Vans Drive Themselves Across the World

bossanovalithium writes "Four driverless electric vans successfully ended a 13,000-kilometer test drive from Italy to China which mirrored the journey carried out by Marco Polo in the Middle Ages. The four vans, packed with navigation gear and other computer software, drove themselves across eastern Europe, Russia, Kazakhstan and the Gobi Desert without getting lost. They had been equipped with four solar-powered laser scanners and seven video cameras that work together to detect and avoid obstacles."

Comments

Very cool, but...

[Anonymous Coward]

... as cool as it sounds, the vans were mostly designed to form a "virtual train" after a human-driven vehicle, so it's not quite autonomous navigation just yet.

Hey at least something cool out of my home country for once!

Re:Very cool, but...

[Anonymous Coward]

... as cool as it sounds, the vans were mostly designed to form a "virtual train" after a human-driven vehicle, so it's not quite autonomous navigation just yet.

I figure the first practical use for autonomous vehicles will be articulated lorries / semi-trailer trucks on the motorway / freeway. They tend to form virtual trains anyway. If they can talk to each other, you could get them just inches apart (with synchronized braking, etc.), which I imagine would offer opportunities for much improved aerodynamics.

Re:Very cool, but...

[gorzek]

Liability is one of the things I worry about with any kind of autonomous road vehicle. The first time one of these automated "road trains" shreds through a family's sedan I expect there will be fighting between the trucking company and whoever developed the automated driving system to decide who is financially liable for it.

More Importantly

[Ltap]

Did they bring back any spices or silk? And we can't trust their tall tales of two-headed men without proof!

Re:More Importantly

[rossdee]

Fortuately since there was no people in them, they didn't bring back the black death or bird flu or otther deadly diseases...

Re:More Importantly

[Anonymous Coward]

Driverless yes, peopleless no, RTFA:

Though the vans were driverless and mapless, they did carry researchers as passengers just in case of emergencies. The experimenters did have to intervene a few times. The vans got snarled in a Moscow traffic jam and humans were needed to handle toll stations. At one point, a van stopped to pick up hitchhikers.

Re:More Importantly

[snspdaarf]

Oh, great. First, my GPS tries to kill me by directing me down a one-way road the wrong way, now my automated van is going to stop for some Manson wanna-be on the side of the road. No, thanks!

Re:More Importantly

[tophermeyer]

At one point, a van stopped to pick up hitchhikers.

I thought you were joking, so I checked TFA. This actually happened. Which is crazy. Horror movies start with stuff like this.

Robot vans picking up hitchhikers? In what twisted universe does a hitchhiker: 1) flag down a van 2) discover that it is driven by nobody and 3a) trust the van's occupants that they are "researchers" 3b) trust that "it's totally cool, nobody's going to steal your organs".

Re:More Importantly

[MyLongNickName]

I thought you were joking, so I checked TFA

Damn noobs.

Re:More Importantly

[Anonymous Coward]

But most men are two-headed.
[Unbuttons pants]

Re:More Importantly

[kryliss]

Commenting on your sig, When the humans had to take over during "Moscow traffic" was probably caused by a blue screen of death. :) Obligatory Microsoft bash completed.

Re:More Importantly

[Anonymous Coward]

Horror movies start with stuff like this.

But so do pornos

Re:More Importantly

[CastrTroy]

This is exactly the problem with self driving vehicles. Even if they are 10000 times safer than having human drivers, they will still not be used by the general public, because any kind of crash will be a huge lawsuit against the company. With human drivers, you can always blame the problem on human error. However, with computerized drivers, it's now the manufacturer who is at fault for every single problem. They can't even get all the systems they have (Think Toyota) working properly. Making cars that drive themselves is going to be an even bigger problem.

Re:More Importantly

[JWSmythe]

If you make it, they will come.

    I can just picture the van with shag carpet and disco lighting, and Number 5 driving.

    and of course, appropriate music.

    (I couldn't resist)

Re:More Importantly

[Anonymous Coward]

This one? What are the odds that the group with the robotic van are anything *but* researchers, in the present time?

Re:More Importantly

[Anonymous Coward]

A universe where the the van programmer's 'plan b', in case of van failure, was to conveniently pick up a 'hitchhiker' (engineer) who could guide it the rest of the way and still claim that it was automated.

Re:More Importantly

[slick7]

At one point, a van stopped to pick up hitchhikers.

I thought you were joking, so I checked TFA. This actually happened. Which is crazy. Horror movies start with stuff like this.

Robot vans picking up hitchhikers? In what twisted universe does a hitchhiker: 1) flag down a van 2) discover that it is driven by nobody and 3a) trust the van's occupants that they are "researchers" 3b) trust that "it's totally cool, nobody's going to steal your organs".

What happens when the Taliban, al queda, Palestinians, Chechens, or whom ever grabs one of these vehicles, loads it with explosives and sends it on its way. So much for symmetric warfare. Or better yet, slapped with national security papers thus preventing patenting (except for the military, of course).

Re:More Importantly

[apoc.famine]

I don't know if lawsuits will be that big of a problem. The automation has already started. It will just be a gradual trickle-up, with a lot of legalese you have to sign off on in order to have these automations.

Mercedes already has the option for active object avoidance systems - cruise which detects slower vehicles in front of you and matches their speed 150' behind them, blind spot avoidance systems, IR night vision, rear camera, parking assistance.

In 10 years, that will all be standard on about every car - just as ABS, airbags, and satnav have started to become standard features on all but the most inexpensive cars. Sure, there will be questions at first, and some lawsuits. There always is with new technology. But new technology also starts with the rich, due to it being prohibitively expensive when it's first introduced. Merceds will hammer out the bugs. Then it will trickle down to the rest. It happens with just about every new technology.

When Mercedes has had all that tech for 10 years, and it's trickled down to mid-class cars, what are they going to do to get rich folks to keep buying their cars? They're going to have to innovate. And that will just be building upon all those features that they've refined for over a decade.

Re:More Importantly

[supertrinko]

I imagine one of these would be less effective at explosives delivery than a remote controlled vehicle would be.

Re:More Importantly

[slick7]

I imagine one of these would be less effective at explosives delivery than a remote controlled vehicle would be.

Autonomous vs. remotely operated is different how?

Re:More Importantly

[Anonymous Coward]

remotely operated has a radio transmissor that reveals your location. autonomous does not.

Re:More Importantly

[Anonymous Coward]

they will still not be used by the general public

Did you mean "will not be available to he general public"? I suppose all it takes for them to be picked up in the US is a change of the liability laws similar to what happened to aviation (General Aviation Revitalization Act of 1994).

Frist Thumbs-up!

[pinkushun]

At one point, a van stopped to pick up hitchhikers.

Re:Frist Thumbs-up!

[EdZ]

Shouldn't that be thumb sideways?

Sponsor

[flyingkillerrobots]

For those of you who want to know who made the vans, it was sponsored by the European Research Council. The lead researcher works at the University of Parma, Italy. Why, oh why do the summaries lack useful information? Yes, I am new here.

Re:Sponsor

[timeOday]

That's what's really interesting and new about this story to me - this wasn't done by the usual characters from CMU or Stanford who have won the DARPA driving challenges in the past (the google car is from that same lineage also). Whether developed independently or replicated, the technology is getting more widespread.

Re:Sponsor

[autophile]

Maybe they got rid of "Who, What, Where, When, How (and in investigative journalism, Why)" since the time I was in school.

Re:Sponsor

[apoc.famine]

It's been like this for a long time. Just in the last six months or so it's seemed to get really bad. At least this summary isn't an outright falsehood like many in the last week.

Not more "safety features" please

[inigopete]

Rather than replacing drivers it is hoped that the technology will be used to study ways to complement drivers' abilities

That's become the problem with ABS, traction control, airbags and many other safety features: make drivers feel like they're safer, they will drive more like idiots. I'd far rather this system was developed to replace drivers; granted it will take more work to make it completely reliable, but it would mean fewer people thinking that because they've got the latest safety systems in their car they don't have to pay as much attention to their driving.

Re:Not more "safety features" please

[Ltap]

The problem is that most people would rather trust a human in life-or-death situations, despite the fact that humans would be hampered by slow decision-making and reflexes.

Re:Not more "safety features" please

[somersault]

Who do you know that drives more like an idiot because their car has safety features? I drove like an idiot even when my car didn't have ABS, and these days even though all cars I drive have ABS, I drive like less of an idiot.

Traction control is no use for driving like an idiot. I switch it off when I want to have some fun.

Re:Not more "safety features" please

[tophermeyer]

It seems to me that this is an "all or nothing" kind of thing. Either all vehicles are computer controlled, or none of them. My experiences seeing average humans interact with computers would make me not feel safe in my computer controlled car while panicky meatbags are on the road with me.

Re:Not more "safety features" please

[ElectricTurtle]

Humans might make slower decisions, but they have a much broader and more integrated matrix of perceptions and conceptions to draw from. Until AIs are strong enough to understand environments intelligently and intuitively as a whole rather than programmed to respond to a few set objects in a few set ways, a human decision and action will be necessarily more complete even if it is slower.

Re:Not more "safety features" please

[Anonymous Coward]

Slashdot users are not normal people. Before Power steering, ABS etc... how many people were eating breakfast as they drove. If they were around back then, I doubt celphones would be common use in vehicles of that time.

Re:Not more "safety features" please

[jeffmeden]

That's become the problem with ABS, traction control, airbags and many other safety features: make drivers feel like they're safer, they will drive more like idiots.

Never mind the fact that traffic deaths (in the US at least) have been decreased INCREDIBLY with the aforementioned technologies. Some do choose to drive like increasingly effective idiots, but not nearly enough to outweigh the safety benefits. I will go with the safety technology versus the notion that the sword of Damocles is effective at preventing accidents, thank you very much.

Re:Not more "safety features" please

[Eevee]

You don't ride taxicabs in Munich

Subsequent analysis of the rating scales showed that drivers of cabs with ABS made sharper turns in curves, were less accurate in their lane-holding behaviour, proceeded at a shorter forward sight distance, made more poorly adjusted merging manoeuvres and created more "traffic conflicts". This is a technical term for a situation in which one or more traffic participants have to take swift action to avoid a collision with another road user.[3] Finally, as compared with the non-ABS cabs, the ABS cabs were driven faster at one of the four measuring points along the route. All these differences were significant.

Re:Not more "safety features" please

[Dishevel]

Are cellphones car safety equipment now???

Re:Not more "safety features" please

[Anonymous Coward]

Completely agreed.
If anything, they should scare the living daylights out of people wanting to take up driving so they pay attention.

"So you want want to drive, eh?
YOU MIGHT DIE! EVER THOUGHT ABOUT HOW PAINFUL IT WOULD BE TO CRASH?!
IMAGINE YOUR DRIVING WHEEL IMPALED IN YOUR CHEST, SOUNDS SORE, RIGHT?
So, rest your foot on the accelerator here, and breaks here, take the handbrake off and gently push the accelerator."
Meanwhile, the person learning quite literally pissed his or her self.

Re:Not more "safety features" please

[Anonymous Coward]

Insurance rates will probably fix that, assuming that the robo-car is orders of magnitude safer than the manual car.

Re:Not more "safety features" please

[cheekyjohnson]

Yet still, the number of accidents caused by human drivers is staggering. An AI, if made properly (as you said), would outdo a human and make far fewer mistakes. I really don't think that absolutely groundbreaking AI is needed for this, though.

Re:Not more "safety features" please

[shentino]

A human still has to program the sucker.

And I would much rather have a human I could watch and monitor than an AI concealed in an opaque chip that I would just have to trust implicitly.

I barely trust people as it is even when I can watch them.

Re:Not more "safety features" please

[balbus000]

I don't think we'll ever have completely automated cars. Even if they reduced accidents by 80%.

People will see a big difference between getting in an accident by human error or by a malfunctioning computer.

The fact that it's completely up to the computer will make it feel like playing a slot machine. Sure there are times when human error by someone else is completely out of your control, but I think people will perceive it differently.

Re:Not more "safety features" please

[cheekyjohnson]

"A human still has to program the sucker."

Yes, and while humans do make mistakes, that is precisely why it would be rigorously tested before mass producing it.

"And I would much rather have a human I could watch and monitor than an AI concealed in an opaque chip that I would just have to trust implicitly."

The AI would do exactly as it was told. In reality, there would likely be far fewer accidents if it was, again, made properly.

Re:Not more "safety features" please

[Kjella]

I drove like an idiot even when my car didn't have ABS, and these days even though all cars I drive have ABS, I drive like less of an idiot.

Translation: You grew up and you got experience. If we could fill the roads with 40 year olds with 20 years driving experience, the accident rates would go way down but it doesn't work that way.

Re:Not more "safety features" please

[ElectricTurtle]

"Staggering"? Hyperbole. In the first place rates vary culturally. Iceland has 3.8 fatalities/year per 100000 people, less than a quarter of the rate of the US at 12.3, which in turn is about a quarter of the rate of the worst country for fatal car accidents: Eritrea at 48.4. Even that highest rate is still only 0.0484%/year, and at the risk of sounding cavalier about human life, I wouldn't call that staggering.

Also your assertion that the AI problem would not require a groundbreaking solution is founded on what knowledge? I think you vastly underestimate the problem. Example scenario: a vehicle is traveling on a rural road in the winter around a tight, blind turn on a mountain road. Suddenly, another vehicle appears heading toward the first in the middle of the road. Does the AI in the first vehicle know it's winter and black ice may interfere with braking? Does the AI know that turning out of the other vehicle's path toward the mountainside may result in the vehicle flipping? Does the AI know that if it turns away from the mountain to avoid the other vehicle that it could cause it to plummet to its doom?

Let's back this off a bit, instead of a mountain, it's a hilly region and the same scenario, turning toward the hill would cause the same risk of flipping, but turning away would probably be rough but survivable. The AI turns away, but the hill is too steep and icy to brake effectively, does it know how to steer under such conditions? Does it know where to steer? Let's say there's a body of water down there, does it recognize that as a hazard to avoid? What if the water is frozen? Does that appear as a solid surface to the AI? What about at night? On and on and on.

Human intuition and integration is so powerful we don't think about most of these things consciously. We have the capacity to act with so many key factors understood naturally and relationally. AI will get there, that's inevitable, but it will be decades more before that happens, and when it does it will be "groundbreaking".

Re:Not more "safety features" please

[Hognoxious]

I drove like an idiot even when my car didn't have ABS, and these days even though all cars I drive have ABS, I drive like less of an idiot.

Perhaps you like got older or something. Maybe you even grew up a little, who knows?

This, children, demonstrates the importance of a control group in experiments.

Re:Not more "safety features" please

[Anonymous Coward]

If it can hold a constant speed and stay between the lines than it's better at driving than a lot of lisenced drivers I know. I'd be much more inclined to trust an AI driver than a random person.

Re:Not more "safety features" please

[somersault]

I've grown up very slightly - in large part due to taking an advanced driving course and having a subsequent driving ban due to speed, rather than simply because I'm older - but the point is that I definitely don't take risks just because I know my vehicle is "safer".

I suppose if someone doesn't actually know what ABS does then they may be more inclined to allow less time for braking etc, but ABS can actually increase stopping distance, for the sake of retaining directional control.

An interesting point from the article someone linked above is that ABS reduced the number of accidents "caused by the driver", but increased the number of accidents which "they had no control over". I'm presuming this to mean that they were better able to avoid obstacles and retain control of their vehicle, but in situations where there was nowhere safe to turn, the extra stopping distance caused them to crash.

Re:Not more "safety features" please

[cheekyjohnson]

" Does the AI in the first vehicle know it's winter and black ice may interfere with braking? Does the AI know that turning out of the other vehicle's path toward the mountainside may result in the vehicle flipping? Does the AI know that if it turns away from the mountain to avoid the other vehicle that it could cause it to plummet to its doom?"

Obviously, it should. I didn't underestimate the problem. I knew exactly what you were trying to say. I mean, sure, it will require more knowledge of AI than what we have now, but I don't really consider that "groundbreaking."

Re:Not more "safety features" please

[robot256]

Humans might make slower decisions, but they have a much broader and more integrated matrix of perceptions and conceptions to draw from. Until AIs are strong enough to understand environments intelligently and intuitively as a whole rather than programmed to respond to a few set objects in a few set ways, a human decision and action will be necessarily more complete even if it is slower.

But sometimes a complete solution is useless if it is too slow, and the limited perception of humans leads to suboptimal solutions. If the AI can sense more accurately react faster than a human, it doesn't need as much predictive capability. Why do you need to anticipate a pedestrian running into the road if you can stop as soon as they actually do? And I think you underestimate the quality of AI these days. But in any case, AI and human drivers have different strengths, and I believe the best solution will take advantage of both of them. I for one would love a car that can both drive itself if I let it, and watch out for my mistakes when I am in control. I don't think I would ever be reading the paper behind the wheel even if it was driving itself, though.

Re:Not more "safety features" please

[somersault]

Well, put it this way. In the same timeframe that I had my Golf (no ABS, no airbags, no aircon, no nothin!), I also had access to my mum's car which had ABS and was faster. I was 19. I didn't consciously take more risks simply because the car had ABS and airbags. Certainly I have become a better driver with experience and further training, but I don't rely on electronic driving aids to keep me on the road. The only times my ABS has been active have been in wide open areas like carparks or muddy areas where I'm purposely messing about.

Re:Not more "safety features" please

[somersault]

Meant to say "only times my ABS has been active in the last year". I have had it kick in on real roads on occasion, usually in very wet or icy conditions.

Re:Not more "safety features" please

[cheekyjohnson]

"Even if they reduced accidents by 80%."

Yeah, humans are illogical like that. Now to wait until the people against technological advancement die off...

Re:Not more "safety features" please

[ElectricTurtle]

I really wish an AI developer/researcher were around to punch you in the dick. I can only intuit the problems, but I do know as most do that imitating human awareness and decision-making capacity in dynamic environments is one of the holy grails of AI. Hell, there have been countless projects and contests year after year since AI development has existed, none ever fully achieving that goal. To continue to trivialize it as not "groundbreaking" is to demonstrate a fundamental ignorance of the field even as an abstract.

Re:Not more "safety features" please

[robot256]

Also your assertion that the AI problem would not require a groundbreaking solution is founded on what knowledge? I think you vastly underestimate the problem. Example scenario: a vehicle is traveling on a rural road in the winter around a tight, blind turn on a mountain road. Suddenly, another vehicle appears heading toward the first in the middle of the road. Does the AI in the first vehicle know it's winter and black ice may interfere with braking? Does the AI know that turning out of the other vehicle's path toward the mountainside may result in the vehicle flipping? Does the AI know that if it turns away from the mountain to avoid the other vehicle that it could cause it to plummet to its doom?

There are many different types of driving situations, some more difficult than others. Why must an AI be able to cope with all of them for it to used at all? I would hope that the operator of such a vehicle would understand the limits of its AI and revert to manual control in situations like you describe. If it were able to out-perform humans in the most common settings--interstate and city during fair to somewhat severe weather--then I see no reason to ban its use outright. Heck, in severe weather the computer may have an advantage--it has radar that can see through fog and rain in all directions, it has the actual data from traction sensors, and yes, it knows where the walls and cliffs are and not to hit them.

Re:Not more "safety features" please

[Anonymous Coward]

Windows Vista taught us that this feature would be immediately turned off and/or ignored by the vast majority of users, including most of the ones who most need the coaching. If you try to teach a non-captive audience with annoying methods, you will lose them.

Re:Not more "safety features" please

[jeffmeden]

Hold your breath for the odds that one (or a handful) of companies want to be directly responsible for the 20% of accidents that are left over. It's not about being a luddite, it's about not wanting the risk involved. With individual drivers making mistakes (that admittedly add up to a lot of accidents) you still have a low risk/responsible party ratio. If every single fatal accident that happened (even if the count were reduced by 80%) resulted in a lawsuit claiming the car was at fault because the human inside had little/nothing to do with operating it, the car companies wouldn't be around for long. We have already witnessed this with things like preventive braking and adaptive cruise control. The technology has been available for decades to allow it to supplement the driver's ability, but the corporate lawyers took one look at it and said "haaaaaaaahaha go ahead and sell one, WE won't get rich but all the other lawyers in the country will!"

That is the illogical attitude you are referring to.

Re:Not more "safety features" please

[Grygus]

I think it's inevitable.

If nothing else, once you have kids growing up with self-driving buses, they won't think twice about self-driving cars and will roll their eyes at their parents' concern.

Re:Not more "safety features" please

[Anonymous Coward]

granted, the decreas in trafic accidents also mirrors the aging demografic.

Re:Not more "safety features" please

[ElectricTurtle]

In general, I agree, as Omar Bradley once said: "The second best decision in time is infinitely better than the perfect decision too late."

However, predictive capacity isn't the point, and in fact I never mentioned it. Where humans excel AIs is interpreting layered dynamic environments intuitively. Your example of the kid running into the street is nice and simple. Obstacle presents itself, car immediately brakes, problem solved, eh? Oh yeah? What if there is a car tailgating at speed? Does the AI split the difference between the braking distance between it and the kid as well as the car behind? What if the road is icy? Does it know to maneuver instead of or in addition to braking? (The next one's a stretch, but bear with me:) What if the kid isn't actually running into the road at all, but quickly moving one side to the other while standing on a flat bed trailer? As an object it would be registered as 'human moving into path' but in fact it wouldn't really be, would the car still brake?

It's very easy to come up with pat simple scenarios and make a program that simply executes little when A do B routines, but combinations of conditions do not always result in a combination of responses. Humans can intuit their way through layered conditions in a dynamic environment in a way that makes up for the time it takes to come up with the solution.

Re:Not more "safety features" please

[Firethorn]

Personally, I think that my question becomes one of 'what do you do when somebody successfully designs an AI driven vehicle that is statistically safer than a human driver and cheap enough to be affordable?'

I'm not saying that it'll be perfect. But it'll never get tired, drunk, or distracted. It may not be able to handle all situations, but those it can it handles far better than humans.

How do you settle liability in such cases? Would you still say it's the 'drivers' fault? The owners*? The car company or the AI maker?

It might be best, assuming high initial cost and like I said, liability concerns, to have them in taxies and long distance trucks first. That way you have a company to absorb the liability while saving money on drivers(assuming you can get rid of them) and rack up enough hours to justify the system relatively quickly.

*While the insurance company cuts him a deal on his rate for having the system.

Re:Not more "safety features" please

[nmos]

I really don't think that absolutely groundbreaking AI is needed for this, though.

Maybe not to handle a modern well marked highway but It'll be a while before they can handle the typical neighbourhoods without lane markings and stop signs overgrown by trees etc much less construction zones and parking lots. IMHO the real question is if we develop cars that can deal with 90% of the driving will our driving skills become so poor that we can't properly handle the 10% where humans are still needed.

Re:Not more "safety features" please

[HiThere]

There are real problems, but you didn't list them. The problems you listed are refinements of the current system. Real problems are what do you do if an animal runs in front of the car? A child? A child in a Halloween costume? How much damage should you take to avoid each? And how do you distinguish cases 1 and three?

Things like that. Complex object recognition, particularly when in disguised form, is an unsolved problem, and *does* require a breakthrough. Driving skills, terrain recognition, standard obstacle recognition are refinements. (And "black ice" isn't necessarily hard to see. If you're combining radar with sonar other things will be difficult, but probably not that.)

Re:Not more "safety features" please

[ElectricTurtle]

[...] it knows where the walls and cliffs are and not to hit them.

You're missing the point. It's a Kobayashi Maru-style scenario. Of course an AI isn't going to just ram itself into a rock wall, but if it has to make a decision between an oncoming vehicle, a rock wall, and a cliff, what is it going to do? Could it really come up with the maneuver that maximally increases survivability? (Which, by the way, is progressive braking swerving as close to the wall-side as possible, hoping that the other vehicle compensates in the opposite. However, if the oncoming vehicle is on a ballistic trajectory (due to hydroplaning or ice skidding) toward the wall side, then the opposite response is called for, including accelerating rather than braking because you want to choose the path of non-intersection because if you succeed in stopping near the edge of the cliff and the other vehicle bounces off the rock wall and pushes you off the cliff... well... you're dead. Further, if the vehicle is in a tractor-trailer configuration, one has to compensate for the potential of trailer swing during its breaking/swerving. That has to be observed and reaction adapted. Can an AI do this yet? Hell no.)

Re:Not more "safety features" please

[cheekyjohnson]

No. I did not say it wasn't impressive or that it wasn't extremely difficult. I merely meant to say that, compared to some other things that an AI could be made to do, it wouldn't be as difficult. I did not mean to imply that it wasn't impressive or incredibly difficult. I know that it is.

Re:Not more "safety features" please

[Anonymous Coward]

You REALLLY think that humans are capable of making such decisions within 0.5s they have for reaction? Maybe race drivers are, but normal driver hasn't been thought to react properly and in most cases will just direct the car in opposite direction of obstacle.

Re:Not more "safety features" please

[ElectricTurtle]

Yes, there are additional problems, and no, that doesn't make "driving skills" and "terrain recognition" less than "real problems" or mere "refinements". Christ, "driving skills" is such a broad category it could include practically everything, and "terrain recognition" is quite a similar problem to "complex object recognition". You're never going to wholly solve one without wholly solving the other, and I did touch on it briefly (iced over body of water vs solid ground, what if the ice is dirty?).

And "black ice" isn't necessarily hard to see.

Tell that to all the dead people on WA's Highway 2.

Re:Not more "safety features" please

[Spectre]

Quite true ... and with raising the age a person can be licensed to drive (used to be 14 in quite a few states of the US, it is now almost universally 16, with a few holdouts that allow restricted licenses a bit younger).

Re:Not more "safety features" please

[TheTrueScotsman]

Only the most skillful or lucky human driver would the survive your mountainside scenario. The experience of assisted driving aids so far (ABS etc) is very positive apart from the decrease in driver awareness such aids engender.

I don't think fully automated driving is very far away apart from social and legal acceptance factors - anti-skid, weather sensors and satellite relief maps combined with some sensible heuristics developed by skillful drivers might even make the mountainside situation survivable (of course, a robot car coming the other way shouldn't be in the middle of the road anyway because it won't have been drinking).

Re:Not more "safety features" please

[Firethorn]

"Staggering"? Hyperbole. In the first place rates vary culturally. Iceland has 3.8 fatalities/year per 100000 people, less than a quarter of the rate of the US at 12.3, which in turn is about a quarter of the rate of the worst country for fatal car accidents: Eritrea at 48.4. Even that highest rate is still only 0.0484%/year, and at the risk of sounding cavalier about human life, I wouldn't call that staggering.

You're looking at fatalities, he stated 'accidents'.
Worldwide, 2004, 1.2M people killed in crashes. 2.2% of deaths. On a worldwide scale, I'd rate anything capable of breaking single digits in death rate as 'significant'. Still, there's a lot more to auto accidents. Most survive - 50M injured a year. In the USA alone, 46k/year dead, 2.4M injured, sixth leading preventable cause of death. Canada - 48% of severe injuries.

Some more stats pulled off the auto accident wiki:
57% due solely to driver factors
97% have driver error, intoxication, and 'other human factors'
Global cost estimated at $518B, USA $230B

So we'd be looking at a decent AI being able to eliminate something like 50% of accidents right off the bat. Presumably, this would save around 600k lives a year and prevent 25M injuries, save the USA $115B.

That would be around $380 per person in the USA, per year. 254M cars, estimated life of a decade, if the system cost could be gotten down to $4k, it'd be totally worth it if it chopped the accident rate in half. Make the worst drivers get them, and you'd break even quicker yet.

That's not even figuring in the cost of a human life - which I'd place around $5M, which would mean $3T for the world, $230B for the USA, so I'd say that even $8k would be worth it.

I think you vastly underestimate the problem. Example scenario: a vehicle is traveling on a rural road in the winter around a tight, blind turn on a mountain road. Suddenly, another vehicle appears heading toward the first in the middle of the road. Does the AI in the first vehicle know it's winter and black ice may interfere with braking? Does the AI know that turning out of the other vehicle's path toward the mountainside may result in the vehicle flipping? Does the AI know that if it turns away from the mountain to avoid the other vehicle that it could cause it to plummet to its doom?

Hmm... I think you underestimate how much of a difference the reaction speed of a computer can make, and overestimate the amount of figuring your average driver does.

Presumably the car would 'know' it's icy out, just like cars today detect such things for the purposes of traction control systems. Drop knowledge would be doable, but I think we're going to need to improve the accuracy of our maps around 10X for it to really work.

Re:Not more "safety features" please

[dwandy]

I don't think we'll ever have completely automated cars.

Fully automated cars are inevitable.
We can debate how they will come about/be implemented (1), how long it will take(2), how they will work(3), if there will be insurance requirements and who would pay it(4), but their arrival is guaranteed.

...by human error or by a malfunctioning computer.

Cars today are already almost 100% computerized. The controls you touch tell a processor what servos to enact.

...will make it feel like playing a slot machine.

I'm not planning on letting M$ build these things...but even if Sync eventually Borgs it's way into car control (instead of feature control) I just won't buy that one ...

Ultimately most driving is monotonous work that requires constant attention. Humans suck at that sort of activity; machines excel at that sort of activity.

My guesses
(1) Incremental: First increment is here: cars that park themselves, lane departure warnings, etc. Next increments like "Sixth Day" where on the highway it was automated and in the city there was manual control. Farm equipment will be an early adopter; trucks and buses will lead on the road.
(2) Well, partly automated is here, and I suggested that the children being born today would be the last full generation that drove themselves. The next generation will be the crossover. Of course I said this two years ago and that is starting to look pessimistic.
(3) Sensors. Lots of Sensors. As a child I thought we would need a "wire" in the road to tell the car where it was but that seems silly now.
(4) Insurance will continue much as it does today, there will be an annual fee paid by the driver based on the risk of their vehicle; don't worry, the insurance and car companies aren't going to kill their golden goose.

Re:Not more "safety features" please

[robot256]

True. But how many humans could actually do that when pressed? Not all of them, that's for sure. Yet they are still allowed to drive as much as they want, since they are willing to take the risk or avoid the situation. The same could be true of an AI. It could simply refuse to drive on what it knew to be prohibitively dangerous icy mountain passes. Or your perfectly cognizant human would recognize the situation and take over from the AI, which prior to this had done a perfect job of avoiding walls, cliffs, and skidding.

Sure, there is infinite room for improvement of AI, but that is hardly a reason to oppose its adoption as long as we understand its limitations.

Re:Not more "safety features" please

[JWSmythe]

    I was wondering how AI's would deal with the same thing. Around an international airport that I live by, they were changing the highways. About 5 highways go to or pass through that area. Pretty much over the last 6 years, all the roads have been torn up and laid back down. New ramps and overpasses have been installed. It was confusing enough that even locals who don't drive that route every day would get confused on where to go. Even when I drove it every day, sometimes I'd find that where the road went straight, today it went off somewhere else.

    For about a year, there was one point where they had shifted all the lanes to the left pretty quickly They didn't paint over the old lines, so you had lane lines going a little to the right, and newer lines going to the left. It wasn't always clearly distinguishable, except I knew where the old lanes were from habit. It was very "interesting" to drive in the rain. With enough rain on the road, the lines would disappear (covered in water), so you had 3 lanes of cars doing 70mph, and you knew everyone had to shift. I don't know how many accidents happened there, but I'm sure there were at least a few.

Re:Not more "safety features" please

[Mirey]

I can assure you that traffic accidents would be significantly reduced if, by law, there had to be a spear head on the steering wheel aimed at the drivers head.

Re:Not more "safety features" please

[nmos]

Never mind the fact that traffic deaths (in the US at least) have been decreased INCREDIBLY with the aforementioned technologies.

Of course, the average age of drivers has also changed during the same period. How can you be sure the change in traffic deaths has to do with tech. rather than the number of older/safer drivers on the road?

Re:Not more "safety features" please

[houghi]

So you don't believe in Darwins law?
If we have no security in cars, within a few generations, all people will be able to jump out of the way of cars.

Re:Not more "safety features" please

[JWSmythe]

    You've also grown up, which changes things. Our teenage immortality is abandon for the reality that we are going to die soon, so lets not make it any more painful than necessary.

    I was in a car accident in the 80's. I learned then that vehicles don't stop as fast as you want them to, and pain can last an awful long time (permanent damage to the fleshy operator). So I drive more and more conservatively as time goes on. But... Incidents have happened that were outside of my control. You can't control nor predict random actions of everyone on the road, especially in areas with heavy traffic. I've been in a few more accidents, which leaves me wanting to always drive where there are no other cars.

    I know that more accidents will hurt me worse. After my last MRI, the doctor gave me very simple advice. If I planned on continuing to walk (or survive), don't get in any more accidents. If there's a hard hit, it's likely I'll either be crippled from the waist down, or from my neck down. The later would probably result in a rather short life expectancy (i.e., seconds).

    Still, I do take my car racing (closed courses, not open street racing). I know the physical limitations of my car, how well the ABS works, etc, etc. If you put a mark on the road, I can be doing 65mph, and then stand on the brakes to stop within an inch of that mark. Do I do that as normal practice? Nope. Have I had to in an emergency? Several times.

    As for traction control, it just annoys me. My car (2000 TransAm WS/6) doesn't have it. My friends car (2000 Camaro SS) does. It doesn't perform all that well with the traction control on. Well, for obvious reasons. It's limiting the possibility of tire slip. I'm practiced in controlling that myself, and can "drift" (I hate that term), but stay within the capabilities of the vehicle.

Re:Not more "safety features" please

[nmos]

You don't ride taxicabs in Munich

Very interesting link. Thanks!

Re:Not more "safety features" please

[robot256]

It's very easy to come up with pat simple scenarios and make a program that simply executes little when A do B routines, but combinations of conditions do not always result in a combination of responses. Humans can intuit their way through layered conditions in a dynamic environment in a way that makes up for the time it takes to come up with the solution.

It's also very easy to ramble on with hypotheses without knowing anything about the AI in question. How do you know it's simply a collection of if A then B routines? Intuition is not magic: it is simply the ability to run an accurate predictive simulation of your environment. This is entirely within the scope of an AI, and in question is only the relative accuracy of the predictions.

An AI can identify objects and classify them into cars, people, animals, unknowns, etc. Each class of objects will have a different behavior profile--a person moving quickly can turn sharper and stop quicker than a car moving quickly, for example. Unknowns would be treated with caution, obviously. For each object it could calculate a range of possible trajectories and account for them in its driving patterns--the kid running on the truck would have a large uncertainty, while a car on a cross street would maintain roughly the same path.

Basically, the only way a vehicle AI could work is as a stochastic simulation constantly being updated with radar and camera data. Within that simulation the car can plan its own behavior just like a human would. When we have systems like this, then we will have viable vehicle AIs. We have all the pieces already, they just need to be perfected and integrated, which is precisely what these researchers are doing.

Re:Not more "safety features" please

[Anonymous Coward]

My uncle is an AI researcher, and I used to share an office with one. Neither one seemed likely at any point to punch someone in the dick.

Re:Not more "safety features" please

[jeffmeden]

The accident rate may go down but you can bet that the fatal accident rate will go up... Is that what you want? To kill anyone who makes the slightest mistake? They got rid of death-trap cars (those with poorly designed gas tanks, plate glass windows, hard metal dash boards, little/no door frame rigidity)... And I don't think I remember "they caused too few accidents" being cited as a reason.

Re:Not more "safety features" please

[Anonymous Coward]

I think this is exactly the point. If we're requiring those kinds of extreme abilities before we allow the AI to be used, we should be requiring them before we issue a license.

Overwhelmingly more dangerous driving is due to driver impairment (drunk, distracted, sleepy, sick) than environmental factors. These situations are dangerous to others, not just the passengers, and while we prohibit them and try to reduce them through enforcement, this takes them off the table entirely.

Furthermore, the icy, mountainous situation described would likely be safer with the AI as you would be more comfortable going slower if you could be reading a book or watching a movie or browsing the web or focusing on the other passengers, rather than being stuck focusing primarily on the road.

Re:Not more "safety features" please

[Anonymous Coward]

As development goes it DO work that way. Why would the assembly line equip the car with an AI without drivning experience when they can equip it with an AI that has the equivalent of 20 years driving experience?

Re:Not more "safety features" please

[mcgrew]

News flash -- idiots drove like idiots before the safety features, too. What's making them more dangerous these days is cell phones. But even accounting for cellphones, there are still far fewer deaths per driven mile than there were before the safety features.

However, I, too would like a self-driving car.

Re:Not more "safety features" please

[apoc.famine]

A woman died near my parent's house a couple of years ago, at 3 in the afternoon, on a dry road, doing 60+ around a corner marked 30. She started to slide, overcorrected, went off the corner and on the way down the bank, rolled the car 90 degrees, and wrapped it around a tree roof-first. Happened to be the same corner my sister went off (headed the other direction) in the rain at 50+ about seven years prior. She only took out the sign marking the corner and a telephone pole with the nose of the car, and was unhurt.

I really, REALLY don't see the scenario the GP listed above happing on that corner more frequently than these two totally preventable accidents. When it comes down to it, yes, the AI will cause accidents. But probably far less than people cause themselves doing stupid shit like going ridiculously fast around sharp bends, putting on makeup while driving, texting, reading newspapers and books, getting/giving blowjobs, etc.

Re:Not more "safety features" please

[Gubbe]

You are both missing the point.

Suddenly trusting a new chip that has to have been tested 100% reliably never enters the picture.

You're not going to wake up tomorrow and see an ad in the paper saying "New Toyota Nauris: Now without steering wheel!"

No, it'll be gradual. Already we're seeing park assist, lane assist, adaptive cruise control, all that. Bit by bit they add these features. First as expensive options, then as standard and later when they are thoroughly road-proven, they come as features that you can't even turn off. This has already happened with ABS and ESP. In most new cars you can only turn them off momentarily in low speeds for special circumstances.

This will slowly happen with all computerized assistance features until at one point the car can drive itself in pretty much all conditions and you only grab the wheel if you feel like it. And then, when your private car automatically takes you to work while you shave and read a book, and when manual driving becomes a boring chore that limits you to the slow detour routes where old manual steering cars are still allowed, you'll find yourself happily buying a car without a steering wheel, just like you now happily buy a car whose engine you don't have to tinker with. Those who still enjoy tuning and piloting a hunk of metal will take up karting or motocross as a hobby.

All these current driverless cars are, are development platforms, from where mature enough components trickle down to the production lines. It won't be a revolution. Now, obviously the same idiots that follow their satnavs off a cliff today will at also misuse these individual features tomorrow. There will be lawsuits, the users will get the blame because they were not following instructions and the manufacturer will make the components more idiot-proof. And then you don't drive your car anymore.

Re:Not more "safety features" please

[cheekyjohnson]

I know. I meant that when it becomes available and has been tested.

Re:Not more "safety features" please

[Anonymous Coward]

Actually, you can have both, to some extent.

The grandparent isn't entirely accurate.

What studies found is that drivers respond to the _feeling_ of safety by behaving more dangerously, but we can make them actually safer, and the lesson is that we need to concentrate very much on actual safety and either not try to make it feel safer, or perhaps even the converse.

For example, airbags are good because there isn't an omnipresent "safer" feel. You drive like usual, and when you hit something the world is suddenly very loud and you're in pain, but you haven't smashed your face into a steel pillar. This doesn't encourage more dangerous driving.

ABS causes dangerous driving from older drivers with no experience of it. They're expecting a margin that no longer exists - where braking causes sliding. If they brake suddenly without sliding, they think they're still inside that margin and this encourages them to believe they were still safe, when a younger driver might think "shit, I nearly hit that" and won't read anything into the lack of sliding because they're used to ABS.

The safety harness used in road cars is likewise OK on drivers who grew up with it. They basically forget they're wearing it, and in a near accident when the belt engages they realise the seriousness of that and will "back of". A multi-point racing-style hardness might encourage recklessness, but essentially no-one uses those in road cars.

On the other hand, traction control gives the false impression to all but the most experienced users that they have unlimited grip. It's definitely worse for this than ABS. Of course in an accident the traction control may "go off a cliff" and they suddenly lose all control over the car. So you will see drivers right at the edge of their car's performance envelope who are unaware how dangerous that is. One unexpected hazard and they'll be in a crash or worse.

Car designs can give the feeling of safety without making the occupants much safer. This is a recipe for disaster, the driver is more reckless, while accidents are already more likely or more serious. Some SUVs are prone to this, which is particularly appalling as they're often used with children as passengers.

Something that doesn't make drivers feel much safer, but does have a huge safety implication (and is responsible for a big reduction in deaths) is road design. Examples:

Wider lanes mean less accidents and less deaths, as traffic in an adjacent lane (whether passing or oncoming) has more room to compensate if you cross lanes through inattention, or in swerving to avoid a hazard.

Design of safety barriers and road furniture makes a big difference. Break-away sections, replacing metal structures with plastic ones, ensuring things crumple, bend, and contain rather than acting as ramps, shattering, or penetrating the accident vehicle. This is mostly invisible to the driver until they're in an accident, so it doesn't cause undesirable reckless behaviour.

Re:Not more "safety features" please

[grumbel]

Also your assertion that the AI problem would not require a groundbreaking solution is founded on what knowledge?

On the DARPA Grand Challenge they went from basically nothing to cars that can drive in an normal urban environment in just three years. Getting all that technology consumer ready might take another 10 or 20 years, but there really doesn't seem to be any fundamental problem with getting cars to drive themselves. The military will probably have driverless trucks in the next few years.

Does the AI in the first vehicle know it's winter and black ice may interfere with braking?

Yes, why wouldn't it? Companies can be stupid, but they probably figure out by themselves that their new AI driven car should be tested in winter and handle such obvious situations.

Does the AI know that if it turns away from the mountain to avoid the other vehicle that it could cause it to plummet to its doom?

That will probably be market clearly on the map and picked up by the LIDAR sensors.

it's a hilly region and the same scenario, turning toward the hill would cause the same risk of flipping, but turning away would probably be rough but survivable.

Do you really think a human is pandering all those option instead of just doing the next best thing he can do to avoid the other car?

And of course the main issue with your scenario: It was caused by a stupid human driver in the first place. If the other car would be AI controlled it would have known the current road conditions long before it got into trouble and would have driven at appropriate speed. Thus never coming in collision danger with the second AI car. And of course the cars might communicate with each other to coordinate the decisions in emergency situations, so they don't need to guess what the other car will do, they know it.

Re:Not more "safety features" please

[StikyPad]

Yeah, that's a pretty ridiculous scenario. I'd be more interested to see an AI distinguish between a plastic bag blowing around and a truck tire that just went airborne.

Re:Not more "safety features" please

[Belial6]

You are correct, and your point about the other car is very important. The GP basically suggests that cars should not drive themselves because human drivers are worse at driving.

Re:Not more "safety features" please

[Belial6]

It is the experience that matters. I'd take a road full of 20 year olds with 5 years experience than a bunch of 40 year olds with 0 any day.

Re:Not more "safety features" please

[HiThere]

The dead people on highway 2 weren't using sonar and radar to detect obstacles. Different things present problems at different wavelengths.

"Driving Skills" can, indeed, be interpreted at widely as you say. I though it was fairly clear that I was talking about things like "recognizing crests and troughs, impenetrable obstacles, places where one doesn't get sufficient traction, etc." and "given the current terrain, at what speed can one stop before encountering an obstacle". (This should be understood as including "how fast is it safe to take that corner?", "What's my stopping distance, including the time it takes to recognize a problem?", etc.)

OTOH, I'm not an expert in the field. However, I expect that you are not either. And there are existence proofs that things approaching marginally acceptable are currently possible (though not yet economic). Automated vehicles now operate in many controlled environments, like warehouses, hospital corridors, etc. These are not even approaching state of the art. They were nearly state of the art two decades ago.

State of the art now is controlled tests in the open, as, e.g., driving from Milan to China. That test was well controlled, and can't be considered "free driving", but it demonstrates many of the features that are necessary. And also where some development work is necessary. The recent DARPA test involving driving in an Potemkin Village with pedestrians (who were all adults) was also quite impressive. And that's last year's state of the art. (Note that a flawless state of the art performance demonstrating capabilities superior to the average holder of a class D license wouldn't suffice to cause wide distribution of automated vehicles. That requires that it also make economic sense.)

P.S.: If sonar + radar couldn't detect "black ice", then that would just mean that some other sensor would need to be added to the mix. I guarantee that something can detect it. Perhaps infra-red Lidar. Or radar at a mix of wave-lengths...including infra-red. Or, if nothing else, spectral analysis. People are built with the sensors that were most useful out of those available when apes were evolving. Black ice detection at high speed wan't one of the selection criteria, so it's no surprise that we don't handle it properly. The surprise should be that some much of what we sense is so generally useful. (And even that is understandable, when properly considered.)

Re:Not more "safety features" please

[Thing 1]

Nice image, thanks: I picture that tire coming towards the car, and it's bouncing, and the car brakes suddenly (communicating this to the car behind it and others, so as not to cause an accident), the tire bounces up higher than the car, the car accelerates to scoot under it, and the tire bounces again behind it. The next car repeats this, and so on.

This is somewhat like the Internet, which routes around damage; the self-driving cars will route around damaged vehicles on the road, driving us to our destination as fast as possible, given the conditions (weather and accidents) -- but won't have any rubbernecking. Well, that is, people in the vehicles will be free to look in whatever direction they care to; the vehicle will still travel at the optimal pace to its destination.

Re:Not more "safety features" please

[Ag3ntAzn]

I agree. Human drivers would make irrational decisions so it would be harder to make an AI if there was a mix of human and AI controlled vehicles. If all cars were AI controlled, the AI wouldnt need to react to those irrational decisions that a human might make.

Re:Not more "safety features" please

[Fat Cow]

Do you have a reference for this decrease in US traffic deaths and some indication that the decrease was caused by these safety features?

I ask because I've previously read (in a book called "Risks" that I can't find on amazon) that only seat belts clearly increased safety, all the others were marginal, statistically insignificant or made things worse. Their hypothesis was that people drove more recklessly to compensate.

wonder how they were fueled

[Anonymous Coward]

wonder how they got fuel and whether fuel usage and estimation was automated as well?

Re:wonder how they were fueled

[Petskull]

"Accidents were few but far from nonexistent, though I suspect the greater stress was from the fact that these vans had to charge 8 hours for every 3 hours of driving."
http://www.crunchgear.com/2010/10/28/driverless-vehicles-complete-trek-from-italy-to-china/

http://edition.cnn.com/2010/TECH/innovation/10/27/driverless.car/index.html
"We weren't worried about not making it," though, Broggi said. "This big trip was an intermediary step in a longer process. We have something new planned for 2012."

Uh oh..

Autonomous vehicles

[ledow]

The problem with autonomous vehicles is not what they can do successfully, it's what happens when they fail.

If I don't press my brakes in time to prevent an accident, I risk going to jail for dangerous / careless driving.
If the autonomous van doesn't... well... what? We can take the human "driver" off the road, sure, but that's not fixed the problem. So the second one person has an accident in an autonomous vehicle, you're looking at major liability and lawsuits directed towards the car manufacturer - whether or not it was their fault and whether or not a human driver could have prevented the accident in *any* car. That manufacturer now has to take responsibility for that car versus every idiot on the road, every pedestrian that runs out and everything that can confuse one of its sensors.

Autonomous driving *is* possible and quite easy - but we need autonomous roads to make it work, with nobody but the autonomous vehicles on it. Nobody, nowhere has actually built a real-life one of those on a real road that people would want to use because you have to use their vehicles to do it and you have to (indirectly) pay for that vehicle, that road, and any mistakes those vehicles make. And those roads don't and won't exist for decades if at all - or, more accurately, it's called the rail network. Automated rail networks are commonplace - London has the Dockland's Light Railway that has no drivers.

If you're going to have to build a road that only automated cars can use, and make some cars to use that road, you've effectively built a railway, or else you're putting billions of pounds of effort into avoiding obstacles and keeping to a strict lane when you could just make the thing run along a rail.

Why is there no call for an automated rail network? You can make it as fast as the super-express trains, it's very safe in comparison to any road, on established technology, you know it's not going to veer off the road, you can pack thousands of trains onto the rails if you do it right and take thousands of passengers in each etc. But instead, people honestly think that it's more sensible to put an automated system of even the best technology on an open road with other idiots and do this on a one-person, one-car basis (hence millions of units and billions of pounds) with complete freedom over how it moves the car, among other traffic that will stop it ever doing anything a human couldn't do? It's ridiculous.

Stop wasting your time and build a personalised rail network when I can get into a "pod" or something, enter my destination and it would take me there on good, solid, metal rails and a bit of signalling. And I don't have to worry that it thinks the man walking along the street with a cardboard cutout is actually a small child running in front of the car, or that it doesn't spot a police tape which has been strung across the road to close it because of a pedestrian parade further up the street.

An automated car has to have a human in it. It's the best call ever made on the introduction of a new technology so far. An automated car needs exclusive automated roads to every destination in order to work anywhere near effectively under autonomous control - that's called a railway and any more "transportation routes" being built just for automated cars is a fantasy world in a modern city. Automated cars have been shown to crash WHEN DEMONSTRATING how they were uncrashable. An automated railway already exists and works perfectly and has an excellent safety record. Use it.

Re:Autonomous vehicles

[caluml]

Stop wasting your time and build a personalised rail network when I can get into a "pod" or something, enter my destination and it would take me there on good, solid, metal rails and a bit of signalling.

Indeed. A packet-switched transport system. Broadcast your destination via Bluetooth, "routers" can receive that and direct you the best way. The pods would be unpowered, but pushed/blown along - possibly compressed air?
If you had a system of tubes under the ground, and some sort of decent bearings, you could make it work. You could also have large "trunk"/"backbone" roads, which smaller roads joined. Basically, model it on the Internet. But without the packet loss, or routing loops. Or collisions.

Re:Autonomous vehicles

[Dr_Barnowl]

There are pilot projects of Personal Rapid Transit systems going on.

Re:Autonomous vehicles

[RivenAleem]

And keep the porn. There should be a mechanism whereby I can get off whenever I desire.

Re:Autonomous vehicles

[OeLeWaPpErKe]

If the autonomous van doesn't... well... what? We can take the human "driver" off the road, sure, but that's not fixed the problem. So the second one person has an accident in an autonomous vehicle, you're looking at major liability and lawsuits directed towards the car manufacturer - whether or not it was their fault and whether or not a human driver could have prevented the accident in *any* car. That manufacturer now has to take responsibility for that car versus every idiot on the road, every pedestrian that runs out and everything that can confuse one of its sensors.

As in every other situation - the owner of the car is responsible "by default" for any accident. (It's not because just about every leasing contract specifies the driver being responsible that this is always true). Obviously, you'd probably want to change this with contracts.

And frankly this "impressive feat" was pulled off by animals with barely a few tenthousand neurons, using much harder to navigate routes than the public road network. While I agree, given the current state of technology, it is unfortunately impressive. That AI is not actually capable of giving us something half as smart as a stupid dog is sad, though.

As for the magical system you're describing, it exists ! It's called "public roads". It requires things with the intelligence of a stupid dog (ie. not a very interesting job for humans, but done by humans for lack of alternatives). Now if we could just make programs that can be just a little bit smarter than the least of the mammals, that'd be great, and it would allow us to build your "pod system" for an investment of $0, losing none of the flexibility we have now. While your system, if we can make it work (it's a much harder problem than you state) requires a few Obama's ($1 trillion per 6 months, for 4 years).

Re:Autonomous vehicles

[jeffmeden]

Basically, model it on the Internet. But without the packet loss, or routing loops. Or collisions.

The success of the Internet thus far has been thanks to the fact that every few years we manage to develop another novel way to pack a digital signal into a tenth of the space it once occupied. If your series of tubes are to be successful, we are going to need miniaturization (of the passengers) and a LOT of it. Sadly the trend (in the US at least) is for the average passenger to get larger over time, not smaller.

Re:Autonomous vehicles

[Anonymous Coward]

There is a key difference between a highway for autonomous cars and an automated rail system. Namely, the last mile: you can get off of an autonomous highway and start driving your car on public roads (manually, even) but you can't do the same with a rail system. I don't think anyone is, at this time, floating a proposal for autonomous cars that MUST drive themselves.

The technology for autonomous driving is also the same for autonomous alerting - i.e. "let me drive, but tell me if I'm doing something wrong."

Re:Autonomous vehicles

[Kjella]

A car needs a semi-drivable asphalt / gravel / dirt road to drive on. There's endless miles of them that it wuold cost trillions to replace and billions to maintain per year. The vast, vast majority of people want a drive that takes them to the doorstep and ends in their driveway, not some railway station far off from which you need to carry all your belongings and goods. Most of those that can comfortably do without already take the bus / tram / metro / railway.

And even if you happened to be on the network, it still wouldn't replace the car unless all yours friends and family and shops and everything else you'd like to visit was on the network too. I don't have a car and as a result I need to have a taxi budget, if I needed it regularly I would without a doubt buy a car and an automated car system like you describe would be no substitute at all, exactly because all the odd places wouldn't be covered.

An automated driving system would have a huge advantage in that we could have it record all the video and sensor input to a black box that'd survive the crash meaning no more word against word. I would think as all the crap drivers got exposed through video recordings, the death and injury tolls and the insurance costs started dropping people would accept that it is not perfect but substantially better and can be continously improved unlike the average driver who is pretty much the way it is. If the US wants to be all legally retarded then it'll happen in Europe or some other area and eventually the US would get dragged along.

Re:Autonomous vehicles

[natehoy]

Basically, model it on the Internet. But without the packet loss, or routing loops. Or collisions.

Or a government kill switch.

Re:Autonomous vehicles

[cheekyjohnson]

"The problem with autonomous vehicles is not what they can do successfully, it's what happens when they fail."

I wonder what people do in the many, many occasions when human drivers make a severe error...

Re:Autonomous vehicles

[rockNme2349]

So the second one person has an accident in an autonomous vehicle, you're looking at major liability and lawsuits directed towards the car manufacturer - whether or not it was their fault and whether or not a human driver could have prevented the accident in *any* car. That manufacturer now has to take responsibility for that car versus every idiot on the road, every pedestrian that runs out and everything that can confuse one of its sensors.

I've thought about this problem for a while, and here is my guess how it will proceed. When cars started being made with cruise control, the responsibility in an accident still belonged to the driver. There are cars being built today which automatically apply brakes when they sense an oncoming collision, but in the event of a malfunction or accident, the human driver is ultimately held responsible.

I don't believe anyone is going to drop an autonomous car into the market, but instead it will simply be more and more iterations of the computer taking control. The human driver will always have a manual override though, and will be responsible for the accidents, simply because that was the status quo. My guess is by the time we do get autonomous cars, people probably won't be paying attention to the road since their cars are driving themselves fine anyway, but they will have signed a disclaimer claiming responsibility anyway. I do think there will be uproars when accidents do occur, like we have seen with the Toyota problem, but not for a long while after we have become comfortable with autonomous vehicles will any law change regarding responsibility.

Re:Autonomous vehicles

[Anonymous Coward]

Google invested in http://shweeb.com recently.

Re:Autonomous vehicles

[burkmat]

Finally, us, the geeks, will be the overlords! Disobey our commands, and we might just ARP-spoof your default gateway into a concrete wall!

Re:Autonomous vehicles

[master811]

London has the Dockland's Light Railway that has no drivers.

That's not entirely true though. Yes they don't have drivers per se, but they still have a trained staff member on board at all times, to manually control the train in the event of an emergency and to generally control when the train leaves and departs, - especially during peak hours, when the network is busy.

Re:Autonomous vehicles

[sznupi]

Rethinking our habits would be required with autonomous cars anyway (for starters - how to deal with 80-90% of drivers thinking they are in the top 50%?). We might as well not limit ourselves to just one wundersolution...

Not giving away the cities to cars, not building them primarily around the requirements of cars, would be a good start. Would help in not inhibiting bikes, too. Which BTW, in the form of folding(*) or "rental" bikes, nicely expand the utility of public transport ((*)and such bike is often a very handy addition to a car - easy to keep in the trunk most of the time, often gets useful at some destination to which only car travel is practical; but once there...). And certainly many more examples.

Re:Autonomous vehicles

[sznupi]

I do think there will be uproars when accidents do occur, like we have seen with the Toyota problem...

More precisely, uproars at perceived problems with autonomous cars. Certainly perceived chiefly by drivers - 80% of which think they are in the top 50%.

Re:Autonomous vehicles

[sznupi]

As was the case with elevators for a long time...even when not strictly needed.

Re:Autonomous vehicles

[sznupi]

Compressed air is really not a very efficient way of storing and transporting energy... (really, why would you want to throw away all the experiences with mechanical design of, well, cars?)

Re:Autonomous vehicles

[snspdaarf]

I wonder what people do in the many, many occasions when human drivers make a severe error...

They die. The driver gets sued. When Robodriver pooches something, it will be a godsend to the victim's family [lawyer] as they will get to sue $AutoManufacturer, $SoftwareCompany, and $ThirdPartySupplier, all of whom have more money than the owner of the vehicle.

Re:Autonomous vehicles

[Firethorn]

Why is there no call for an automated rail network?

There actually is - Personal Rapid Transit

These vans are attempts to design a system that can move on actual roads, with people still driving on them. Obviously, this is far more difficult than to design a constrained system.

Re:Autonomous vehicles

[global variable]

To their defense, I see this situation very much along the same lines as to why we still have legacy Intel x86 architecture in our new chips which obviously could benefit from completely re-hauling it. In reality Intel even tried to re-haul it with their Itanium line, but as well all know, that didn't catch on. Everybody already has software that they don't wish or need to replace. So in order for them to move forward and sell new stuff, they have to conform. Although I noticed it's tricky getting 16-bit stuff to work in Windows 7 (who remembers "You don't know Jack"?!) We're stuck with the roads whether we like it or not. But man, I would love a society where I didn't have to drive anywhere. We should petition this to be put into the moon colony long term design :)

Re:Autonomous vehicles

[houghi]

If I don't press my brakes in time to prevent an accident, I risk going to jail for dangerous / careless driving.

If that is your reason for pressing the brakes, then there is something seriously wrong with you. You mean if those were NOT consequences you would not break?
Why not say: If I don't press my brakes in time to prevent an accident, I risk hurting people or damage property (be it my own or others).

Re:Autonomous vehicles

[dmitriy]

This is second successful trial of autonomous vehicles in traffic we've heard about (the first is Google). It's clear that technology is viable. It's clear what benefits it will give.

The arguments against this technology remind me the early days of the Web: it was widely believed by skeptics that no one will ever use the Web for commerce because (1) there's no framework for financial responsibility, (2) internet backbone is closed for commercial traffic, and (3) there is no accepted technical solution for HTTP encryption, and encryption packages are considered controlled technology by US government and are prohibited to be freely distributed. All these objections were solved or waived in a couple of years.

Re:Autonomous vehicles

[Anonymous Coward]

"If you had a system of tubes under the ground,
[...]
Basically, model it on the Internet."

It's a series of tubes, you know?

Re:Autonomous vehicles

[mcgrew]

The problem with autonomous vehicles is not what they can do successfully, it's what happens when they fail.

That's no different than when anything else on your car fails. If your brakes fail, you'll be damned lucky not to hit something or someone. A left front tire blowing out almost killed me in 1976. A year later, a friend did die when the motor mount on his hopped up 396 Camaro broke and pulled the throttle wide open; he hit the 17th car of a freight train at 96 mph according to the accident investigator.

If your steering fails at 65 mph only a miracle will save you.

I once narrowly missed having a tire that had come off of another vehicke crash through my windshield.

Autonomous driving *is* possible and quite easy - but we need autonomous roads to make it work

TFA proves you wrong.

Why is there no call for an automated rail network?

There are automated rail networks.

Re:Autonomous vehicles

[wootcat]

Think small steps. The information being gathered by these researchers and others may eventually lead us to extremely reliable self-driving cars. But first, I believe you're going to see that information being used to develop crash-avoidance systems and technology that can handle highly random situations. Put that technology in vehicles and soon you could see an amazing drop in vehicular accidents. Another interim step may be sections of freeway managed by smart systems to control/maintain traffic flow during high-volume time periods in a more controlled environment (limited entrance/exit points, no traffic lights or cross traffic). Traffic flow systems coupled with crash-avoidance could mean cars traveling at or near speed limits with a foot or so space between them. Combine that with navigation systems used to move cars or virtually connected fleets of cars heading to the same exit ramp and you now have a system able to move them into optimal lanes. I think we'll eventually see self-driving cars, but there are still a lot of steps in between.

Re:Autonomous vehicles

[mcgrew]

As in every other situation - the owner of the car is responsible "by default" for any accident.

Mod parent "informative". If my brakes fail and I T-bone you in an intersection, it'll be my insurance company that pays you, not my car manufacturer's insurance company.

Re:Autonomous vehicles

[Fat Cow]

It's the migration that's the problem. We already have the roads whereas building rail to every house runs into the chicken-egg problem.

Also, if you end up building rail to every house, the trains or pods will still have to deal with people crossing in front of them

Why all-electric vehicles aren't there yet

[Sonny Yatsen]

This last line caught my eye.

The vehicles ran at maximum speeds of 60 kilometres per hour and had to be recharged for eight hours after every two to three hours of driving.

I think Marco Polo probably made better time with camels. Still an impressive feat, though.

Re:Why all-electric vehicles aren't there yet

[chemicaldave]

I think Marco Polo probably made better time with camels. Still an impressive feat, though.

But they weren't audromedous camels.

Re:Why all-electric vehicles aren't there yet

[cindyann]

answers.com says it took Polo four years to get to China -- even with getting stuck in Moscow traffic the vans win.

Re:Why all-electric vehicles aren't there yet

[Anonymous Coward]

Great generalisation. Maybe you should have scrolled down a few inches on the same slashdot homepage before typing that. http://tech.slashdot.org/story/10/10/28/2258252/Electric-Car-Goes-375-Miles-On-One-6-Minute-Charge

Re:Why all-electric vehicles aren't there yet

[Sonny Yatsen]

That was a joke - not to be taken literally.

Re:Why all-electric vehicles aren't there yet

[geekoid]

Actually, no.

Re:Automated fill-ups, too?

[IAmGarethAdams]

No, because as the first 4 words of the summary states, they are electric vans

Re:Automated fill-ups, too?

[cindyann]

Didn't read the article eh? Or even the summary?

Electric vans.

really...

[pablo_max]

Packed with navigation gear and didn't get lost you say? Wow! Seems like a very achievable goal to not get lost when you are packed with navigational aids. I suppose what is more interesting is if they make the journey without human intervention..i.e..not needed a human to get them unstuck.

Re:really...

[RivenAleem]

Does this mean there was always a woman on standby to ask for directions?

THREE words

[ctid]

You couldn't be bothered to read three words before commenting?

Re:THREE words

[MyLongNickName]

I know... isn't it brilliant? First we stopped RingTFA. Then we skimmed the summaries only. Then we only read the headline. The next logical step is random posting to stories. Then, just picking a random website to post to.

I LOVE PROGRESS!!!!

Re:Automated fill-ups, too?

[2gravey]

had to be recharged for eight hours after every two to three hours of driving.

Man that's a long freakin road trip!

Great fuel milage

[Clipless]

They only had to stop for 8 hours after every 2 to 3 hours of driving. That sounds like a freaking blast.

Re:Great fuel milage

[MyLongNickName]

Sound like when I drove the inlaws to the Outer Banks for vacation. I swear to God that their bladders held only 2 mls..... COMBINED.

I wish they followed someone else's path..

[JargonScott]

Every time I see the name Marco Polo I'm instantly 12 again, screaming MARCO!!!! while at the city pool. All my "friends" left me and went to the snack bar.

Re:I wish they followed someone else's path..

[paimin]

Cartman?

We all know what happens next...

[Stenchwarrior]

Johnny Cab!!

holy crap

[grumpyman]

That's miles better than when they did the DARPA challenge???

Robo-Van is a must-have

[digitaldc]

I need one of these vans for my morning commute. I can sleep in the back while the van takes all the strees from the bad drivers, horrific traffic, scared deer and occasional road-rage :)

So what about water?

[hesaigo999ca]

So how did they cross the water, did they know which boat to take, and when it would be waiting, or was it that part that was manual for the experiment?

skynet approves...

[Anonymous Coward]

..enough said.

Proud

[xnpu]

Proud of this accomplishment! Mostly though that they arrived without any flat tires!

russia...

[Loki_666]

Bet this messed with he heads of the (Road police). How did the vans react when they were pulled over every 50km by a cop seeking a bribe....

Did the vans react and stop? How can they identify a policeman directing them to stop?

And more to the point, can they dispense cash for bribes to continue on their way?

Re:A human still has to program the sucker.

[DocSavage64109]

From my experience most accidents are due to distracted, impaired, unskilled drivers or skilled drivers exceeding safe speeds for the conditions. A well programmed AI would take out the driver skill variable and should make the car's safety equal to all but the best drivers -- much like a computerized chess AI is better than all but the best humans.

Only 13,000K from italy to china

[Nursie]

I drove 27000 around the coast of australia this year. maybe instead I should have driven from the UK to Perth.

The article almost blew my mind

[IndustrialComplex]

They had been equipped with four solar-powered laser scanners and seven video cameras that work together to detect and avoid obstacles."

I read that as:

They had been equipped with four solar-powered laser cannons and seven video cameras that work together to detect and avoid obstacles."

Those vans were almost a hell of a lot more awesome.

If this van's a rockin'..

[gearsmithy]

call tech support.

Mapless?

[Anonymous Coward]

packed with navigation gear

from the article:

the vans were driverless and mapless

How does mapless navigation work? Were they programmed to just go east all the time?

Silk Road

[Anonymous Coward]

I hope they stopped for yum cha.