

Does everyone really believe fully autonomous cars are just around the corner? I thought Google was pretty much just snowing us with how close they really are. They have a car trained to drive the same road, over and over again.

Every time I drive now, I noice one or two random events I have to negotiate and wonder if the Google car is going to be able to handle them. Things like misleading turn signals and body language from human drivers, the confusion that occurs with multiple people messing up a 4 way stop, etc.

One thing i've noticed on Hacker News whenever subjects about AI come up, particularly with autonomous cars, is that commenters who claim to actually work in the field always call BS on the hype.

It has been interesting to see how many friends and relatives have an initial "Why would I trust a driverless car on the road" reaction until I point out the idiocy of "other" drivers on the road. People believe they are the best drivers, yet readily agree that everyone else shouldn't be driving.

I'm really looking forward to seeing the progression of this. Seems like it could be the next big change to society norms.

What about liability? Today if a driver makes a mistake and hurts someone, the driver is responsible. If a self-driving car makes a mistake, is the automaker responsible?

In 2013 auto insurance companies paid out $124 billion in liability and property damage claims.[1] Plus who knows how much courts awarded in tort trials for serious automobile accidents.

Even if a self-driving car is 10X safer than a human driver, that's still equivalent to 10% of today's insurance and litigation burden being shifted to automakers. Won't this have a significant impact on the cost of autonomous cars?

I can't see courts upholding the standard software license agreement that indemnifies software makers from bugs when it comes to autonomous vehicles. Injury & death are not the same as data loss, and in any event that would only cover the driver, not all victims.

For this reason alone I think the future will be one of semi-autonomous driving, where humans are still required to be trained and in control and ultimately liable for decisions. "Cruise control on steroids."

[1] http://www.iii.org/table-archive/21338

> What about liability? Today if a driver makes a mistake and hurts someone, the driver is responsible.

Most likely, liability will be addressed to some extent in statute authorizing general use of driverless vehicles, but there's been quite a lot of analysis of this issue just based on generally applicable liability principles.

For one example, see http://www.brookings.edu/research/papers/2014/04/products-li...

> Even if a self-driving car is 10X safer than a human driver, that's still equivalent to 10% of today's insurance and litigation burden being shifted to automakers. Won't this have a significant impact on the cost of autonomous cars?

At the numbers you give, not particularly. Let's use that 10% number, and assume that the $124 billion paid out in 2013 was spread out over just the new cars sold in the US 2013, rather than all cars sold in 2013, or all cars on the road in 2013. 15.6 million cars were sold in that year, and we've got to spread $12.4 billion in additional automaker liability across them -- that's less than $1,000 per car. (And, of course, if the automakers are taking all the liability, that means 100% of the cost of liability insurance to drivers goes away for driverless car purchasers, so the total cost to the consumer will be dramatically lower, even with the cost added to pay for the automaker's liability burden.)

I feel like there needs to be a stronger phrase than "cognitive dissonance" to describe someone who thinks that the software that can write captions of pictures of cats is evidence of a possible existential threat to humanity but advocates for letting software control what he himself calls "two-ton death machine[s]."

We then need a set of stronger phrases for "failing at reading comprehension 101", "uncharitable presentation" and "failing to think on a topic for even 30 seconds".

Also, I see no cognitive dissonance here - just a realistic understanding of how powerful software can be.

I am always struck that you always see the autonomous driving demonstrations in favorable climate conditions. I have never heard of a test at night with low visibility due freezing rain on snow packed roads like you might experience in the northern Midwest or Canada.

Guess what, people really suck at that too. Most decide to stay home when it's like that because it's not safe. No reason to focus on that now. Just say, nope, this isn't safe you should stay home. If it's an emergency, you can drive yourself in these terrible conditions.

The hardest problem isn't a technical one; it has to do with liability. There will be crashes and deaths, no doubt about it. But who is at fault now? The car maker (or whomever wrote the software)?

Because of this, laws regulating autonomous driving will be the bottleneck. I can envision something transitional: approval for long stretches of "easy" driving will come first. For example, driving on the I-5 from LA to SF.

For me, this is the perfect use case for a self-driving car so I can enjoy doing something else rather than paying attention to a boring drive. Personally, self-driving cars in a downtown area isn't as crucial. Unless I have no idea where the hell I'm going in a confusing city.

Tesla could become an auto insurance company that only covers their own cars. I don't think liability insurance is as much of an issue as regulation.

How is the liability issue different from a mechanical flaw that causes an accident? Say a tire blowout. Is it the fault of the car manufacturer or the tire maker? None of this is particularly new.

I don't doubt laws will slow things down, but there will be a point where it is so clear that autonomous cars are overall safer, that they'll figure things out.

There is no problem.

They both can be. The owner is responsible for the car they put out in the road. The maker (software, hardware, hell even the retailer) will be liable under products liability for any flaws in the design.

There will be some debate about negligence and autonomous car crashes. But you don't even have to pass laws. Courts will just make the rules up as they go along. It's the beauty of the common law system. Judges fill in the gaps when new fact patterns are applied.

I'm guessing car companies will end up paying for it because they have the big pockets and they can best handle it.

I'm not sure there aren't big technical problems still ahead. The stupid Uber app won't stop sending drivers down alleys in Washington DC. LOL at there being a computer driver in the next 10 years.

I both love this future because I think it will save lives and be awesome, and hate it because I am sure, totally sure, that we won't get a physical disconnect button to turn the car off, open the doors and get out if the car is malfunctioning, on fire, or taken over by hackers or the 'Car Safety Administration and Prevention' department.

I think it's interesting there's so much billionaire scare-talk about skynet and the risk of AI and none of these companies mention meatspace-level security with their cool autonomous AI's they're building.

I genuinely wonder why that is; is it just preferring control (over the robot product by the company)?

For the guy that leads a company that has a rocket that can come back from space, i reckon a car that can drive itself is a comparatively simple problem. The real problem is laws. We're making progress but its going to be damned slow.

I reckon there will be a future time where we look at human controlled cars as the same as we look at horse and carts now, but also as very dangerous things that are eventually outlawed unless you're on a racetrack. I think eventually motorsport will be the last place you'll see human driven cars.

I cant work out if Formula E with completely autonomous cars would be interesting, or really boring...

>I cant work out if Formula E with completely autonomous cars would be interesting, or really boring...

in various papers there are announcements of how robots discover cooperative behavior ... I somehow expect that with further development of AI the robots should also discover the benefits of the opposite behavior too :)

> I cant work out if Formula E with completely autonomous cars would be interesting, or really boring...

Well it can't get any more boring than it already is.

I think producing a consumer-ready autonomous automobile is a significantly more difficult problem than sending a rocket to space and back.

‘In the distant future people may outlaw driver cars. You can’t have a person operating a two-ton death machine!’

I don't care how good the design is, people will always require a way to take control when they wish. The trick will probably be to make the UX (passenger experience - PX?) so much easier than driving that nobody will want to.

But outlaw them? Nah...

Pretty easy to imagine manually driven cars (one day) becoming "not street legal", while still legal to own and drive on private tracks etc as with current exotic/powerful cars.

Will they? You have zero control over the bus, or if they're driven by a chauffeur. For all intents and purposes, such vehicles are self-driving, just with some meat brain instead of a silicon one doing the thinking.

Also note that in both of my examples, you're not the owner of the vehicle. With self-driving cars there will be little reason to own a car, and such the need for having control will be smaller.

Right, an autonomous ride-share car service would be ideal in its capacity to reduce the need for personal car ownership. But even so, technology is fallible in extenuating circumstances, and human intervention will always be needed as a recourse.

> But even so, technology is fallible in extenuating circumstances, and human intervention will always be needed as a recourse.

I disagree. The basic assumption for banning human drivers on public road is that the technology will reach the level of being strictly superior in terms of driving safely. At this point letting humans drive is just very stupid.

I am not allowed to take control of the plane I am flying in today. It is actually illegal to do so and I face severe penalties for even trying.

Same deal for taxis, busses, trams, cable-cars, ferris wheels and ferries.

We have modes of transport we do not control, where attempting to take control is illegal.

Why not the same for cars? They will simply be self-driving taxis after all.

Because you don't own any of those things. If you own your own plane and have a pilot's license, you are more than welcome to fly it yourself and to disengage the autopilot should the situation demand it.

I wouldn't mind an autonomous car-share service. Just saying that Musk's futurist vision of mandatory self-driving vehicles is incompatible with personal control of personal technology, and practically, politically infeasible in the US anytime soon.

I wonder if saying "it is a solved problem" might be a bit discouraging/demotivating to the engineers who are continuing to work to make it better.

Shouldn't Elon be concerned that the car might decide the human is superfluous, open a door and swerve wildly until the body is ejected

Reply on news.ycombinator.com