Aren't self-driving cars impractical?

[Nathyn]

http://money.cnn.com/2007/02/14/mag...tomated.biz2/index.htm?postversion=2007021506

The 39-year-old Stanford professor of electrical engineering is the world's most successful manufacturer of self-driving vehicles: He and his graduate students built Stanley, the first car to complete the DARPA Grand Challenge, a 131-mile robot car race across the desert near Las Vegas organized by the research arm of the Pentagon.

That historic success in 2005 netted Thrun a $2 million prize. He reinvested some of it in an even more intelligent Stanley, which will be unveiled at the next iteration of the robot car contest - one that takes place on city streets. Obstacles include stop signs, lights, and cars driven by humans.

I think it will be a long, long time before we ever see a self-driving car, if ever, for two main reasons:

#1. Detecting road signs and traffic lights. It would be incredibly expensive to outfit every street in America with radio transmitters to tell self-driving cars the proper speed limit. Not to mention this huge project would be impractical: Why should the government spend billions just so lazy people can have "self-driving cars"? Road signs couldn't be programmed directly into the car itself, like in the OnStar system, either because of how they frequently change (I.E. during construction). Plus you can't
really detect traffic lights.

#2. I don't believe we have the kind of optics to be able to always accurately distinguish between various objects on the road, certainly not good enough to accurately track movement. It was a real challenge for us to develop a UGV which could even move through a static obstacle course... That's worlds away from ever building one that could detect the color of traffic lights, distinguish between bicyclists, autos, pedestrians and animals, and be able to move accordingly.

 

[Horatius]

It will come as an evolution thing. We'll start having self-driving cars where we need them most - highways - which have the least number of complications such as you describe. It will start with inter-city highways, and then expand to in-city highways. With some of the bugs worked out, it will gradually expand onto major city streets, and then later, minor ones.

So expect a few years of a system in which you drive yourself to a highway on-ramp, and then the car takes over. Once you're close to your destination, the car returns control to you for the street-level driving.

It would make inter-city driving a lot nicer, and reduce accidents due to driver fatigue and what not, as well as eliminating a lot of stupid-mistake accidents.

At least, that's what I'm hoping for.

 

[timhau]

It will come as an evolution thing. We'll start having self-driving cars where we need them most - highways - which have the least number of complications such as you describe. It will start with inter-city highways, and then expand to in-city highways. With some of the bugs worked out, it will gradually expand onto major city streets, and then later, minor ones.

I think it'll be the endpoint of a long, gradual process where more and more functions are transferred from the driver to the car itself. And the process is happening right now.

We've had automatic transmissions for a good while. In the 1990s, anti-lock braking systems became standard. Now, electronic stability control is becoming more and more widespread; I suppose most new cars have it by now. Cruise control has been available for a couple of decades, but now we're getting cruise controls with sensors that measure your distance to the car in front of you and adjust accordingly. Some top-of-the line Lexus models can be equipped with cameras that monitor the road and send a warning if you're about to cross a line on the road without giving a turn signal.

 

[Roboramma]

As to the singage issue, I can see a similar evolution.

Why not have cars with a database that stores the location of street signs (stop sides, speed limit zones, etc) which is updated via the internet? Car owners could be required to update it every month or something.
But if we did need some sort of radio transmitter on the signs, they might first appear in accident prone areas. Cars properly equiped could then offer a warning to their drivers, rathar than this specifically being for "self-driving" cars.

The thing about cars driving themselves is that they could be designed to maximize the efficiency of traffic flow. They could also be designed to increase road safety. We may not be there yet, but when we are they will have something to offer to far more than just lazy drivers.

 

[Horatius]

We've had automatic transmissions for a good while. In the 1990s, anti-lock braking systems became standard. Now, electronic stability control is becoming more and more widespread; I suppose most new cars have it by now. Cruise control has been available for a couple of decades, but now we're getting cruise controls with sensors that measure your distance to the car in front of you and adjust accordingly. Some top-of-the line Lexus models can be equipped with cameras that monitor the road and send a warning if you're about to cross a line on the road without giving a turn signal.

Exactly. So far, most automatic systems have dealt only with the speed of the car - transmissions, accelerators and breaks. Now we're starting to see some high-end cars that can automate aspects of the steering - Have you seen the ads for the car that parallel-parks itself?

Add in some side-scan radar, and you could have cars that steer to maintain their lane positons, as well as the separation between leading and following cars.

I've also imagined a WiFi system for cars. Each car talks to the cars around itself, to pass info on road conditions. Imagine if a road was too icy to drive on - your car sends that info back along the line, so later cars can plan to exit the highway and take an alternate route. With good enough sensors, you could detect all sorts of such hazards, and share the info between all the cars on that road. This info could also be used to update a real-time road info file that anyone could access, such as for road repair work or police alerts.

And it wouldn't take too many cars with such sensors to have a big effect. Even if only 1% of cars had top-of-the-line packages, other cars could still make use of the info. When you think of how many cars pass any given point on a major road every day, 1% still gives you pretty accurate up-to-the-minute information. We'd have a lot better knowledge of road conditions than we do now.

 

[MRC_Hans]

And it wouldn't take too many cars with such sensors to have a big effect. Even if only 1% of cars had top-of-the-line packages, other cars could still make use of the info. When you think of how many cars pass any given point on a major road every day, 1% still gives you pretty accurate up-to-the-minute information. We'd have a lot better knowledge of road conditions than we do now.

This is an important point. We had an experiment here some years ago, where they equipped some cars with automatic speed-limit alarms (you don't need local transmiters, just a GPS and a map with speed limit info). They forgures out that just 10% of the cars equipped with this would significantly impact he average speed on roads. Even if you end up overtaking him, that pesky law-abiding driver ahead of you will keep you slowed for some time.

Hans

 

[Horatius]

This is an important point. We had an experiment here some years ago, where they equipped some cars with automatic speed-limit alarms (you don't need local transmiters, just a GPS and a map with speed limit info). They forgures out that just 10% of the cars equipped with this would significantly impact he average speed on roads. Even if you end up overtaking him, that pesky law-abiding driver ahead of you will keep you slowed for some time.

Hans

Under our current set-up, I'd find that incredibly annoying.

I tend to speed like hell when going between cities, simply because long drives like that are one of the most boring things in the universe. I go really fast in order to make the drive as short as possible.

If the car could drive itself, an extra hour or two on the road wouldn't be so bad, because I'd be able to read a book or watch TV or something, rather than just stare at the road in front of me. So I wouldn't care so much if I was only doing the speed limit.

 

[Ripley Twenty-Nine]

Under our current set-up, I'd find that incredibly annoying.

I tend to speed like hell when going between cities, simply because long drives like that are one of the most boring things in the universe. I go really fast in order to make the drive as short as possible.

If the car could drive itself, an extra hour or two on the road wouldn't be so bad, because I'd be able to read a book or watch TV or something, rather than just stare at the road in front of me. So I wouldn't care so much if I was only doing the speed limit.

I think one of the major selling points of these 'self-driving cars' would be that speed limits could be increased quite a bit. For instance, if you are on a highway where only self-driving cars are permitted to drive on, why not raise the speed limit to 180km/hr? I've seen videos of lines of these self-driving cars driving at extremely fast speeds, in perfect precision about a foot away from each other.

Also, if these onboard computers had a live always on Internet link (Which could easily be done today through Cellemetry or Satellite), and up to the minute road, traffic and weather conditions, speeds limits could be variable by the second, and every car would automagically adjust.

It opens up an awful lot of exciting possibilities for sure.

 

[Dark Jaguar]

The final step when this is all in place is to eliminate drivers altogether, which is to say, make it illegal to drive so as to eliminate the source of all accidents, human error, as best as possible. I don't see this as stepping on civil liberties as driving on roads is currently a privilage as it is. I wouldn't go so far as to apply it to one's own property of course.

 

[Loss Leader]

Isn't the most usefull aplication for self-driving cars military. We send out Humvees in Bahgdad just to drive around and look for trouble. With self-driving cars, they could drive a circuit while soldiers back at camp watched the streets through nice HD cameras. The cars could function in normal traffic with a little help from onboard maps. They could be significantly lighter because they would have a reduced need for armor. They could also speed away with little or no contoller input at the first sign of danger.

That's seems the best use for them in the near-term.

 

[Horatius]

I think one of the major selling points of these 'self-driving cars' would be that speed limits could be increased quite a bit. For instance, if you are on a highway where only self-driving cars are permitted to drive on, why not raise the speed limit to 180km/hr? I've seen videos of lines of these self-driving cars driving at extremely fast speeds, in perfect precision about a foot away from each other.

Well, that sort of top speed might need all new roads, as most aren't designed for that speed. But we could easily do 120-140km/h, as that's pretty common already. I think the hard part would be getting the political will to make these changes.

The final step when this is all in place is to eliminate drivers altogether, which is to say, make it illegal to drive so as to eliminate the source of all accidents, human error, as best as possible. I don't see this as stepping on civil liberties as driving on roads is currently a privilage as it is. I wouldn't go so far as to apply it to one's own property of course.

This would also probably evolve over time. I don't see it being illegal everywhere, but gradually, more options will come into play, that make people more likely to choose such options voluntarily. Picture, as an initial step, a dedicated lane (like HOV lanes) for self-drive cars. As we upgrade roads, such lanes could be physically isolated from others, like the Express lanes on the 401 through Toronto. Eventually, some all-new self drive highways could be built, on the same basis as Toll roads. You wouldn't need to use them, but it would be a lot easier for some purposes, so you would choose to use them.

Eventually, it would just become the "new normal", no actual laws required.

 

[Orangutan]

Aren't self-driving cars impractical?

It's not about a self driving car, It's about the things that get learned when you try and build a self driving car.
imagine asking "Isn't going to the moon impractical" 60 years ago.
Here's the page to the DARPA grand challenge
http://www.google.com/search?hl=en&...t=result&cd=1&q=darpa+grand+challenge&spell=1
Last time it was in a desert, this time it's going urban.

 

[Michael Redman]

Cars that prevented almost all accidents would pay for themselves many time over, even if the system were very expensive to build.

Think of the cost of insurance. Now think of not having to have insurance.

 

[ZeeGerman]

I think one of the major selling points of these 'self-driving cars' would be that speed limits could be increased quite a bit. For instance, if you are on a highway where only self-driving cars are permitted to drive on, why not raise the speed limit to 180km/hr? I've seen videos of lines of these self-driving cars driving at extremely fast speeds, in perfect precision about a foot away from each other.

Why would I want to wait for self-driving cars to go 180? And what kind of sissy speed limit is that anyway? I just made the 100 km from Audi back to my office in under 45 minutes topping 220 for long stretches, all legally of course
(Mein Auto fährt auch ohne Wald).

Anyway, I agree that the development will be evolutionary and we might see "hands free" freeway driving within 10 years from now, probably first in trucks (platooning, "electronic draw bar"). Also, car to car communication (C2CC) might be around by then. The problem here is to agree on a industry standard communications protocol. Not easy, but it can be done, even world wide as the example of KWP2000 as diagnosis protocol proved.
Infrastructure investments are probably not as necessary as it might seem, since wirless communication (GSM, WiMax) together with onboard GPS-navigation and map material (which will be a commodity pretty soon in new cars) could broadcast information like speed limits, warnings, road blocks etc.
The camera technology is actually quite more advanced than the OP suggested. My company is currently experimenting with a camera system that provides true 3D information without stereo vision by giving a grayscale image + a distance measurement for each pixel. We will see what can be done with such a device on embedded automotive ECUs pretty soon. Also, infrared systems are quite helpful to detect pedestrians/cyclists or animals. Night vision systems are available already and given Moore's law, the computing power on low price automotive grade HW will sonn be enough to run sophisticated algorithms. And these will be provided by guys like Sebastian Thrun (referred to in the OP). I know him personally - I even proof-read his master thesis back in 92, how's that for boasting - extremely smart guy who probably knows more about autonomous mobile robots than any man on earth right now.


Zee

 

[Ziggurat]

http://money.cnn.com/2007/02/14/mag...tomated.biz2/index.htm?postversion=2007021506

I think it will be a long, long time before we ever see a self-driving car, if ever, for two main reasons:

#1. Detecting road signs and traffic lights. It would be incredibly expensive to outfit every street in America with radio transmitters to tell self-driving cars the proper speed limit. Not to mention this huge project would be impractical: Why should the government spend billions just so lazy people can have "self-driving cars"? Road signs couldn't be programmed directly into the car itself, like in the OnStar system, either because of how they frequently change (I.E. during construction). Plus you can't
really detect traffic lights.

#2. I don't believe we have the kind of optics to be able to always accurately distinguish between various objects on the road, certainly not good enough to accurately track movement. It was a real challenge for us to develop a UGV which could even move through a static obstacle course... That's worlds away from ever building one that could detect the color of traffic lights, distinguish between bicyclists, autos, pedestrians and animals, and be able to move accordingly.

Several points: first, keep in mind who's funding this: DARPA. Does DARPA want a Lexus that can drive itself? No, it doesn't care about that. DARPA wants a transport vehicle that can move supplies from one location to another without a driver, or a scout vehicle that can patrol an area autonomously. And it wants this sort of thing for military applications. Rigorous adherence to traffic laws isn't their priority.

Furthermore, when driving on roads, the robot doesn't need to be able to recognize all the signs. Roads and street signs are essentially static, you don't need to put radio transmitters everywhere, all you need is a database which contains that information. And we already HAVE something along those lines: it's called MapQuest. Store all the information about stuff like stop signs, speed limits, etc. in a database, and GPS will tell you what you need to know. As for stoplights, well, actually that's NOT that hard. Stop lights are fairly regular objects, if you know the intersection has one (from your database), looking for it and noting which light is on isn't actually that hard (not much harder, at any rate, than recognizing unknown obstacles).

And lastly, if you can get obstacle recognition working (which they've already done), dealing with moving obstacles really isn't a lot harder. You need more processing power to track movement, sure, but the hard part is recognizing the object in the first place. And you don't NEED to be able to distinguish between a deer and a bicyclist: you just need to know that both of them are objects big enough that you shouldn't hit them.

 

[Earthborn]

When these things appear, they will be incredibly practical. Think about it; never again will you need to drive around to find a parking space. The car drops you off at your destination, and finds a parking space for itself. Often cars could be tightly packed in automated parking garages. When you want to go back home, you'll just press the button on a transmitter no bigger than those central lock thingies are now, and your car will pick you up. No time to pick up the kids from school? Use your webbrowser at work to program your car remotely to pick them up for you.

A self-driving car becomes even more practical when it communicates with all the other cars on the road. Such networked cars can work out the most efficient all encompassing transport plan together. Not all people would need to own a car, but rather a witkar concept can be made to work. The witkar concept failed primarily because on many times of the day many people try to move in the same direction which causes shortages of witkars in one area, and surplusses in the other. Networked cars can automatically divide themselves evenly, and if there is always a small percentage of cars on the move, there can be one nearby for everyone.

I once saw a documentary in which automated cars were shown that were used on test tracks just to test new cars. One of the technicians argued that fully automated cars would still be a long way off, because cars on the road are not always in good condition. They may have poor tires, people don't always refill their oil on time. A robot car would not work safely if someone doesn't maintain it, because its driving characteristics would change in ways it would not have been programmed for. I don't think this is necessarily a problem, as long as the car has good self diagnostics. A robot car does not have to wait for its owner to take it to a garage for a check up, it can drive itself. The car finds that it is time for a check up, and it automatically requests a loan car so the owner doesn't have to go without. The loan car from the garage drives itself to the owner's house, and when it arrives the owner's own car drive itself to the garage. Nobody needs to be a second without a car just because it needs to be in the garage.

I think the biggest influence of autonomous vehicles will be for freight. The constant pressure on truckers to ignore drive time regulations means that the possibility of having constant 24/7 movement of goods will just be too good to pass. Trucks will probably start to look similar to this as no cab for the driver is necessary. (Passenger cars may also lose the front/back distinction.) Trucks like this might be on the move for much longer than any driver can stay awake, and be much safer.

If this sort of technology can be further miniaturised, an entirely new concept of transport may become possible that would solve many of the problem with current mail packages. I imagine suitcase sized vehicles that anyone can fill and order to move to any adress. It automatically drives itself on foot or bicycle paths to the destination adress. For adresses in another city it can automatically load itself onto trucks or trains or even take ordinary public transport. When it arrives it can wait until the adressee is home, and only open itself to the correct person. Perhaps it could even be send to a specific person instead of an adress and find that person (wearing a GPS locator) whereever s/he is. It opens up the possibility of package delivery where one does not have to wait at home for something to arrive or need large mail boxes that thieves might open as easily as the postman. Sending goods might become as easy as sending email (although obviously slower).

 

[casebro]

Just think of the potential for terrorism!

Some geek comes up with a virus, and how many commuters die? One bullet into the computer of the lead car, how do the rest respond when it suddenly locks up?

Or the 'pop-up ads' aspect:

The shopping center wants you to drive all over the lot, to show you all the stores.... Sorry, we steer you off the freeway, through the busines district because, well, it's good for business... No, the scenic route is filled to it's quota, you'll have to go on the freeway....

Or, even more earth shaking, we won't each need out own car:

Auto-Cabs will take over, and pick you up at your door, drop you at your door, why have a car sitting in your driveway all day? or the parking garage? Do away with mass transit, each cab would know the most efficient route to pick up more riders, for each individual's destination, No more single occupancy vehicles.

Of course, Big Brother will be watching, planning out the route, you're gonna take real soon, checking to see who is naughty or nice...

 

[Molinaro]

I think road safety could be greatly increased without having to go to self-driving cars. Simply implant every car with a GPS receiver, and have detectors at every intersection of every road. When you pass by it calculates your average speed between that detector and the last one you passed. If it calculates that you were over the limit, you get a fine in the mail.

I would very much like something like that in place, everywhere.

 

[Ripley Twenty-Nine]

I think road safety could be greatly increased without having to go to self-driving cars. Simply implant every car with a GPS receiver, and have detectors at every intersection of every road. When you pass by it calculates your average speed between that detector and the last one you passed. If it calculates that you were over the limit, you get a fine in the mail.

I would very much like something like that in place, everywhere.

Well of course you would. You live in Toronto, and never get above 30km/hr anyway.

 

[aggle-rithm]

I think road safety could be greatly increased without having to go to self-driving cars. Simply implant every car with a GPS receiver, and have detectors at every intersection of every road. When you pass by it calculates your average speed between that detector and the last one you passed. If it calculates that you were over the limit, you get a fine in the mail.

I would very much like something like that in place, everywhere.

I think a simpler solution would be to have an adjustable governor in each car that prevents it from exceeding the speed limit in it's current zone.

Unless you LIKE bureaucracy!