Google: What We're Driving At | Hacker News Reader

Google: What We're Driving At

googleblog.blogspot.com

259 pointspiinbinary84 comments

One of the key issues for Google, from a corporate strategy standpoint, is "freeing up people's time." Driving is one of last places where we spend significant time awake without being able to use the Internet and hence any of Google's services (except if you use a smartphone, which is now illegal in some states, and in any event isn't an ideal place to be clicking on ads). There is a safety issue here as well. Although using a smartphone while driving is illegal in some states, people are driving while using their smartphones with increasing frequency. We need our Internet "fix."

I am sure this driving technology also taps into several of Google's key capabilities: e.g., programming expertise, its voice recognition technologies, search, its mapping software (Google Maps and Navigation), etc.

I sadly feel we'll never have cars that drive themselves.

The first time, ever, one crashes, regardless of the circumstances: "COMPUTER DRIVEN CAR DEATH! DO COMPUTER CARS MAKE YOU UNSAFE?" is shouted from the media rooftops, citizens get outraged, laws are passed, and we're all doomed to sit in traffic and continue to lose many lives to manual driving forever.

This hasn't happened for other transportation methods so I don't see why cars should be different.

Early planes crashed more often than not and even today we get at least one huge crash per year, highly covered by media but no one suggests that we should ban planes even after they were used on 9/11 to destroy twin towers.

Busses crash, trains get derailed etc.

I think you're painting humanity with overly pessimistic brush as if reacting in the worst possible way is the only way can possibly react.

A vision for a safer, more efficient transportation is very compelling and something that many people and plenty of media will get behind and advocate.

I think this can only be addressed by stating it clearly and upfront. "Automated cars will kill people. They will kill people every year. The only thing they have going for them is that every year they will kill 10 times less people than people driven cars. Life is dangerous, and it's your choice."

Car makers initially fought putting seatbelts in cars because it made them look dangerous. Now seat belts fracture thousands of ribs a year and are required by law. Both initial fear and actual danger were over come due to the massive demonstrable benefit of the technology.

I can see that too, but I don't think it's impossible. I don't think it will be an instant transition from driver to computer, it will be more gradual. More and more things in your car will become automated, starting with the simplest until we've reached the point that the driver is there simply to be transported.

For comparison, I think about other methods of transportation that have already been automated. Autopilot on planes, fully automated airport trams, etc. To be sure, these are dangerous systems if something goes wrong, but the infrequency of catastrophic occurrences let's us mostly trust in their safety.

> I can see that too, but I don't think it's impossible.

Honest question, how do you automate this: http://www.youtube.com/watch?v=85dzsA-qKPw&feature=relat... ?

Easy, you apply a bit of game theory. A system similar to say a poker bot can make the high level decisions.

Not really. Computer cars will very likely be manufactured mainly by the existing auto giants. And those have billions of ad dollars, so the media will be very careful.

In the USA this is true but European countries have already given the green light to a somewhat watered-down version of self-driving cars on public roads. The USA will eventually follow I would think.

Something no one seems to have mentioned: the post is by Sebastian Thrun, a professor at Stanford. His team won the DARPA Grand Challenge in 2005 and placed 2nd in the DARPA Urban Challenge in 2007, so he's one of the most experienced people in the world in the field of autonomous driving. He's also done important work in robotics and mapping, two of the key skills needed for autonomous driving.

I didn't realize he was also at Google now, but this is probably good news for everyone, since Google will likely support him better than DARPA could.

Links:

http://en.wikipedia.org/wiki/Sebastian_Thrun

http://en.wikipedia.org/wiki/DARPA_Grand_Challenge

Awww.

I liked the idea of waiting to April 1 to announce this, complete with videos of the cars driving.

(Disclaimer, I work for Google and have known about this project for some time. I have no connection to it.)

This looks like something that could be game-changing. But Google seems to dabble with ideas that don't ultimately go anywhere. Do you have a sense for how seriously they're taking this project?

Is "as seriously as possible" a satisfactory answer? This stuff isn't easy ;)

This would give a whole new level of meaning to Google's favorite product subheading... "Beta"

April 1 trick used during the launch of Gmail. Twice wouldn't be that great. (It was a great marketing idea, by the way.)

I think it would make April 1 much more fun for people to be always sorting through Google's jokes trying to figure out which one was for real.

According to the World Health Organization, more than 1.2 million lives are lost every year in road traffic accidents.

That is an incredible number. 1.2 million out of 7 billion. If the average person lives to age 70, that means that on average, one has a 1.2% chance of dying from a traffic accident.

Not to mention most of those 7 billion likely don't drive, which effectively increases your risk if you DO drive.

Having been in 3 major car accidents, one for each decade I've been alive, where I was at the wheel only once and alcohol was not involved in any of them... I have to say I do not find these numbers at all surprising. I would have guessed higher.

Another way to look at it is you've been in 3 major accidents and survived all of them, so it may be that the chances of having a fatal accident may be lower than you thought?

Unlikely. You're probably under the belief that That Can't Really Happen To You, just like almost everyone. At least experiencing the high risk of it a few times probably will help you to consider seriously the height of the risk.

That said, being human you might overcompensate.

And that doesn't count the incredible number of serious-but-not-fatal accidents which surely cause much injury.

I'm not a statistician, but something about that doesn't sound right to me.

  >>> p = 1.2e6/7e9 # Probability of dying in any one year
  >>> scipy.exp(log(1-p)*70) # Probability of NOT dying 70 years in a row
  0.98807069644395429
  >>> 1-_
  0.011929303556045712

There might be some roundoff error in the exponentiation. I am too lazy to check.

Out of interest, why do you prefer the second line to

    (1-p)**70

? Is there some kind of improved numerical stability?

Great, thanks for the clarification.

I thought it sounded high too, but it's correct.

Does that number count bus accidents?

It's interesting to me that the cars are legal under California law. It means they could theoretically partner with car makers to sell it as an expensive option -- advanced cruise control. Much less scary sounding than automated vehicles. Once you get enough early adopters on the road it seems like the public opinion would likely shift pretty quickly -- is there anything people resent more than driving?

> ...is there anything people resent more than driving?

I for one absolutely love driving; it's the traffic-filled commutes to and from the office every day that I resent. If I could optionally automate the driving for times when I don't want to be driving, ie, in highway traffic jams, on long road trips, or when I'm tired, then I would gladly pay extra for the option.

This is exactly how I feel, and I think it plays right into the "Advanced Cruise Control" perspective. A mode that one might use on highways (whether on a long trip or stuck in stop-and-go traffic), where there are much lower risks of unexpected pedestrians/bicyclists/blind driveways/etc (and thus lower liability risks).

Also, I have to add that this is the kind of research that I love that Google is doing. I might not love everything they do, but I don't think there are more than a handful of corporations actually pushing things forward with far-out, jump-into-the-deep-end projects the way they are.

Surely "when I'm tired" is one of the situations in which this tech should not be used. In Australia, for example, where long-distance driving is common, fatigue is one of the biggest contributors to road accident statistics.

Apart from the risk of someone actually falling asleep while the car is on auto-pilot, anything that reduces one's alertness over the duration of a drive makes this technology more dangerous (which is not to categorise it as dangerous).

The road recognition features would deal with simple cases like drifting out of lane but not so much in built-up areas or around unpredictable drivers. I think a big part of safe driving is adapting to accommodate the behaviour of other road users. Hopefully some tech will be developed in this area such as recognising inappropriate signalling, drifting out of lane, tailgating (ahead) etc.

I really like the "Advanced Cruise Control" branding idea--it makes the concept much more palatable to frame it that way. I hope whoever introduces this runs with that idea.

I would guess car rental companies would be a better first partner, they already put cameras in their car. A lot of the videos you see on youtube, with drivers crashing thanks to behaviour which could get them a Darwin award, are from rental cars.

Then they could partner with FedEx, UPS, or DHL.

Then car manufactures can put in optional parental override, but a better marketing name would be computer assist. Especially for old people, they wouldn't get computer override, they'd have computer assist.

And a few years after that, when the statistic are in and computer driven cars are much, much, much less likely to kill anyone then human driven cars. Even after you discount drivers under the influence. Then we'll see political pressure to have computer "assistance" in every car.

It means they could theoretically partner with car makers to sell it as an expensive option -- advanced cruise control.

The first uses will probably involve traffic between large distribution centers located out in the suburbs/exurbs. To begin with, the trucks could park themselves at a staging area out on the tarmac, and a human driver could go out and bring the truck to the loading dock. The USPS, UPS, and Fedex could be the first big customers along with Walmart.

It's worth noting that they are already starting to do this in very small ways. Many high end cars will now automatically apply the brakes if you get too close to the car in front of you on the highway, for example.

The idea of "highway trains" is promising. Cars are still inefficient - only 1% of the energy they consume is actually used to propel the driver.

It's hard to get too excited about anything involving cars.

Car industry haven't made enough progress in the last 100 years, but your 1% is way off. A modern car diesel engine (like the one in BMW 320d ED) can exceed 45% efficiency.

Car weight = ~4000

Average person = ~200

That's not a problem with the efficiency of the engines or other parts. That's a social issue of people who drive with no passengers.

Once you've overcome the social issue of each person wanting their very own vehicle, you don't need cars at all. You can use buses and trains.

Except where you want to go where buses and train don't go...

Funny, no one ever complains about the times you want to go where highways don't go.

Seriously, though, countries with functioning infrastructure have trains that go nearly everywhere, or at least close enough you can walk or bicycle the rest of the way.

I'm in the US. The closest bus stop to the college I work at is > 90 miles away and the closest train stop is 13 miles away but is unstaffed and not for local transit. Roads are the base transportation infrastructure in the US, so using that existing infrastructure is a good bet.

I'm in the US too, and it sucks. That's why I said "countries with functioning infrastructure" (because with anything bigger than a small town, driving around isn't functioning infrastructure). But you're not going to have any kind of sustainable infrastructure when it takes a 1-2 ton vehicle to move a 100-300 pound person any distance, whether or not the vehicle is automated.

We are going to have to come up with something "sustainable" with only 1 or 2 people in the vehicle, because the US still has a lot of rural areas.

Rural areas make up very little of the population. Although there's some improvement there--I live in a rural town that's quite navigable by bicycle and has a good bus system.

The real problem is the suburbs.

Uhm, there are no buses here (see other posts) and bicycles don't work in certain conditions (high winds, snow, -40F below weather). Plus, let's also remember that a lot of places don't have facilities for people to bike into work (like showers), not to mention the people who can't use bikes.

So what? We live in the sticks. There aren't enough of us to matter. If we keep driving around with one SUV per person it doesn't add up to much, comparatively. The low hanging fruit is fixing the metropolitan areas. Most people live there, and demographically, people are still moving there. So fucking around with making cars and rural areas more efficient isn't going to make much difference compared to making metropolitan areas more efficient and reducing car usage within them.

That said, rural college towns can have pretty awesome bus systems that run all year, even during subzero temperatures, and they can also have bike paths that go mostly everywhere, which are usable for nine months of the year. Because mine has all that. I'm sorry yours doesn't; but obviously, there is room for improvement, which is exactly my point.

>Seriously, though, countries with functioning >infrastructure have trains that go nearly everywhere, >or at least close enough you can walk or bicycle >the rest of the way.

Oh you mean those countries without poverty, unemployment, corruption that built pollution free cities where everybody is happy?

Seriously.

First not everybody can or wants to use a bicycle.

Additionally you can't possibly hope to give access train to every part of the "country", that would be 1/extremely expensive 2/extremely uncomfortable for people living close to lines.

You need a density threshold. You can't have a bus line for two people once a week. You can see the density threshold problem in action in the USA.

Lastly comes the problem of service time. It's impossible to offer 24/24 7/7 service except for the most busy lines.

But perhaps you're willing to put 50% of the GDP into public transportation?

I live in a city with outstanding public transportation (Paris). Metro, buses, mini-buses, taxis, bicycle rental, you name it.

Yet, if you have several appointments within the same day in the city, it's not doable by public transportation because of the waiting and walking time that adds up extremely quickly.

If you go out to see friends that live outside the city, it's not doable to come back after midnight.

If you want to visit someone that lives outside the city, it might take a lot of time to use public transportation.

You're dreaming if you believe you can build a public transportation system that solves all transportation problems for everybody. If your system is good, you'll help people to go to and back from work and move around the city for a decent price.

But you'll always need individual transportation (In my case I ride a motorcycle. It saves me a lot of time and headaches and it's just damn cool and fun!).

Welcome to the real world, where things are difficult.

OK--you need personal vehicles a lot less. Seeing the nightmare of gridlock and freeway my country is, though, cutting out the easy 80% is something Americans have tragically left undone. (The USA has low density because of vast areas of flyover country where virtually no one lives--most Americans live in metropolitan areas.)

This explains Eric Schmidt's "casual" remarks about cars that should drive themselves recently at the TC Disrupt conference ( http://techcrunch.com/2010/09/28/schmidt-on-future/ ).

It would help a lot if every car get connected to GPS and a global network identifies cars that are in the nearby. Now the server should decide how these cars should go. And in addition, there should be an independent regulation in the car itself in case Internet is down.

If every person gets connected to GPS using his mobile phone, then they can be integrated in this complex network reducing the risks of accidents.

The server would also connect with others to check for roads constructions, weather... and decide the best and fastest path to take.

It would be really amazing and complicated. But believing is how we can get these things done.

It would also help if the map attached to the GPS was actually accurate. I seem to find a lot of secondary highways in rural states that are not located correctly. It is a true pain when your GPS starts screaming about "getting back on the road". Really a pain having to turn off the sound when you come to certain highways.

I've always imagined that there would surely be a way to optimize traffic to the extreme if it was completely automated. For instance, in many circumstances I could see that instead of a set of traffic lights, instead of slowing down, cars could communicate with each-other and adjust their speed accordingly, so that they criss cross each-other perfectly. (think Traffic Rush on iphone/android)

That sort of precision would require a ton of computing power, but Moores Law, the cloud and wireless internet are making that all possible. Exciting stuff!

Considering that San Francisco has arguably the most difficult city streets to drive in the country, I'd say this is no small feat

Why would their technology just lower the death rate instead of eliminating it almost completely?

Why the downvote, its an honest question. When all cars are automated, the accidents left should me near zero...

Bizarre road conditions, mechanical failures, the percentage of humans who insist on driving old cars or operating their smart cars manually, improbable edge cases which the software fails to account for ("race conditions"?), illegal drag racing, hackers playing pranks which turn deadly...

Probably lots of other scenarios. You'll never eliminate error completely, but I think we'd all be a lot happier if car crashes happened with the frequency of airplane crashes.

Don't forget, also, pedestrians who run out onto the road and get hit by an automated driver.

"In terms of time efficiency, the U.S. Department of Transportation estimates that people spend on average 52 minutes each working day commuting. Imagine being able to spend that time more productively."

You can, it's called taking the bus, and in the process it actually employs some people such as drivers, cleaning personnel, etc.

Not every place in the US has buses. I commute about 45 minutes each way and no public transit available.

Of course, but as the statement was said it was if there was no way of commuting without oneself being the driver.

So we are getting there, at last :)

http://www.paleofuture.com/blog/2007/5/11/disneys-magic-high...

http://www.youtube.com/watch?v=F6pUMlPBMQA

I can drive a car with no driver, no driver, no driver... http://www.youtube.com/watch?v=HLUX0y4EptA

Well it's a good song anyway, even if it isn't talking about an AI... yet.

I wonder if there is something that can be done to the actual road to make it more "friendly" to robot vehicles? Roads are constantly being worked on and it would be interesting to attack parts of the problem from that angle.

A few months ago, one of my friends who works for Google told me, "In the next year, Google is going to announce something that will change the world." Is this it?

I thought your ideal future was everyone riding bicycles ;)

http://news.ycombinator.com/item?id=1699114

I wasn't making a value judgment, just wondering out loud. I thought it was going to be a new app for my phone ;)

Robo Prius in 2008: http://news.cnet.com/8301-11386_3-10042320-76.html

Surprised I'm being down-voted. This story is real. Both cars have the same Lidar scanner and other hardware.

http://bit.ly/d8nIZp

Trace through the people in both stories and you'll see the connections.

Is there any information on the algorithms behind this?

I'd like to know how often the human drivers have had to take over.

For all the sexiness that is the idea of automated cars, there is one fatal flaw: way more people are going to be driving drunk. Yes, eventually, maybe the tech is good enough that it won't matter what state you're in, but for now there should be a human operator there for emergencies; one that is able to respond quickly and isn't intoxicated.

Fixed by a ID-system and a blowtest for alcohol.

Why not?

I'm surprised nobody has mentioned one of the biggest issues with general acceptance of this system:

Police will be able to force anyone to pull over anywhere, at any time, and for any or no reason.

Even though they can do that now, people still decide to pull over. They are not forced to.

And what if there becomes a database that police forces compile, showing the history of movements of everyone? That database could potentially be leaked. I think the privacy implications of this should get at least some thought.

I read this backward the first time, because the way you say it really doesn't make any sense.

Police need a reason to pull you over, and a car that's autonomously driven by a computer that won't allow it to drive at an illegal speed or unsafe manner is essentially immune to routine traffic stops. Provably so, because it has telemetry and video demonstrating that it's done nothing wrong.

So unless you manage to get the model with the broken tail light, these cars will actually make the police less happy because they can't pull people over randomly anymore.

When a police officer indicates you must stop your car, you don't really have a choice currently. They doubtlessly will do their best to force you to stop, and if you choose not to, you will be chased, possibly collided with, and will likely end up with a gun pointed at you. As far as what happens after you do stop without causing an incident, you do have a choice, somewhat.

I do agree that this could create an environment where authority is in absolute control of the movements of citizens, a power which could (and surely would) be abused.

"Even though they can do that now, people still decide to pull over. They are not forced to."

If you decide not to, then you are forced to. Unless you have Jason Bourne-esque driving abilities.

Reality isn't really the issue. People's perception of reality is what will determine whether this system is accepted.

Since you're recycling your material (http://news.ycombinator.com/item?id=1776383), I will too (http://news.ycombinator.com/item?id=1776429).

I don't know how many episodes of World's Wildest Police Videos you've seen, but police can already force you to pull over anywhere, at any time, and for any or no reason. Spike strips, PIT maneuvers, etc. Letting them pull you over with a remote command would be a huge improvement--high speed chases are unsafe for all involved and (due to frustration and adrenalin) increase the risk of police brutality.

Of course, with computer-controlled cars, you can eliminate traffic infractions and hence most if any need for police to pull you over anyway.

It's kind of depressing to see a blithe assumption that the police will be given that kind of power, and that if they are, nobody will hack their own cars. Granted the first assumption is plausible.

So, like gun control, will a coming political battle be for 'auto control'?

Reply on news.ycombinator.com