Google Cars Drive Themselves, In Traffic
An anonymous reader noted that "At the TED 2011 conference this week, Google has been giving extremely rare demos of its self-driving cars. TED attendees have even been allowed to travel inside them, on a closed course. The car is a project of Google, which has been working in secret but in plain view on vehicles that can drive themselves, using artificial-intelligence software that can sense anything near the car and mimic the decisions made by a human driver."
It's pretty amazing how they've stretched the limits of technology!
Classic fear mongering. The car always had a driver in it (with override capabilities) while on public roads.
Have you seen the lunatics out there? Give me a robot any day! We are given a licence (one test only) in our youth and then out you go, rain, hail or shine, fit or unfit, tired or not, drunk or senile or both. That's ignoring the meatheads who want to deliberately drive dangerously and those not paying attention on a mobile phone texting "RORL" (roll off road laughing). I see your point but lets move on.
I can't drive due to my disabilities. This would be useful. Of course, it has to be bug free (OK almost). It probably won't be ready until after I am dead though. I always wanted KITT type of car! :(
Ant(Dude) @ Quality Foraged Links (AQFL.net) & The Ant Farm (antfarm.ma.cx / antfarm.home.dhs.org).
The 2nd order effects from this are going to be interesting. If you only have robot drivers (and you will, cause with lower accident rates, you'll have lower insurance rates if you always let the computer drive), you won't need visible signs or traffic lights. How would this affect pedestrian crossings? Would pedestrians feel irrationally unsafe crossing a road with robot drivers on it? Will we remove speed limits as computer reaction and cognitive ability gets faster?
So how do humans do it?
They start with a small number of basic "rules" and acquire the majority of their learning from experience.
In general it seems like an expert systems AI project. You have a domain that has an incomplete definition, many variables and many inputs. You hand code some basic rules as a starting point. You rig up sensors so that a computer can observe the environment and the human and it generates new rules based on its observations(*). And/Or you let the computer loose in a simulated environment and it learns through trial and error.
(*) In the movie Starman the alien learns to drive via observation. Red means stop, green means go, and yellow means go faster. Selection of the expert to observe is a critical step.
Perhaps not, but it's likely to be a hell of a lot better at not doing the idiotic things that cause the overwhelming majority of accidents in the first place.
Almost all accidents other than collisions with animals that run out in front of you are due to human stupidity. Black ice may be an exception, maybe, except that if the conditions allow it to happen a prudent driver accounts for the possibility (note that if you hit a patch of black ice the accident is considered your fault esp. for purposes of determining liability). Everything from:
You name it. It's plain old human stupidity. It's a particularly egregious kind of imbecility too, the kind that fails to recognize that other people exist and can be harmed by your poor decision-making. If "robots" can be programmed not to do these things I'm all for it. Alternatively, if robots can be programmed to beat the living shit out of people who do these things, I'm all for that too.
That's a new one to me. I have heard complaints that many train systems would be uneconomical, in the sense that they'd never survive without some kind of subsidy. I haven't heard anyone actually refer to alternate transportation as a tenet of Communism, however.
It is a miracle that curiosity survives formal education. - Einstein
Absolutely false. The car is completely capable of detecting pedestrians, deer, stopped cars, etc. This thing knows how to stop in the event that some shit goes down (see link below). You're just making up a lot of bullshit based on literally no research.
SOURCE: http://www.youtube.com/watch?v=Atmk07Otu9U Skip to 2:10 to see where the ABC reporter makes a move like she's going to run out in front of the car. The thing slams on the brakes.
-1 disagree is not a modifier for a reason. -1 troll, flaimbait, redundant, overrated are NOT acceptable substitutes.
Thirty thousand people dead each year in US car accidents. That's over half a million dead each generation. Robots could not do worse. And I think they could do a lot better. Especially if the cars talk to each other.
In fifty years people may well look back upon our manual driving culture as next to insane. That said. I love to drive. But really. It's hell out there
"No fear. No envy. No meanness." Liam Clancy
That's convenient except for the fact that there are lots of legal reasons to exceed the speed limit and it is not for you to decide if other drivers are being prudent. That is why we have police patrolling streets. There is no legal reason for you to be in the left lane as you stated yourself, it is for passing. If you are not passing then you should not be in said lane performing a rolling road block.
While yes, poor impulse control is a huge issue it is often caused by people that shouldn't be on the road to begin with. When someone is so scared to drive that they can't maintain speed then they shouldn't be on the road at all. They force people to pass them and clog up road ways when they fail to merge properly. Every time I see someone stop and wait for an opening I know I get a little more mad.
This is just funded by Google. It's the group at Stanford which did the DARPA Urban Challenge that's doing the work. It's essentially the same technology. They're getting very good at this.
The thing on top of the car is a rotating cone of LIDAR scanners. The original version of that was developed by Team DAD for the 2005 DARPA Grand Challenge. The prototype, which was a much bigger wheel of scanners, fell off the vehicle. But they then built a more compact production version, the Velodyne scanner, with 64 lasers. It costs about $100K per unit, but automatic driving became much better once that came out. Most of the teams in the DARPA Urban Challenge used that.
Personally, I think the rotating machinery approach is too expensive for production, and that the Advanced Scientific Concepts flash LIDAR has more promise as a production product. The ASC system requires some exotic custom imaging ICs, with a time-of-flight timer behind each pixel. That's the kind of thing that's incredibly expensive when you make 10 of them, and cheap when you make 10 million.
Taxi: Welcome to the free automatic Google Taxi. Where do you want to go? ...
Passenger: To the train station.
Taxi: OK. By the way, there's a Starbucks on the way. They currently have a special offer, two coffees for the price of one. Maybe you want to go to there first?
Passenger: No, I just want to go to the train station.
Taxi: If you are interested in train stations, maybe the railway museum would interest you. It's only five minutes from here.
Passenger: I'm not interested in the train station, I just want to get my train.
Taxi: Maybe you are interested in Morton's model railway shop? They have great models, and I can get you there in only ten minutes.
Passenger: I don't want a model train, I want to use the real train!
Taxi: Did you know that just this week, the Railway Academy opened? In the first year they give discount for their locomotive driver courses. I can send you the application form to your phone.
Passenger: I don't want to drive the train, I just want to take the train. And if you don't drive me there soon, I'll miss it.
Taxi: Did you know that you can buy train tickets with 5% discount at train-ticket.com?
Passenger: I already have the train ticket. I just want to get to the fucking train.
Taxi: Oh, you are interested in fucking? There's a whorehouse not far from here
The Tao of math: The numbers you can count are not the real numbers.
Well, the good news is that you'll probably have all the telemetry data from the sensors, including the visible spectrum imaging system. This should give you enough evidence to prove that the car did the best possible thing... or to prove that it failed and that it's the fault of the company that sold you the car.* Once this happens once or twice every manufacturer will be required to provide a system that can take the sensor data and generate a 3d simulation of the situation for use in court. Hopefully that will help stop nonsense lawsuits.
*(Not that I agree with holding the manufacturer liable for a pedestrian jumping out in front of a car. I just think it's better that the family frivolously sue a car company with plenty of lawyers than an average citizen with no money for extended legal shenanigans.)
-1 disagree is not a modifier for a reason. -1 troll, flaimbait, redundant, overrated are NOT acceptable substitutes.