Now where's my flying car?

March 4, 2011 7:38 AM

Inside Google's Auto Driving Car
posted by empath (33 comments total)

There's something missing.
posted by zarq at 7:49 AM on March 4, 2011

I think this could have used more background info. Here's the source of the video, Google's blog entry about the project, and a New York Times article.
posted by exhilaration at 7:50 AM on March 4, 2011 [2 favorites]

Eh, I'd be much more interested in a video from when they have it in real road situations. A closed course with no disturbances is pretty much where an auto-driver is expected to excel - similar to flying an aircraft. The only uncertainty is the traction of the pavement.
posted by anthill at 7:52 AM on March 4, 2011 [1 favorite]

We've already been inside Google's car for awhile now.
posted by Burhanistan at 7:52 AM on March 4, 2011

Eh, I'd be much more interested in a video from when they have it in real road situations.

I think their TED 2011 talk showed examples of that, but I don't know if its online yet.
posted by memebake at 7:55 AM on March 4, 2011

I like that "auto driving car" can be interpreted in two different ways, both of them correct.
posted by mingo_clambake at 8:00 AM on March 4, 2011 [5 favorites]

In the future computers will burn with white hot fury at the selfish, aggressive driving of other computers. Droid rage, or something like that. Drone rage. Dammit.
posted by tigrefacile at 8:11 AM on March 4, 2011

I'm extremely skeptical this will go anywhere (SWIDT?). Computers are good at limited domains. Driving is not a limited domain. Worse, driving is like 90% a limited domain, 9% "routine emergencies" and 1% extreme excitement. So if you have a car that handles the drudgery part, the driver's attention will wander and s/he won't be there when needed: pow!
posted by DU at 8:13 AM on March 4, 2011

Eh, I'd be much more interested in a video from when they have it in real road situations.

Here's a short video from the nyt with the car driving in Mountain View, California.

Also, from the Google Blog:

With someone behind the wheel to take control if something goes awry and a technician in the passenger seat to monitor the navigation system, seven test cars have driven 1,000 miles without human intervention and more than 140,000 miles with only occasional human control. One even drove itself down Lombard Street in San Francisco, one of the steepest and curviest streets in the nation. The only accident, engineers said, was when one Google car was rear-ended while stopped at a traffic light.

posted by memebake at 8:17 AM on March 4, 2011

I have a theory: Metafilter posts in the morning get "meh, that's no big deal" reactions. Evenings get enthusiasm.

This video is fun because we get to see what it's like inside the car, with some random outsider experiencing the joy of a car that drives itself very fast. Yeah, it's a closed course, but it's also a pretty challenging one at that speed. The Google car has also demonstrated itself in real traffic, that's a much bigger challenge this team has met. It's pretty amazing.

Self-driving cars seem scary, but self-flying airplanes are common. Automated flying is easier than driving since there's no collision avoidance and very few turns. But it's given us a lot of experience in how to make sure a human can take over. The pilot sits there constantly monitoring what the autopilot is doing and can take over at any moment to hand fly it. But most of the time, the plane flies itself.
posted by Nelson at 8:20 AM on March 4, 2011 [1 favorite]

Miles? Pff. I want to see stop-and-go traffic. One way signs. Construction zones (both for special speed limits and obeying human instructions). Moving over to let the car in the lane next to you move over to let someone merge onto the highway. Etc.
posted by DU at 8:21 AM on March 4, 2011

I'm extremely skeptical this will go anywhere (SWIDT?). Computers are good at limited domains. Driving is not a limited domain. Worse, driving is like 90% a limited domain, 9% "routine emergencies" and 1% extreme excitement. So if you have a car that handles the drudgery part, the driver's attention will wander and s/he won't be there when needed: pow!

Humans don't like to admit this, but we're absolutely shit at driving, just in terms of reaction time. Similar to the Watson thing, this is an example of those problems that people will say can never be done by a computer until it's actually done, then people forget all about how insurmountable it was.
posted by odinsdream at 8:27 AM on March 4, 2011 [4 favorites]

Of course it will eventually be done by a computer. But that computer will actually be intelligent (which is to say, good at unlimited or ill-defined domains).
posted by DU at 8:33 AM on March 4, 2011

I can't wait till my hands can be free to do more important things like mobile coffee brewing or eye surgery.
posted by mateja at 8:37 AM on March 4, 2011 [1 favorite]

I can't remember where I read this, but.. the real excitement for self-driving cars is they could enable the end of private car ownership. A fleet of self-driving taxis that will arrive at your door in < 2 minutes and take you wherever you want to go could be a really big change.
posted by Nelson at 8:42 AM on March 4, 2011 [5 favorites]

"Droid rage, or something like that. Drone rage. Dammit."

'roid range.
Works best if said in an insultingly bad Kiwi accent
posted by sodium lights the horizon at 8:48 AM on March 4, 2011

A fleet of self-driving taxis that will arrive at your door in < 2 minutes and take you wherever you want to go could be a really big change.

That sounds like heaven to me. Double heaven if the cars make Jetsons noises. (I've grown to sort of hate cars)

But that computer will actually be intelligent (which is to say, good at unlimited or ill-defined domains).

An "intelligent computer" is a philosophical problem, not an engineering one.

I don't think it would be insurmountable to create an algorithm which recognized the common street signs (they are highly reflective, well typeset, and are systematic geometric shapes), OCR the text and have a reasonable set of programmed responses. If my email client can learn to identify junk mail over time, then a driving system can learn to identify obstacles, construction hazards and what not. That "knowledge" could be easily shared between cars.

I also don't think that recognizing a human shaped blob waving flags around is an insurmountable challenge either.
posted by device55 at 8:49 AM on March 4, 2011

Well, balls... "roid rage" not "roid range". I mean, really. I've just spent the last 2 hours picking interviewees apart on attention to detail and then that... ugh. Thank god it's Friday and I'm on this side of the interview...
posted by sodium lights the horizon at 8:50 AM on March 4, 2011

Some people here are really jaded, but they'll turn around when Google's car driving auto comes out.
posted by grouse at 8:51 AM on March 4, 2011

The pilot sits there constantly monitoring what the autopilot is doing and can take over at any moment to hand fly it.

This is not correct. The notion of 'the human backup is always there, engaged with the current situation, with practiced skills, and ready to intervene at a moments notice' is false. Humans don't pay attention unnecessarily. There will be several (tens) of seconds of delay before even an experienced driver will be able to effectively take over. In an airplane, this is OK (until something happens autopilot doesn't know how to handle and you stall). On the ground, this amount of time might not be enough.

Lisanne Bainbridge outlined the "ironies of automation" nearly 30 years ago, and most of them are still valid to badly designed automation today.

But Google designers are smart, and have probably read the 30 years of research in designing automation and automation interfaces in aviation, military, and process control. Looking forward to seeing how this proceeds. There is a small fortune waiting for them in open-pit mining if they can get this to work.
posted by anthill at 9:11 AM on March 4, 2011 [2 favorites]

I don't think it would be insurmountable to create an algorithm which recognized the common street signs ... or recognizing a human shaped blob waving flags around ...

It's never an insurmountable challenge to create an algorithm that can deal with situations you, as designer, anticipate. The problem Google designers are dealing with is that the closer you get to an evolved, natural world (like London downtown streets), the more variation and strangeness you get. I look forward to auto-driving on interstate freeways. I just hope every street doesn't have to get turned into one.
posted by anthill at 9:16 AM on March 4, 2011 [2 favorites]

Autonomous City Explorer
posted by zennie at 9:21 AM on March 4, 2011

It's never an insurmountable challenge to create an algorithm that can deal with situations you, as designer, anticipate.

this x 1e9
posted by DU at 9:49 AM on March 4, 2011

Miles? Pff. I want to see stop-and-go traffic. One way signs. Construction zones (both for special speed limits and obeying human instructions). Moving over to let the car in the lane next to you move over to let someone merge onto the highway. Etc.

One of the Android guys has a reply he likes to use every time someone complains about Android. You want to see these videos? http://www.google.com/jobs. Because I've seen those videos.
posted by GuyZero at 10:46 AM on March 4, 2011 [1 favorite]

I was having a conversation today about how I was promised twenty years ago that there would be self driving cars by now. It was with one of the blind pupils at my school who was telling me how she's determined to move out of her tiny village when she's older because it's so hard to get around without a car (and way too expensive to use a taxi everywhere). It would be amazing if they could create a model with enough fail safes that she could hop in and go where she needs to at some point in the future.
posted by nangua at 11:27 AM on March 4, 2011

Isn't the reason we don't have self driving cars, aside from the technology being relatively young, that manufacturers worried about liabilities?
posted by mccarty.tim at 11:36 AM on March 4, 2011

Liability is an issue, but surely not an insurmountable one. We already have smart cruise control on cars, the trick is to slowly expand the scope of it to include things like basic steering on highways, situational awareness of other cars (either visually or via wireless).

Eventually these will expand to the point that folks in the driver seat are pretty superfluous, and then it will be much easier to find the political will to adopt full self-driving vehicles

The alternative is to permit full self-drivers, but only initially for restricted domains. I'm thinking buses in special lanes or airport shuttles. Here too, eventually people's comfort with self-drivers will allow the eventual expansion into wider domains.

Likely both of the above will be happening simultaneously.

As stated before, humans suck at driving. I would bet that the trade-off in decreased rate of accidents via self-driving cars would more than offset the rare 'mad car' accident liability. You'd just need someone with money and willpower to create insurance pools of self-driving cars.

Here's looking at you Google. Motor vehicle accidents are one of the top 10 causes of death in this country, and the #1 accidental cause of death. You've got the opportunity to save over 40,000 lives per year. Go do some good.
posted by leotrotsky at 12:06 PM on March 4, 2011

I'm extremely skeptical this will go anywhere (SWIDT?). Computers are good at limited domains. Driving is not a limited domain. Worse, driving is like 90% a limited domain, 9% "routine emergencies" and 1% extreme excitement.

Well, you clearly haven't been keeping up on this stuff. Google's self-driving car has been driving on real roads in real traffic for months. Human beings also respond poorly to unexpected events and in fact tens of thousands of people die every year in auto accidents.

Frankly this AI skepticism is just getting old. How much cool stuff do computers have to be able to do before people just admit they were wrong about computers not being able to solve complex problems the way humans can?

Isn't the reason we don't have self driving cars, aside from the technology being relatively young, that manufacturers worried about liabilities?

That's an interesting point. Right now, if you get into a crash, you're at fault. But what if the AI goes haywire? Who would be legally responsible? I suppose car companies won't sell these unless the driver agrees to accept liability for the cars mistakes (which should be less common then actual human driver's mistakes). I think autodrive cars will greatly reduce the carnage on the roads.

The great thing about this, IMO is that we'll be able to get home after the bar, and we'll be able to use the internet while driving. Long drives will be so much more convenient.
posted by delmoi at 1:05 PM on March 4, 2011 [1 favorite]

Isn't the reason we don't have self driving cars, aside from the technology being relatively young, that manufacturers worried about liabilities?

One of the reasons there are so few YouTube competitors isn't just that it takes a lot of storage. It's that it's a legal minefield. And crawling potentially copyrighted content on the web in multiple legal jurisdictions without the authors' permission to create an index, which may be viewed as a derived work?

Google isn't just a technological achievement - it's a regulatory achievement.

So consider again why Google might manage to make self-driving cars work when other companies can't.
posted by GuyZero at 1:53 PM on March 4, 2011 [2 favorites]

How much cool stuff do computers have to be able to do before people just admit they were wrong about computers not being able to solve complex problems the way humans can?

Well they don't do it he way humans can, which is why there is so much skepticism. People don't understand how it's possible to do these things without human consciousness. But most of what people do is already done without consciousness.
posted by empath at 1:56 PM on March 4, 2011

Most of driving is definitely done without 'conscious' consciousness. We're pretty much a crappy, limited manual controller coupled to remarkably effective, adaptive visual and recognition senses.

How much cool stuff do computers have to be able to do before people just admit they were wrong about computers not being able to solve complex problems the way humans can

The challenge here (as well as many other places) is less on complex problem-solving, and more in how well computers can be taught to formulate complex problems.
posted by anthill at 2:17 PM on March 4, 2011

Isn't the reason we don't have self driving cars, aside from the technology being relatively young, that manufacturers worried about liabilities?

Also, I recall reading that the array of lasers that the car uses to "see" was quite expensive ($50'000 or something)... though the auto industry

As for liability, I'd assume you'd see the first self-driving cars being used commercially in some country other than the U.S.
posted by bobo123 at 7:44 PM on March 4, 2011

bobo123: Yep, from the exterior video it looks like they're using the standard-issue Velodyne HDL-64E LIDAR. Last I heard it was going for about $75,000. IIRC, the company recently introduced a $30,000 model with half as many lasers.

Right now the state of the art in computer vision simply can't provide the level of reliability and accuracy that we get from LIDAR. As soon as somebody figures out how to replace that fragile $30k laser assembly with something like a $2k array of cameras, autonomous cars will be much closer to being a real consumer product.
posted by teraflop at 9:01 PM on March 5, 2011