Aren't self-driving cars impractical?

[timhau]

No, it doesn't. But the car that automatically drove most of the way coast to coast through the US in 1997 was said to be unable to do it, so it is great to see that there has been progress in this matter.

I didn't know there was such a car. What kind of a route did it take? I assume you could go most of the way on a coast-to-coast trip on wide-open, straight highways, which I, without claiming any sort of expertise whatsoever, assume to be the easiest to automate (esp. if you can count on having a human observer ready to take over if it looks like the automaton will fail).

BTW, that video reminds me of what my wife and I went through when we purchased a high-tech all-in-one remote for the TV, DVD player, and digital receiver. All it convinces me of is that these guys didn't read the manual. (Of course, the fact that you can't just drive the car next to a vacant spot and tell it "OK, park yourself" shows that it really isn't fully self-parking.)

 

[timhau]

In short, it could easily be made today and all in all would it probably just make the cars 1-200$ more expensive but,

Really? I mean, really? I'm about to go home. I'll walk, but since it's about a 35 minute trip on foot and it's pretty cold outside, I'd appreciate a ride. Can you really make a car that would get me home through the downtown area of our 175,000 inhabitant town? Go straight for many, many blocks (most of them with traffic lights), then turn right to a three-lane one-way street, go several blocks (again, most intersections have traffic lights), turn left and go 2.5 blocks and park, avoiding not only all other cars out there but also pedestrians (some of whom have kids and/or dogs in tow) and suicide cyclists -- and all this for a 200-dollar extra investment to our non-self-driving Yaris?

You'll have to do a lot better to convince me than to appeal to The Man wanting to keep driving fun.

 

[ponderingturtle]

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.

Signs are easy, you don't need RF transmitters, you just need to teach a vision system to recognise a sign, and they are standard so it is not that hard.

 

[Crazycowbob]

In short, it could easily be made today and all in all would it probably just make the cars 1-200$ more expensive but, but, but, -it hasn't got a chance.

Wow, when was the last time you shopped for cars? 1-200 dollars won't even upgrade you to powered seats.

Not to mention, personally, I don't think governers solve anything. While speed definately contributes to the severity of a crash, it's not the cause, poor driving is. If everyone on the road is going 50mph, then the safest speed for you is 50mph. People driving 10mph faster/slower are driving dangerously, as well as those cutting through lanes, or stopping in the middle of a busy street.

Sorry, but that's just one of my pet peeves, because around here, cops only care about speeding. I've seen a person cut across 3 lanes on the highway, in sight of a cop, and get away clean. Sure you need to keep your speed within reason, but the real killers out there are the people driving like idiots, the people the cops ignore because they're not easy or profitable to prosecute. The only way to get rid of them is to automate driving completely, though I'd have to wonder how many of them would still opt for the "manual" drive option, just so they can cut around all the other cars and get to their destination those all important 5min earlier...

 

[Cuddles]

Signs are easy, you don't need RF transmitters, you just need to teach a vision system to recognise a sign, and they are standard so it is not that hard.

And then when it's foggy...

 

[ponderingturtle]

And then when it's foggy...

How is having a sign obscured suddenly a new problem with computer driven cars?

They might also be able to use non visible light to look for such things.

As for how expensive it would be, probably a few thousand dollars per unit. This will greatly depend on how many are made as it is all about the cameras and computers and such.

 

[Ziggurat]

And then when it's foggy...

You do the same thing people do when it's foggy: slow down.

 

[Ripley Twenty-Nine]

Signs are easy, you don't need RF transmitters, you just need to teach a vision system to recognise a sign, and they are standard so it is not that hard.

As others have pointed out, I think this is the wrong approach.

I already have mapping software on my Pocket PC that when connected to my matchbook sized GPS device can tell me the following:
1) Exactly where I am within 1-2 feet accuracy
2) The name of the street I am on
3) Whether the street is One Way or not.
4) Where the next intersection is
5) The current speed limit
6) My current speed
7) The exact route to get to my destination
8) How long it will take me to get to my destination based on my average speed, and the speed limits on the roads I will take to get there

So we're not that far off as it is. The only problem with it, is that the maps are not completely up to date. But who has this information? Municipalities. When new roads are being built, or construction is happened on existing roads they have all of this information. If this information was constantly being updated in a publically available database, the car would have absolutely no need to read street signs.

 

[Ladewig]

So suppose that we have a self-driving car that waits for the light to turn green, goes, and is slammed by someone running the red. The "driving passenger" of the self-driving car wasn't looking and doesn't know what the light did. The person who was running the red claims that they had the light. The car manufacturer claims that their car won't go unless the light is in their favour. Discovery turns up an obscure email where an engineer talks about how a combination of poor light placement and an unwashed car can confuse the car about whether it sees a regular light (don't go on red) or a blinking red (you can go on red).

I am having a hard time buying it. The marginal cost of wide-angle pulse radar would be low enough that automated cars waiting for a green light could easily detect vehicles accellerating towards the intersection. Also, by that time the most dangerous intersections will have "red light cameras." Many U.S. cities have already started installing them.

 

[Horatius]

And then when it's foggy...

Except that, unlike the Mark I eyeball, the sign sensors aren't restricted to visible light. Use a wider spectrum, and you could still make out most signs. You might need to check the IR/UV reflectance of the paint or retroreflectors used, but other than that, no problem.

 

[Horatius]

I am having a hard time buying it. The marginal cost of wide-angle pulse radar would be low enough that automated cars waiting for a green light could easily detect vehicles accellerating towards the intersection. Also, by that time the most dangerous intersections will have "red light cameras." Many U.S. cities have already started installing them.

It's also unlikely that you'd have both human and machine driven cars on the same streets.

Also, even if they did share roads, if the human-driven car was still plugged into the car-to-car network, it could send a "stupid human trick" warning along the net, so other cars would stay out of his way. And any cop cars nearby would get the same warning

 

[ponderingturtle]

As others have pointed out, I think this is the wrong approach.

I already have mapping software on my Pocket PC that when connected to my matchbook sized GPS device can tell me the following:
1) Exactly where I am within 1-2 feet accuracy
2) The name of the street I am on
3) Whether the street is One Way or not.
4) Where the next intersection is
5) The current speed limit
6) My current speed
7) The exact route to get to my destination
8) How long it will take me to get to my destination based on my average speed, and the speed limits on the roads I will take to get there

So we're not that far off as it is. The only problem with it, is that the maps are not completely up to date. But who has this information? Municipalities. When new roads are being built, or construction is happened on existing roads they have all of this information. If this information was constantly being updated in a publically available database, the car would have absolutely no need to read street signs.

I guess. Good thing that muncipalities can instantly update things like detours, construction issues, when a staircase is improperly labeled as a road and the car decides it wants to take a left and drive down it. Also roads that are on maps but where never built.

You need to use a vision system to recognise all sorts of issues involving things like not running people over and not driving down a staircase. Why is having it recognise signs, something that software I work with every day can do such a bad approach?

 

[Earthborn]

What kind of a route did it take?

It's on their T-shirt design on the official website. I was wrong though, they already did it in 1995.

I assume you could go most of the way on a coast-to-coast trip on wide-open, straight highways, which I, without claiming any sort of expertise whatsoever, assume to be the easiest to automate (esp. if you can count on having a human observer ready to take over if it looks like the automaton will fail).

That's pretty much how they did it.

Of course, the fact that you can't just drive the car next to a vacant spot and tell it "OK, park yourself" shows that it really isn't fully self-parking.

Well, it isn't fully self-parking if you can't get out of the car and the car finds a spot itself and parks itself there. But it is a step in the right direction.

 

[ponderingturtle]

It's also unlikely that you'd have both human and machine driven cars on the same streets.

So you are not proposing mearly putting in added features to roads for robotic cars, but their own entire infarstructure?

Any self driving car will have to be able to deal with human drivers and other issues of an uncontroled enviroment.

 

[Horatius]

So you are not proposing mearly putting in added features to roads for robotic cars, but their own entire infarstructure?

Any self driving car will have to be able to deal with human drivers and other issues of an uncontroled enviroment.

I picture it more as a gradual switch over. As robocars become more common, more dedicated lanes will be created. As new roads are built, some will be dedicated to robocars, while others will still be traditional roads.

Of course, it's hard to predict how this will all happen. As the tech matures, it may be possible to have both types of cars on the same streets, but if this system actually works well, I'd expect most people to adopt it. The die-hards who insist on driving themselves will be a quaint sub-set of car owners, sort of like the Classic Car owners these days - interesting people, but not a large number.

Think of how many people today already try to do other things while driving - using cell phones, eating, drinking, watching TV, yelling at their kids, playing with the dog, whatever. This sort of system would allow them to do that, without the insane risks. I think they'd leap at the chance.

Heck, just think of the improvement that reduced drunk driving could make. Have a big button labelled "Go Home" on the dash board, and even someone who's so drunk they can't even stand up could make it home safely.

 

[Earthborn]

I picture it more as a gradual switch over. As robocars become more common, more dedicated lanes will be created. As new roads are built, some will be dedicated to robocars, while others will still be traditional roads.

It's certainly a possibility that there will be highways with dedicated lanes where robot cars will be allowed to drive 500 km/h. But if they are going to have the same door-to-door functionality as present day cars, those things will have to mix in with the human driven cars at some point in the trip.

 

[Ripley Twenty-Nine]

I guess. Good thing that muncipalities can instantly update things like detours, construction issues, when a staircase is improperly labeled as a road and the car decides it wants to take a left and drive down it. Also roads that are on maps but where never built.

You need to use a vision system to recognise all sorts of issues involving things like not running people over and not driving down a staircase. Why is having it recognise signs, something that software I work with every day can do such a bad approach?

Well, I guess there would be a potential downside with either solution. What if a Stop sign gets knocked down? What if a Speed Limit sign is covered in snow? These are organic things that can change from one minute to the next, and those changes are much more difficult to convey.

Data is much more static. Once a road is built, it is built. Construction starts on a date, and construction ends on a date. That data already exists for municipalities, so why not make it publicly available in the form of a database?

If municipalities are identifying staircases as roads, then they've got way bigger issues.. If this kind of bad data was allowed to be entered which led to an injury or a death, then they are liable. Luckily all public services have checks and balances in place to prevent such things; imagine a doctor that enters an extra zero to your dosage by accident.

 

[ponderingturtle]

I picture it more as a gradual switch over. As robocars become more common, more dedicated lanes will be created. As new roads are built, some will be dedicated to robocars, while others will still be traditional roads.

Then you are haveing a mixed enviroment durring the transition period.

 

[ponderingturtle]

Well, I guess there would be a potential downside with either solution. What if a Stop sign gets knocked down? What if a Speed Limit sign is covered in snow? These are organic things that can change from one minute to the next, and those changes are much more difficult to convey.

What happens now?

Data is much more static. Once a road is built, it is built. Construction starts on a date, and construction ends on a date. That data already exists for municipalities, so why not make it publicly available in the form of a database?

So you have the cars ignore things like downed wires that require immediate changes

If municipalities are identifying staircases as roads, then they've got way bigger issues.. If this kind of bad data was allowed to be entered which led to an injury or a death, then they are liable. Luckily all public services have checks and balances in place to prevent such things; imagine a doctor that enters an extra zero to your dosage by accident.

It happens. There was a report that showed mapquest, google maps and I think yahoo all had this one staircase marked as a road. This is because they all buy their databases from the same company.

Link on failures of driving dirrection software

 

[Horatius]

It's certainly a possibility that there will be highways with dedicated lanes where robot cars will be allowed to drive 500 km/h. But if they are going to have the same door-to-door functionality as present day cars, those things will have to mix in with the human driven cars at some point in the trip.

Then you are haveing a mixed enviroment durring the transition period.

These comments both assume the robo-car will be driving itself all the time. You could have a system that changes from human to machine control at certain points on the road - like on/off ramps, or certain major intersections, so that while you do have a mixed sytem during the transition, the particular road you're on at the moment will be one or the other, not both.

You'd need to engineer a system so that such transfers are safe, but that's a sovable problem, IMO.