Will a car ever be able to drive itself?

[Cainkane1]

I read a science fiction story about a car that could drive without a driver. It would not allow drunks to drive themselves and would also drive to and from places with no driver at all. Could this happen anytime soon?

[Darat]

I think the answer is yes.

ETA: We already have cars that can prevent collisions, park themselves and so on.

[shadron]

http://www.roboticsproceedings.org/rss04/p23.pdf

This concerns the NASA Autonomous Driving challenge that was won, I believe, in 2007. I don't know that it kept drunks out, but that would be trivial. As to when it might become generally available, a long time. There is a big difference between country autonomous driving and urban autonomous.

[Zax63]

There was a TV series called Robocars about the DARPA challenge for self driving cars. It was pretty amazing what they can do already. One of the qualifying tests, just to get into the competition, was to stop at a 4 way intersection with other cars already there, wait until it's proper time and then make a turn. In my opinion basic driving skills are 80% here. Try searching "DARPA autonomous vehicle".
The questions for me comes when you get to the really unusual and unexpected. What does it do with a massive pothole in it's way? What if it is filled with water and you can't tell how deep it is? How about if a large plastic garbage bag blows in front of the car? Do you slam on the brakes to avoid a harmless collision? Then what about when a kid or an animal runs out in the street? How does a computer tell the difference?

[Darat]

Zax63 - good points but I would like to know more about how well the "average" driver copes with those unexpected events. I have a sneaking suspicion that it isn't as well as we'd like to think.

[Puppycow]

Ever? They already can.

Driving well is a work in progress but this is not far-off Star Trek stuff.

[Darat]

Ever? They already can.

Driving well is a work in progress but this is not far-off Star Trek stuff.

So better than most people on the road today?

I would have thought a Sinclair ZX81, without a RAM pack, connected to a C5 would be able to do that! :D

[arthwollipot]

This came up in my feed yesterday:

Ten Coolest New Car Technologies (http://www.geeksaresexy.net/2009/08/12/ten-coolest-new-car-technologies/)

With collision avoidance, lane detection and GPS navigation, I don't see how we can avoid it.

[Jungle Jim]

I can forsee a time in the future (50-100 yrs. ?) when you won't even drive your car. Sensors in the road will control speed and navigation. You just input your destination and off you go.

[richardm]

There is a big difference between country autonomous driving and urban autonomous.

Yeah, but I suppose these systems don't have to work perfectly all the time to be useful. I can see how an automatic car would work well in the low-speed urban commute that we see so often, and I can see how it could work well on a motorway. Perhaps on twisty, badly-marked country roads it wouldn't work so well, but then you can't use adaptive cruise control so well on those roads either.

How does it cope with roadworks and other unexpected changes to road conditions? Same way your cruise control works when you come to roadworks or a hole in the ground, perhaps - you switch it off as you approach. Yes, you'll get idiots who'll plough into the flooded road and get stuck. We have those already. At least they won't plough into the back of a queue of traffic so often.

If your idea of an automatic car is that you can sit sipping cocktails and doing the crossword as it wafts you along then I think you have a longer wait, but something that would mean you didn't have to concentrate on controlling the car all the time would be very desirable and a pretty good aid to road safety, I think. And not too far away either.

I can forsee a time in the future (50-100 yrs. ?) when you won't even drive your car. Sensors in the road will control speed and navigation. You just input your destination and off you go.

That'll be great for new roads, but surely a hideous expense to go and retro-fit to old ones.

[ponderingturtle]

There was a TV series called Robocars about the DARPA challenge for self driving cars. It was pretty amazing what they can do already. One of the qualifying tests, just to get into the competition, was to stop at a 4 way intersection with other cars already there, wait until it's proper time and then make a turn. In my opinion basic driving skills are 80% here. Try searching "DARPA autonomous vehicle".
The questions for me comes when you get to the really unusual and unexpected. What does it do with a massive pothole in it's way? What if it is filled with water and you can't tell how deep it is? How about if a large plastic garbage bag blows in front of the car? Do you slam on the brakes to avoid a harmless collision? Then what about when a kid or an animal runs out in the street? How does a computer tell the difference?

I would not be supprised if we see some use of this for the military in the next 10 years or so. Civilian use would be more complicated, who is libel if your self driving car rearends someone?

[ponderingturtle]

That'll be great for new roads, but surely a hideous expense to go and retro-fit to old ones.

Of course it will. Decades ago they had cars that could follow magnets built into lanes in specialy built roads. Never caught on anywhere. It needs to be something that will work on existing road networks.

[William Parcher]

It would seem that such a vehicle would require a manual override (autopilot off) for a variety of reasons. You need to have that thing operable just like a traditional vehicle for certain situations.

[MRC_Hans]

I think we can already build a car that is safer than a lousy driver. We may soon approach the safety of an average driver, and once there, the lightning reactions of a computer will soon make it surpass even excellent drivers in safety.

Effectivity is another matter. We already have autonomous Metro trains in Copenhagen, and they don't have accidents, but they DO have a lot of unnecessary stoppages because the system is overcautious. Like someone mentioned above, your automatic car will probably slam on the brakes if a large plastic bag blows in front of it, because even if the sensors can get a good cue that this is not a critical obstacle, who is going to take the chance?

And this brings us to the biggest obstacle: Autonomous cars have to be MUCH better than even excellent human drivers before people are going to be willing to leave the wheel to them (remember, in their own view, most people are excellent drivers).

We simply HATE not to be in control.

For military purposes, they are probably already there, and for research, we have had autonomous cars on Mars.

Hans

[MRC_Hans]

Of course it will. Decades ago they had cars that could follow magnets built into lanes in specialy built roads. Never caught on anywhere. It needs to be something that will work on existing road networks.That is a small problem: Differential GPS requires transmitters at intervals along the road, but that is manageable, and it can determine your position with millimeter precision.

Hans

[MRC_Hans]

It would seem that such a vehicle would require a manual override (autopilot off) for a variety of reasons. You need to have that thing operable just like a traditional vehicle for certain situations.Yes, of course. If only to be able to manage the car if the system breaks down, and in off-road situations.

Hans

[William Parcher]

Another scenario where you might want/need an override is in the event of a crime. Any armed robber or murderer could step in front of the vehicle and force a stop. The "driver" (who is actually a passenger during autopilot) could not engage any evasive/defensive maneuver including mashing the accelerator to waste the bastard.

[arthwollipot]

I can forsee a time in the future (50-100 yrs. ?) when you won't even drive your car. Sensors in the road will control speed and navigation. You just input your destination and off you go.That would require a massive expeniture on infrastructure. I think it's more likely that we'll develop autonomous vehicles that don't require external guidance. Like, say, what we've already got...

[Freethinker]

I'd guess that we might first see self guided vehicles in limited areas such as the US Interstate Highway System. No traffic signals or stop signs. Commercial trucks would seem to be likely first adopters of a system that included onboard intelligence and sensors, and an intelligent and transducer equipped road.

Imagine a dozen or more trucks running in a "trucks only" lane at constant speed with only inches between bumpers, in continuous contact to maintain spacing, across the US plains. Braking reaction time would be virtually instantaneous, and sensors in the road could analyze traffic for miles so vehicles could anticipate other traffic. It could even schedule pull-offs to allow faster vehicles to move ahead of slower ones and accomodate human drivers. A huge benefit of this would be that automobile traffic flow would be much smoother without the commercial truck traffic in their lanes.

[Zax63]

If anyone is interested here is a gallery of video and pictures from the DARPA Urban challenge (http://www.darpa.mil/grandchallenge/gallery.asp).

I think one of the biggest hurdles in adopting self driving cars will be public relations. Even though it may save tens of thousands of lives per year vs human drivers, one "Robot car crashes into shopping mall, 4 killed" headline is going to weigh very heavily in the minds of the average person.
Another issue is legal liability. If insurance companies won't cover them or it is prohibitively expensive then it is dead in the water. It also opens up the car / software makers to a lot of liability. They may need new laws to limit their exposure before it becomes economically viable.

[ksbluesfan]

This would be wonderful for blind people.

[BobG]

This would be wonderful for blind people.

How about the elderly, many of who stay on the roads long past their capabilities allow.

Here in New Jersey drivers never get retested once they get their liscenses.
I realize that never is a strong word, but I've never seen it done.

Bob

[BobG]

http://www.roboticsproceedings.org/rss04/p23.pdf

This concerns the NASA Autonomous Driving challenge that was won, I believe, in 2007. I don't know that it kept drunks out, but that would be trivial. As to when it might become generally available, a long time. There is a big difference between country autonomous driving and urban autonomous.

Guys,

This link has caused my browser to lock up twice!

The second time was after a reboot.

I'm going to break away and run a virus check on my computer.

Bob

[maxfrost]

I for one welcome our autonomous automobile masters.

[Richard Masters]

I would not be supprised if we see some use of this for the military in the next 10 years or so. Civilian use would be more complicated, who is libel if your self driving car rearends someone?

You would be liable, but then cars like these would have sensors (black boxes) that record the incident, so you would know if it was human error or a bug in the system, or the other driver's fault.

[Richard Masters]

I think we can already build a car that is safer than a lousy driver. We may soon approach the safety of an average driver, and once there, the lightning reactions of a computer will soon make it surpass even excellent drivers in safety.

Effectivity is another matter. We already have autonomous Metro trains in Copenhagen, and they don't have accidents, but they DO have a lot of unnecessary stoppages because the system is overcautious. Like someone mentioned above, your automatic car will probably slam on the brakes if a large plastic bag blows in front of it, because even if the sensors can get a good cue that this is not a critical obstacle, who is going to take the chance?

And this brings us to the biggest obstacle: Autonomous cars have to be MUCH better than even excellent human drivers before people are going to be willing to leave the wheel to them (remember, in their own view, most people are excellent drivers).

We simply HATE not to be in control.

For military purposes, they are probably already there, and for research, we have had autonomous cars on Mars.

Hans

Don't forget that people will intentionally mod their cars to drive more aggressively, just like they currently install all kinds of custom parts to make the car lighter, faster, or more powerful. A smart aggressive car cutting you off, will certainly be an interesting sight.

[Richard Masters]

How about the elderly, many of who stay on the roads long past their capabilities allow.

Here in New Jersey drivers never get retested once they get their liscenses.
I realize that never is a strong word, but I've never seen it done.

Bob

I think New Jersey is among the worst states in terms of driving.

[ohms]

Vehicles avoid moving objects autonomously (http://www.defensenews.com/story.php?i=4232197&c=AME&s=LAN).

[BobG]

I think New Jersey is among the worst states in terms of driving.

I do not know about that but we do not give drivers a genuine driving test.

There is no test on a real road. A new driver is required to drive a vehicle in a very large parking lot with markings, cones, etc. and there is no traffic to speak of.

The tester has no idea how this person drives on real roads.

Does this make any sense to anybody?

Bob

[BobG]

Don't forget that people will intentionally mod their cars to drive more aggressively, just like they currently install all kinds of custom parts to make the car lighter, faster, or more powerful. A smart aggressive car cutting you off, will certainly be an interesting sight.

This car would be software controlled and the software would need to be modified. The designers certainly would not make that easy; in fact if the code is not in the pubblic domain, how could anybody modify it?

Bob

[BobG]

Vehicles avoid moving objects autonomously (http://www.defensenews.com/story.php?i=4232197&c=AME&s=LAN).

This certainly answer the original question of this thread!

Bob

[Rat]

I do not know about that but we do not give drivers a genuine driving test.

There is no test on a real road. A new driver is required to drive a vehicle in a very large parking lot with markings, cones, etc. and there is no traffic to speak of.

The tester has no idea how this person drives on real roads.

Does this make any sense to anybody?

Bob
Not to me. I think there should be such areas for very beginners to start in, but for a test, that's ludicrous. I, and I assume everybody else in the UK, had to go through medium to heavy traffic, at speeds between 30 and 50 mph, through several busy roundabouts and difficult junctions, while everybody else on the road was a real person going about their everyday business, so an accident was a real accident. Even so, I didn't feel confident driving for some time afterwards. People who do the NJ test must often be an absolute menace once on real roads.

[BobG]

I for one welcome our autonomous automobile masters.

I agree.

Technology is rarely responsible for accidents nowadays; it's human failure.

Safety would improve greatly and the accident/injury/fatality rate would drop dramaticly.

Bob

[lightfire22000]

There was a TV series called Robocars about the DARPA challenge for self driving cars. It was pretty amazing what they can do already. One of the qualifying tests, just to get into the competition, was to stop at a 4 way intersection with other cars already there, wait until it's proper time and then make a turn. In my opinion basic driving skills are 80% here. Try searching "DARPA autonomous vehicle".
The questions for me comes when you get to the really unusual and unexpected. What does it do with a massive pothole in it's way? What if it is filled with water and you can't tell how deep it is? How about if a large plastic garbage bag blows in front of the car? Do you slam on the brakes to avoid a harmless collision? Then what about when a kid or an animal runs out in the street? How does a computer tell the difference?

There making a lot of progress already on those fronts. We have robotic drivers that are probably better than some of the human drivers out on the road.

[BobG]

Not to me. I think there should be such areas for very beginners to start in, but for a test, that's ludicrous.

This is a very good idea.

I took my driving test back in the 60s in New York on a real road in real traffic. I'm sure that, at that time, it was this way in the whole country.

It was very dangerous for the testers for obvious reasons as I'm sure is the case across the pond.

So now that you have mentioned it, I see a two part test; first on the artificial road and then, after the testee has demonstrated a basic skill level, the tester would take him on the road.

I've often wondered if the danger to the testers on real roads was the provocative factor for this ridiculous system we now have.

Bob

[Richard Masters]

This car would be software controlled and the software would need to be modified. The designers certainly would not make that easy; in fact if the code is not in the pubblic domain, how could anybody modify it?

Bob

There are several ways. XBOX consoles have been modified by removing protection chips, inserting new ones, reformatting them into Linux-based systems capable of storing entire game DVD images.

You can jailbreak out of iPhones and add capabilities that are typically not available otherwise.

The NES and SNES consoles are not public domain and they were reverse-engineered to provide the public with NES and SNES emulators. Today, you can download and play any SNES video game on your computer.

DRM protection systems in the music industry have failed completely because some genius hacks them within a day or two of their release.

There are programs that let you convert iPod music into full-fledged mp3s without any licensing restrictions.

It's not something just anybody can do, but there are people who have the time and experience hacking some of these systems, who can easily do so for the community, given enough motivation. Once the hacks are released, anybody with or without programming knowledge can further mod the behavior of their cars.

[makaya325]

The answer likely is: No

I could see the development of self-driving cars by the use of magnetic rails, however, If we wanted to design self driving cars, it would require some kind of supercomputer memory regarding routes, roads, turnpikes, etc.

[BobG]

There are several ways. XBOX consoles have been modified by removing protection chips, inserting new ones, reformatting them into Linux-based systems capable of storing entire game DVD images.

You can jailbreak out of iPhones and add capabilities that are typically not available otherwise.

The NES and SNES consoles are not public domain and they were reverse-engineered to provide the public with NES and SNES emulators. Today, you can download and play any SNES video game on your computer.

DRM protection systems in the music industry have failed completely because some genius hacks them within a day or two of their release.

There are programs that let you convert iPod music into full-fledged mp3s without any licensing restrictions.

It's not something just anybody can do, but there are people who have the time and experience hacking some of these systems, who can easily do so for the community, given enough motivation. Once the hacks are released, anybody with or without programming knowledge can further mod the behavior of their cars.

OK. You seem to have the expertise so it must be so.

But I still don't see this being an impediment. Yes, you will have accidents because of this but we have accidents now and, in my opinion, it would still be a safer system; technilogical error versus human error. In other words, we will have less accidents.

Also, there would be legal ramifications to suping up cars driven on the public road so people would be taking a big chance. I'm talking about criminal liabilities! And the roads would still be patrolled by policemen which would limit this practice.

Bob

[Rat]

This is a very good idea.

I took my driving test back in the 60s in New York on a real road in real traffic. I'm sure that, at that time, it was this way in the whole country.

It was very dangerous for the testers for obvious reasons as I'm sure is the case across the pond.
Indeed, I took my very first lesson, the first time I'd ever been in the driving seat, in my own street here (http://maps.google.co.uk/?ie=UTF8&ll=52.589349,-1.105542&spn=0.001253,0.003484&t=h&z=19), a quiet but narrow road with cars parked both sides. I was absolutely terrified. I think people forget how alien it is the first time you ease off on the clutch and the car starts moving forwards. I think a test where one can demonstrate one's ability to do tricky manoeuvres in a safe environment, followed by a test of one's ability to cope with real-world traffic, would be a perfect combination.

[Richard Masters]

OK. You seem to have the expertise so it must be so.

But I still don't see this being an impediment. Yes, you will have accidents because of this but we have accidents now and, in my opinion, it would still be a safer system; technilogical error versus human error. In other words, we will have less accidents.

Also, there would be legal ramifications to suping up cars driven on the public road so people would be taking a big chance. I'm talking about criminal liabilities! And the roads would still be patrolled by policemen which would limit this practice.

Bob

I don't see this as an impediment either. In fact, some custom modifications may render the cars safer than the vanilla settings.

[makaya325]

How would a self-driving car recognize what road it is on? Would it have a built in memory to recognize the physical world?

[Richard Masters]

How would a self-driving car recognize what road it is on? Would it have a built in memory to recognize the physical world?

GPS for geographical location and Google Maps for identifying the current road.

[Richard Masters]

GPS for geographical location and Google Maps for identifying the current road.

ETA: a quick snapshot of the signs at the intersection would be an additional safety measure to account for any discrepancies in the Google Maps database.

[Olowkow]

This would be wonderful for blind people.
...once they are out of the garage and down the driveway and on to the automated roadway...:) Oh, then the braille road signs.

[makaya325]

GPS for geographical location and Google Maps for identifying the current road.

But how will it know where it is going? That, my friend, requires artificial intelligence, something that is in the baby stages

[epepke]

Let's say you have self-driving car.

Now the first person dies in an accident involving one.

And the lawyer proved to 12 morons, good and true, that it was due to a programming error.

Where do you think the tiny shards of the entire industry are going to be stored after the award?

[BobG]

The answer likely is: No

If we wanted to design self driving cars, it would require some kind of supercomputer memory regarding routes, roads, turnpikes, etc.

That's not a problem. I tend to think that the computer technology is already here.

And if not, computer power doubles every one to two years and this exponential increase is going to continue for quite some time.

Remember! Throughout history there always have been naysayers stating that this or that could not be done. Yet, it's all been done!

Bob

[Earthborn]

Here (http://www.cs.cmu.edu/afs/cs/usr/tjochem/www/nhaa/Journal.html) you can read about what was already achieved in 1995, apparently even before the technology of "pretty pictures on the internet" was perfected. So, yes, cars will one day in the not too distant future be able to drive themselves.

This car would be software controlled and the software would need to be modified. The designers certainly would not make that easy; in fact if the code is not in the pubblic domain, how could anybody modify it?I think it is more likely that the software will be in open source, so it can be independently tested. The car computer can be in a sealed box of course to discourage non-experts to temper with it.

[makaya325]

Remember! Throughout history there always have been naysayers stating that this or that could not be done. Yet, it's all been done!

There is a difference between being realistic and living in a fantasy world. For a self-driving car to work, it would require incredible intelligence that would recognize every street with ease.

[Richard Masters]

But how will it know where it is going? That, my friend, requires artificial intelligence, something that is in the baby stages

The same way that Google Maps knows where you are going. You give it a destination.

[BobG]

Let's say you have self-driving car.

Now the first person dies in an accident involving one.

And the lawyer proved to 12 morons, good and true, that it was due to a programming error.

Where do you think the tiny shards of the entire industry are going to be stored after the award?

Why should the consequences of a programming error be any different from those of the current errors manufacturers are making and being sued?

Bob

[Richard Masters]

There is a difference between being realistic and living in a fantasy world. For a self-driving car to work, it would require incredible intelligence that would recognize every street with ease.

I don't recognize most streets with ease (because I haven't seen them before), yet I invariably reach my destination.

[makaya325]

I don't recognize most streets with ease (because I haven't seen them before), yet I invariably reach my destination.

I can see the creation of cars that drive themselves with the help of their being rails on the roads (Magnetic levitation)

[Roboramma]

But how will it know where it is going? That, my friend, requires artificial intelligence, something that is in the baby stages

How is it hard for it understand: "take me to 404 18th street"?
It will find the place that you tell it to take you to the same way google maps does: it will give you the most likely matches, and you can confirm. It doesn't need to do anything more than a taxi driver does (ie. sometimes it may not be able to locate your destination, and you will have to plot it manually, which isn't a big deal).

[makaya325]

How is it hard for it understand: "take me to 404 18th street"?
It will find the place that you tell it to take you to the same way google maps does: it will give you the most likely matches, and you can confirm. It doesn't need to do anything more than a taxi driver does (ie. sometimes it may not be able to locate your destination, and you will have to plot it manually, which isn't a big deal).

That would be one expensive car

[Zax63]

The answer likely is: No

I could see the development of self-driving cars by the use of magnetic rails, however, If we wanted to design self driving cars, it would require some kind of supercomputer memory regarding routes, roads, turnpikes, etc.

Take a look at this video about the winner of the DARPA Urban Challenge.
lULl63ERek0
An actual self-driving car and they say it uses 10 Core 2 Duo computers, a lot of power but hardly in the super computer range.

How would a self-driving car recognize what road it is on? Would it have a built in memory to recognize the physical world?
There are tons of GPS units that do this now for less than $200. The last GPS I purchased had all of the map data for the U.S. on 1 4.7GB DVD. No problem at all in an era of cheap 8GB flash drives, or regular hard drives with 1,000GB or more. Add in a 3G Internet connection for updates and current traffic, weather and road conditions and you are all set.

But how will it know where it is going? That, my friend, requires artificial intelligence, something that is in the baby stages
Again, cheap GPS units do this right now no AI required. Roads are a network. Network routing is pretty well solved or we wouldn't have the Internet.

There is a difference between being realistic and living in a fantasy world. For a self-driving car to work, it would require incredible intelligence that would recognize every street with ease.
It's not only realistic, it's reality. That car in that video was programmed in 2007. They claim they built it in less than 1.5 years. I can only imagine where state of the art is today. They may not be ready for general use but self driving cars exist right now.

That would be one expensive car
Here I would agree with you. I wouldn't be surprised at a $10,000 - $20,000 premium over a similar normal car when first offered for sale. 5-10 years after that it may be standard equipment on all cars.

[makaya325]

The GPS that would be used would have to come installed in the car, adding another 2000-3000 bucks.

[arthwollipot]

The answer likely is: No

I could see the development of self-driving cars by the use of magnetic rails, however, If we wanted to design self driving cars, it would require some kind of supercomputer memory regarding routes, roads, turnpikes, etc.http://maps.google.com/

[BobG]

The GPS that would be used would have to come installed in the car, adding another 2000-3000 bucks.

So!

[arthwollipot]

The GPS that would be used would have to come installed in the car, adding another 2000-3000 bucks.$500 for the installed GPS unit that could have come with my 2006 Toyota Prius, but wasn't.

[ddt]

And as to systems that actually already operate on the road: here's a (Dutch) newspaper clipping (http://tweakers.net/nieuws/55071/onbemande-bussen-terug-de-weg-op.html) from August last year. It's about an unmanned bus that is operated by the public transport company in Rotterdam. The service was first instated in 2005, but taken off the road again within a week; and last year it was reinstated.

This shuttlebus makes use of various sensors that are placed in the road, so it can see where it drives (instead of using GPS), and it drives on a separate lane.

[SK.]

I don't recognize most streets with ease (because I haven't seen them before), yet I invariably reach my destination.
The car will keep track of it´s position by wheel odometry ("counting wheel revolutions"), visual odometry, possibly laser scans and (low cost) IMU data. Even in absence of an absolute position update (like GPS) it will be able to estimate it´s own position for quite some time from this relative information (even if errors will add up over time). When it has a decent map it will eben be able to localize itself without GPS, by matching what it has observed with features in the map ( e.g. characteristic houses, traffic lights etc. ).

But that´s only the worst case scenario, most of the time you have GPS available. Personal navigation assistants work today, and as long as GPS is available the problem is basically solved. Estimating own position and computing a route to a target position can be done on today´s cell phones.

Global large scale path planning and position estimation isn´t as much a problem as reliable short term cognition and decision making is. After all, sounds great to have a vehicle that´s 99.9% reliable. Doesn´t sound so great if you let it drive 99.9m and then it crashes against a wall.

[MRC_Hans]

I'd guess that we might first see self guided vehicles in limited areas such as the US Interstate Highway System. No traffic signals or stop signs. Commercial trucks would seem to be likely first adopters of a system that included onboard intelligence and sensors, and an intelligent and transducer equipped road.

Imagine a dozen or more trucks running in a "trucks only" lane at constant speed with only inches between bumpers, in continuous contact to maintain spacing, across the US plains. Braking reaction time would be virtually instantaneous, and sensors in the road could analyze traffic for miles so vehicles could anticipate other traffic. It could even schedule pull-offs to allow faster vehicles to move ahead of slower ones and accomodate human drivers. A huge benefit of this would be that automobile traffic flow would be much smoother without the commercial truck traffic in their lanes.

Better still, couple them together, and let one larger and more effctive engine pull them all, then make a steel road surface and steel wheels to reduce friction, and ..... Wait! Hmmmm, we already have that.

Hans

[Puppycow]

That'll be great for new roads, but surely a hideous expense to go and retro-fit to old ones.

I don't know about that. Roads have to be resurfaced from time to time anyway. I'm pretty sure that resurfacing an existing road is much less expensive than building a new one from scratch.

[SK.]

I don't know about that. Roads have to be resurfaced from time to time anyway. I'm pretty sure that resurfacing an existing road is much less expensive than building a new one from scratch.
It´s actually an interesting question how to achieve autonomous capability. There´s a wide spectrum of possibilities from completely autonomous cars doing all cognition and planning themselves (possibly mesh-networked to other cars in the vicinity) to "dumb" cars, that only receive motion commands from the road that has all the sensors built in. And then there´s the inbetweens, like RFID tags in the road that get sensed by the car.
Recently, research seems to go more into the direction of completely autonomous cars, while we´ve seen more rigged road experiments in the 90s. Or at least that´s my perception, but I might be a little biased towards autonomous systems :)

[arthwollipot]

Phalanxing. I like the idea of phalanxing.

[Puppycow]

But how will it know where it is going? That, my friend, requires artificial intelligence, something that is in the baby stages

It just requires the correct algorithm, resources, sensors and instructions. It doesn't have to decide where to go. The person decides where to go and the vehicle goes there.

[Puppycow]

The GPS that would be used would have to come installed in the car, adding another 2000-3000 bucks.

No big deal. Technology gets cheaper every year. And consider that people pay drivers (chauffers) a lot more than that in a year. So even at that price a lot of people would find it to be worth it.

[DC]

I read a science fiction story about a car that could drive without a driver. It would not allow drunks to drive themselves and would also drive to and from places with no driver at all. Could this happen anytime soon?

i saw a Docu from the USA, where they altered a part of a highway and made cars able to drive themself, i think at 160 mph with only 1 meter between the cars.

and meanwhile we are already pretty far with video cameras and computer "recognizing" the traffic and such things,

I also saw a application for trucks to follow eachother automaticly.

i think, soon we can.

i more ask , can we make a self driving car with less polution? :D

[Puppycow]

Phalanxing. I like the idea of phalanxing.

You mean convoys?

[skbuncks]

Whilst not exactly urban driving, no autonomous car thread would be complete without this:

http://videos.streetfire.net/video/126-Top-Gear-Self-Driving_180380.htm

skb

[arthwollipot]

You mean convoys?Basically, yeah. They call it phalanxing. Multiple autonomous vehicles on the same stretch of road will seek each other out, connect wirelessly to each other, and move as a unit.

[GlennB]

There is a difference between being realistic and living in a fantasy world. For a self-driving car to work, it would require incredible intelligence that would recognize every street with ease.

Agreed. And the flexibility to cope with every eventuality is not something that can be programmed. One example - it's illegal to drive through a red light, and it's illegal to drive up onto the pavement (sidewalk), and robo-car would of course be programmed to avoid both. Yet doing either (or both) of these things could well be the correct response when an ambulance or fire engine is howling just behind you and everybody else has cleared a way to let it through. So we would absolutely need a rapidly available manual override, yet one that could not accidentally be triggered by a 'driver' who would then be out of control of the vehicle.

And ... has anybody here experienced faulty car electronics? It happens. The first time it causes a serious accident or death then the robo-cars will all be switched to manual control. Permanently, I suspect.

And ... the car would need to be able to read and understand written roadside instructions, and hand signals from police and traffic people. "Police check. Please pull in if requested", "Road ahead flooded" and so on and so on.

But the killer blow would be that absolutely nobody would feel comfortable in such a car. In fact we'd all be pissing our pants, the whole time.

[ddt]

Whilst not exactly urban driving, no autonomous car thread would be complete without this:

http://videos.streetfire.net/video/126-Top-Gear-Self-Driving_180380.htm

skb

Jeremy Clarkson has psychic powers. Everybody knows that!

[BobG]

$500 for the installed GPS unit that could have come with my 2006 Toyota Prius, but wasn't.

I took it for granted that a car that drives itself is going to be much more expensive so it's immaterial to be talking about the cost of a GPS.

But, even though I expect this to happen someday, the cost would be stupendous. Not just the cost of the car but the cost of the roads.

I don't know why but the thought of bumper cars at an amusement park just came into my mind.:)

Bob

[Horatius]

But the killer blow would be that absolutely nobody would feel comfortable in such a car. In fact we'd all be pissing our pants, the whole time.



Old people like us probably won't feel overly comfortable, but people who grow up with this technology will just take it for granted.

I imagine that, after a few years, it will be no worse than letting someone else drive. I hate that too, but I can do it.

[richardm]

But how will it know where it is going? That, my friend, requires artificial intelligence, something that is in the baby stages

Figuring out how to get from A to B is a pretty trivial exercise - have you never come across in-car satellite navigation systems?

Or am I missing your point?

And this:

The GPS that would be used would have to come installed in the car, adding another 2000-3000 bucks.

Lots of cars have GPS installed already - they cost nothing like that much. I feel sure I'm missing some subtlety in your argument.

[Safe-Keeper]

I read an article in a popular science mag not too long ago about the testing of an "autopilot" system for trucks. The trucks would form a coloumn on a test course, with the leading truck piloted by a human, then the drivers of the trucks following it switched the "autopilot" on and each vehicle followed the one in front of it, braking when necessary. The article spoke of how the reaction time of a human being are a second of two, while a machine can react nearly instantly. It was amazing. ETA: I guess this is the "phalanxing" referred to above?

Figuring out how to get from A to B is a pretty trivial exercise - have you never come across in-car satellite navigation systems?Exactly. Open Google Earth, pick two locations, and tell the program to find a route between them. On a computer of average strength, you'll have your result in a couple of seconds.

There are some glitches in the software, such as its inherent optimism when telling you to drive your city car through pedestrian-only roads and forest trails and whatnot, but getting around this surely can't be more difficult than assigning the roads in the system a "type of road" variable (0=Don't drive here, 1=Poor condition, etc.).

Nothing sci-fi about pathfinding, far as I can tell. I once rode a car with a computer telling us where to go, and upon receiving its orders its very first instruction to the driver was "Please back out of the driveway".

But the killer blow would be that absolutely nobody would feel comfortable in such a car. In fact we'd all be pissing our pants, the whole time. Hundred years ago, people said the same thing about hypothetical cars travelling at 100 km/h;). Sure, those of us who haven't grown up with the technology will have to take some time to get used to it, even the younger ones among us, but if/when it's proven to be safe I'm sure people will adapt. Unless an "organic" crowd forms and demands we stop using the technology, the same way they're demanding we go back to the developing country stage when producing foods and medicine.

My uneducated two cents.

[SK.]

But, even though I expect this to happen someday, the cost would be stupendous. Not just the cost of the car but the cost of the roads.
Well depends on which approach will win out. The IMHO most elegant and most flexible approach will only use onboard sensors, so roads wouldn´t have to be modified.
Of course, the necessary sensors aren´t there yet. Today´s fancy laserscanners and other sensors mounted on Urban Challenge class autonomous cars amount to some >200.000$ per car. Looking at the advances in MEMS sensors, digital cameras, computing power etc in the last 10 years, it isn´t much of stretch to expect workable and affordable stuff in another 10 years.

[Puppycow]

And ... has anybody here experienced faulty car electronics? It happens. The first time it causes a serious accident or death then the robo-cars will all be switched to manual control. Permanently, I suspect.

And ... the car would need to be able to read and understand written roadside instructions, and hand signals from police and traffic people. "Police check. Please pull in if requested", "Road ahead flooded" and so on and so on.

But the killer blow would be that absolutely nobody would feel comfortable in such a car. In fact we'd all be pissing our pants, the whole time.

Only at first. Once we get used to risks we discount them. How many people die in traffic accidents currently? Yet people are not afraid to get in cars. Unless the rate of accidents is higher, it will still make at least as much sense to let the car drive, and if the rate is lower, it makes even more sense. It doesn't have to be absolutely risk-free to be a safety improvement over the status quo. So yes, at first people will illogically overreact to the risk, but eventually they will get used to it and stop pissing their pants.

[Safe-Keeper]

Also keep in mind that modern-day airliners are already 90% autopilot-driven. Far as I know, they even handle landing with autopilot now.

[Christian Klippel]

The GPS that would be used would have to come installed in the car, adding another 2000-3000 bucks.

That's total nonsense. You can buy GPS navigators for a few hundred bucks. Not only do they have all the maps installed that you need, they also have a radio receiver built in that receives a TMC signal. With that it "detects" traffic jams, blocked roads, etc, and recalculates the route to go around them.

It even knows gas stations, supermarkets, etc. The technology is already there, widely used and working. Add to this that the technology is quite cheap nowdays, and gets cheaper every day.

No fancy AI or hardware in the 2k+ bucks range needed. Really, please educate yourself a little bit on the topic at hand before you start posting such obvious nonsense.

The only problem that i can see with autonomous cars is the question of liability. Let's assume an accident happens. Who is responsible? The driver, because he didn't interfere with the car's actions to avoid the accident? But he's supposed to "trust" the system, isn't he? Or the company who wrote the software? Or a single programmer in that company? But they are supposed to "trust" the hardware sensors and the driver, right? The manufacturers of sensors and stuff which might have failed and thus caused the accident? But there ought to be a "fail-safe mode" in the software for such thing, right? Or the car company, for building and selling the car without having it tested in every imaginable and unimaginable circumstance? But they should have trusted the quality of the used components...

But i'm sure that there will be a solution to that question once the time comes.

Greetings,

Chris

[Ziggurat]

The only problem that i can see with autonomous cars is the question of liability.

That's the only problem you can see? Well, I can see quite a few more than that. Starting with obstacle recognition and avoidance. You're right that GPS is cheap, but GPS doesn't cut it. GPS can't tell me if there's a stalled car in front of me. Or a dear, or a fallen tree, or a box that fell off a truck. Dealing with that stuff takes a hell of a lot more than GPS, it takes fancy detectors of some sort, along with very high power AI. And yes, that costs a hell of a lot more money than GPS, and no, we can't do it very well, or even very reliably, yet.

[BobG]

Also keep in mind that modern-day airliners are already 90% autopilot-driven. Far as I know, they even handle landing with autopilot now.

Correct and even on aircraft carriers!

[Christian Klippel]

That's the only problem you can see? Well, I can see quite a few more than that. Starting with obstacle recognition and avoidance. You're right that GPS is cheap, but GPS doesn't cut it. GPS can't tell me if there's a stalled car in front of me. Or a dear, or a fallen tree, or a box that fell off a truck. Dealing with that stuff takes a hell of a lot more than GPS, it takes fancy detectors of some sort, along with very high power AI. And yes, that costs a hell of a lot more money than GPS, and no, we can't do it very well, or even very reliably, yet.

Well, assuming that such a car exists and works, yes, that's the only real problem that i can see. Indeed i make the assumption here that such a system is available and working, which is not the case at the present time.

However, i think that if such cars come available, that by this time the whole infrastructure has also changed. The problem of a stalling car can be solved by low range RF systems. Then the stalling car would simply signal the other cars that it is stalling. Think of streets and cars as parts of a big network.

We already have cameras on the streets that capture the traffic. Take it a step further and have a computer system that evaluates these images (which is already done partially), and broadcasts the results to all cars in that street segment. That could solve the problem of animals or other unwanted objects on the street. No fancy sensors or stuff in the car required to do the hard work. By the time the car approaches the spot where the animal is on the street, it would already be "aware", preferably slowed down it's speed, and would use rather simple object detection to either stop or drive around the obstacle.

While it is more complicated to detect what is actually on the street (object recognition), it is rather simple to detect that something is there that does not belong to the street. There are already systems that work well which can drive a car on a street safely, simply by detecting and evaluating the markings that you have on streets nowdays. A modified system of that is already used in trucks to warn the driver if he/she starts to steer outside of the "safe area" of the lane.

I think it is just wrong to assume that such a car would come into existence, but everything else around it won't change. Distributed systems are the way to go here, simply to offload big, complicated tasks from the single cars. No need to have multiple cars do the same detection on their own, for their own use, if that can be done in a external system and made available to all cars in the street segment in question.

The street's and surroundings have to become active parts of the system, not just isolated cars that try to steer their way through the traffic. And as it is, streets are already become "active". There are lots of sensors used in streets nowdays, to detect and evaluate the traffic, avoid traffic jam's, control the signs accordingly, monitor for accidents, etc. Sure, not everywhere, but it's going into that direction already.

Greetings,

Chris

[temporalillusion]

Think of streets and cars as parts of a big network.

That's what I had in mind as well.. each car isn't exactly autonomous, there could/would be an overall "car-traffic control" in communication with all robot cars all the time, constantly being updated with information being gathered by every car on the road, and constantly updating all cars.

Then it could analyze things and give instructions to cars like slow down because of an accident ahead that isn't visible to the car yet, optimize traffic flow, even those scary scenes in movies where automatic cars whiz through an intersection at full speed with no lights because they are controlled by a single overseer.

[Ziggurat]

No fancy sensors or stuff in the car required to do the hard work.

We have more road than we have cars. Offloading the problem to roads is moving in the wrong direction.

While it is more complicated to detect what is actually on the street (object recognition), it is rather simple to detect that something is there that does not belong to the street.

No, actually, it is not simple. How do you distinguish between a shadow and an object, for example? It's a very difficult problem to recognize obstacles accurately.

There are already systems that work well which can drive a car on a street safely, simply by detecting and evaluating the markings that you have on streets nowdays.

That is not the hard part. W do not know how to do the hard part (obstacle recognition and avoidance) well, and we can only do it poorly at great expense.

[Darth Rotor]

GPS for geographical location and Google Maps for identifying the current road.
In that case, all roads lead to Porn.



For the OP, what is the purpose of this device? I saw one answer upthread, in driving blind people where they need to go.

Any other purpose?

DR

[Jason Smith]

I guess I'm going to have to be the only one to answer the OP with a guarded no.

I'm at work so I don't have the book handy, but there's a fascinating book just released on paperback called Traffic:Why We Drive the Way We Do (and What It Says About Us) (http://www.amazon.com/Traffic-Drive-What-Says-About/dp/0307264785/ref=ed_oe_h) by Tom Vanderbilt. In the first chapter or two he talks about why robot cars are decades away from being practical. There's simply a lot of cognitive tasks going on as we drive (research estimates from 1500 - 2500). I seem to recall that even the best driving systems can't distinguish between a pedestrian standing at a bus stop versus one walking towards the road. Imagine the fun that would cause in traffic. Again, I can't go into too many details, but he glosses over his thought process in the link I provided.

If you're at all interested in driving, you should definitely check this book out. Go ahead, find out why you're a far worse driver than you think you are. :)

Jason

[GlennB]

I guess I'm going to have to be the only one to answer the OP with a guarded no.
....
There's simply a lot of cognitive tasks going on as we drive (research estimates from 1500 - 2500). I seem to recall that even the best driving systems can't distinguish between a pedestrian standing at a bus stop versus one walking towards the road....

Personally I'd answer the OP with a resounding 'no'.

Yet another example -- I believe that in the UK drivers are recommended not to risk an accident with emergency braking if a cat or dog runs in front of the car. Let's suppose this is true. Then the robo-car will need to be able to distinguish between a dog and a small child taking a tumble in the road. With 'sensors' covered in road grime after driving through wet and muddy conditions?

Likely? Ever? Nope, unless every human can be guaranteed to be carrying a transmitter that the car can register. And I haven't seen an answer to the point about those occasions when it's correct to drive through a red traffic light, or drive onto the pavement/sidewalk.

Totally auto-piloting an aircraft is a complete piece of cake in comparison to this robo-car's requirements.

I'm afraid that there's an awful lot of fantasy thinking going on in this thread.

[Earthborn]

In the first chapter or two he talks about why robot cars are decades away from being practical.Mere decades, eh? I'd say he's quite an optimist. I don't see how that supports your "no" answer to the OP. It's a pity though that we'll probably have to wait that long, because with all the phones and car entertainment systems it is quite obvious that there are lots of people out there who would rather do other things in their car besides driving.

I seem to recall that even the best driving systems can't distinguish between a pedestrian standing at a bus stop versus one walking towards the road.I don't think many of the systems experimented today even bother to distinguish between human beings and other obstacles. Programming them to avoid bumping into things seems difficult enough.

[Christian Klippel]

We have more road than we have cars. Offloading the problem to roads is moving in the wrong direction.

How so? Intelligent traffic control systems are already in use. Here in Germany, on a few select highway segments, we have systems that employ cameras and a sophisticated image recognition to detect what's going on. Why not tap into that already existing resource, but instead increase the workload of a car's system to basically do the same thing? Not to mention that a car can only evaluate it's close surroundings, and thus would have to react and process much faster. Using already available information as well would give it a "look ahead" information.

No, actually, it is not simple. How do you distinguish between a shadow and an object, for example? It's a very difficult problem to recognize obstacles accurately.

Shadows don't produce a radar signature, for example. Shadows don't interrupt the line-of-sight of laser or IR beams, for another example. Such systems could be installed at the side of the roads to continuously monitor what's going on. Combined with cameras and image recognition that would be a very workable solution.

Also note that i said it is simpler to detect any random obstacle (something is there or not) than it is to detect what the obstacle actually is (human, box, tree, ...). Not that is very easy in general. "Obstacle recognition" gives me almost 2.5 million hits in Google. Many of them are related to traffic/vehicles. "Intelligent traffic control" yields about 1.3 million hits, another 3.3 million for "automated traffic control". "Lane tracking" is over 6 million hits, albeit not all of them are about streets.

So, these systems are either already in use, evaluated/tested or being developed. Not using the information gathered by those systems, by networking it and making it available to all the hypothetical autonomous cars, would be a huge waste of resources and would make things much more complicated in said car.

That is not the hard part. W do not know how to do the hard part (obstacle recognition and avoidance) well, and we can only do it poorly at great expense.

Part of the problem surely is that for the most part we use isolated systems that only work on and for their own in a passive environment. Having many systems communicate with each other, building a network, and making the environment active would surely make it simpler and more efficient. Add to that the fact that computer power is increasing, so what is time consuming today would be quite fast tomorrow.

Greetings,

Chris

[Christian Klippel]

I don't think many of the systems experimented today even bother to distinguish between human beings and other obstacles. Programming them to avoid bumping into things seems difficult enough.

Just a quick Googling. Here (http://library.witpress.com/pages/PaperInfo.asp?PaperID=19566) is one abstract, there (http://dli.iiit.ac.in/ijcai/IJCAI-05/PDF/post-0001.pdf) is a complete PDF of another paper. And i only looked at the first two pages of results for this (http://www.google.de/search?hl=de&q=obstacle+recognition&btnG=Google-Suche&meta=&aq=f&oq=) search.

Sure, not perfect systems, still a lot of work to do. But it's going into that direction, and we are not at the very beginning anymore.

Greetings,

Chris

[Ziggurat]

How so? Intelligent traffic control systems are already in use. Here in Germany, on a few select highway segments

A few select highways. But if you want to replace drivers, a few select highways isn't enough. It's got to be ubiquitous. Furthermore, Germany is a lot more densely populated than many places.

Not to mention that a car can only evaluate it's close surroundings, and thus would have to react and process much faster.

You need fast reaction capability anyways, because obstacles don't have to be static.

Shadows don't produce a radar signature, for example. Shadows don't interrupt the line-of-sight of laser or IR beams, for another example.

In other words, you need fancier (and more expensive) sensors than simple cameras.

Such systems could be installed at the side of the roads to continuously monitor what's going on.

If you've ever driven across the US, I think you would understand the futility of that enterprise.

Also note that i said it is simpler to detect any random obstacle (something is there or not) than it is to detect what the obstacle actually is (human, box, tree, ...).

But that's not really good enough either. Let's say a small tree falls in the road, and you can't avoid it. What's better to hit, the trunk or the bushy part? Well, you better be able to interpret what that object is, or you'll make the wrong call. The note about someone standing by the side of the road is another good example. Or how about a ball bouncing towards the road: do I need to swerve to avoid it even though it's not in the road yet? What about the kid chasing the ball? Will he catch the ball before it goes into the road, or will I have to worry about him running into the road? If all I'm doing is recognizing that objects exist, without any recognition of what those objects are, the system won't be able to perform nearly as well as humans.

Part of the problem surely is that for the most part we use isolated systems that only work on and for their own in a passive environment.

If you want the system to be robust, it had better operate that way. It's one thing to include external input to enhance performance, but you had damned well better have the system functioning on a per-car basis. Otherwise you're opening yourself up to catastrophic failure. Plus, of course, deployment is a HELL of a lot easier when it can be done incrementally, which is much easier to do if the cars don't need a full-blown netwrork of external sensors.

[GlennB]

...
You need fast reaction capability anyways, because obstacles don't have
to be static.
....


Everything you said in your post. Then there's the question of knowledge. I know that, after heavy rain, water will gush down the hill and across the road on a certain bend. I know that if I brake hard in that bend I'm quite likely to aquaplane and shoot onto the wrong side of the road. I know that horses are walked on the road in that area. So when it's wet I take that bend very carefully indeed. I cannot, for the life of me, see a computerised driving system 'knowing' all this stuff.

[Christian Klippel]

A few select highways. But if you want to replace drivers, a few select highways isn't enough. It's got to be ubiquitous. Furthermore, Germany is a lot more densely populated than many places.

Even worse, only segments of these highways. But my point was that such systems exist today, and get installed today. And where they exist their data should be re-used.

Yes, of course such systems have to be ubiquitous at one point. But one has to start somewhere, right? I mean, traffic lights didn't pop out of the streets everywhere at the same time too. They were installed at a few places first, then gradually the system of traffic lights grew into what we have today.

You need fast reaction capability anyways, because obstacles don't have to be static.

In other words, you need fancier (and more expensive) sensors than simple cameras.

Well, i think that of course the car's system has to have a fast reaction time. But there is a difference between a highway and a street in the city. The faster you drive, the faster reactions you need. At places where fast reaction is required, it could make use of external systems. If these are not available, then it simply slows down or the driver has to do the job.

If you've ever driven across the US, I think you would understand the futility of that enterprise.

See my above comment about traffic lights.

But that's not really good enough either. Let's say a small tree falls in the road, and you can't avoid it. What's better to hit, the trunk or the bushy part? Well, you better be able to interpret what that object is, or you'll make the wrong call.

True. However, very sudden appearances of objects on the road lead to accidents most of the times anyways. If a deer jumps on the street in front of your car, you are very likely to hit it anyways. Humans don't have that fast a reaction time as well. There is always a worst-case situation in which even the most sophisticated electronics can't help you anymore. Cars simply don't stop in an instant when you hit the brakes. There's the mass of the vehicle and the momentum it has and all that.

In case of your tree example, if it appears on the street that suddenly that you have to hit it anyways, i really doubt that you have much time or ways to select which part of to hit anyways.

The note about someone standing by the side of the road is another good example. Or how about a ball bouncing towards the road: do I need to swerve to avoid it even though it's not in the road yet? What about the kid chasing the ball? Will he catch the ball before it goes into the road, or will I have to worry about him running into the road?

As learned in drivers school, if a ball appears on the street, slow down and watch out carefully, because most of the time a kid is following the ball. If someone is standing at the side of the road, also slow down and be careful. An autonomous car should surely do the same. Being slower gives it more time to process the incoming information and thus do a full analysis of the object's moving towards/onto the street.

And again, the example with someone standing at the side of the road is another case where an external, networked system could help.


If you want the system to be robust, it had better operate that way. It's one thing to include external input to enhance performance, but you had damned well better have the system functioning on a per-car basis. Otherwise you're opening yourself up to catastrophic failure. Plus, of course, deployment is a HELL of a lot easier when it can be done incrementally, which is much easier to do if the cars don't need a full-blown netwrork of external sensors.

My guess is that such cars and systems would be "hybrid" anyways in the beginning. Start with automated driving on highways. Much easier to do than driving in a city. Highways can be better "controlled", overseen and protected from animals or people entering the lanes. Having a steady flow of traffic on highways, due to cars being autonomous there, will already save the people a lot of time. Start with heavily used highways/segments first, then extend to the remaining parts. After that start in the cities.

Also keep in mind that the more complex you make a single system (in this case the car), the more potential points of failure you have. So, packing all into the car is some kind of draw-off as well. However, there should be no problem to have the car work on it's own, but perform better if it "finds" external support in the form of sensors, network, etc. The smallest common denominator is the driver actually doing the driving on its own. It then can drive in a "slower safe mode" autonomously. Or it can go faster when it has the suitable environment.

I know that it will take a long time to reach the point of even having automated highway traffic with autonomous cars. I don't expect that stuff to happen in my lifetime either. But systems go into that direction today. Nowdays it's just driver assistance to stay in the lane, hit the brakes early enough if the car in front of you stops, etc. It will evolve from there.

However, and that gets me back to my original point, assuming that such cars and systems exist, who is liable in case of an accident? That is something that, in my opinion, can not be really solved. All the automation is, after all, meant to make the driver more and more simply a passenger. If there is a failure that leads to an accident, who to blame? The manufacturers surely will say that the driver has to be careful all the time anyways. The driver will say "if i have to do that, why have all that crap supposedly doing my job in the first place?"

If the driver can not be held responsible, who to blame then? The manufacturer of the car? Or someone who implemented some part of the very complex overall system? The manufacturer will say that he has to trust the systems he bought and built into the car. The ones who made these systems will say that the car manufacturer is the one who made it finally work on it's own. Etc, etc...

It wouldn't be the first time that something usable doesn't really get onto the market broadly because of unresolved liability issues in case something goes wrong.

The current system is easy: You drive, you are the first one to be responsible. You may then prove that the other party is guilty, however, you get the blame first.

Greetings,

Chris

[makaya325]

Correct and even on aircraft carriers!

True, but everyone will not be able to afford these hypothetical cars. What is used on planes and ships will not always be used on cars

[BobG]

So when it's wet I take that bend very carefully indeed. I cannot, for the life of me, see a computerised driving system 'knowing' all this stuff.

This already exists in braking systems. You can jam on the brakes as hard as you like and you will not skid. Computerized braking enables a car to stop in the shortest time/distance possible. Humans cannot do better.

As far as going around a curve on a wet road, I believe that cars have computer systems for this also. Maybe someone else could elaborate.

[BobG]

True, but everyone will not be able to afford these hypothetical cars. What is used on planes and ships will not always be used on cars

I presume that this will result in less accidents. Think of the savings there.

[Christian Klippel]

Everything you said in your post. Then there's the question of knowledge. I know that, after heavy rain, water will gush down the hill and across the road on a certain bend. I know that if I brake hard in that bend I'm quite likely to aquaplane and shoot onto the wrong side of the road. I know that horses are walked on the road in that area. So when it's wet I take that bend very carefully indeed. I cannot, for the life of me, see a computerised driving system 'knowing' all this stuff.

To put it simple: It doesn't need to know these details at all. The only thing it really needs to know that to drive slower when the street is wet, i.e. when it is raining, and even slower when it approaches a bend/curve in the road. All the other, additional details could be given by an external system. That way it could go faster at places where it is not so critical (i.e. a straight, even road) or slower at more critical places (straight, even road but where horses could appear).

It's similar to your or my computer on the internet. It doesn't need to know what IP the site www.randi.org has. It only needs to know where to get that information: the DNS. And even that one doesn't need to know, it can ask the root nameserver. So, such a car could constantly "ask" the external systems about the current situation to have a more complete "picture" of what is going on, in addition to it's on board systems.

Greetings,

Chris

[Christian Klippel]

This already exists in braking systems. You can jam on the brakes as hard as you like and you will not skid. Computerized braking enables a car to stop in the shortest time/distance possible. Humans cannot do better.

Right. That's the Anti-lock braking system (http://en.wikipedia.org/wiki/Anti-lock_braking_system).

As far as going around a curve on a wet road, I believe that cars have computer systems for this also. Maybe someone else could elaborate.

That would be the electronic stability control (http://en.wikipedia.org/wiki/Electronic_stability_control).

Btw, here (http://en.wikipedia.org/wiki/Category:Automotive_safety_technologies) is a overview on Wikipedia about the car safety systems that exist today. Probably not everything is in that list, and surely not stuff that is developed right now or will be available soon. However, it's already a nice list of what is available now.

Greetings,

Chris

[Jason Smith]

Mere decades, eh? I'd say he's quite an optimist. I don't see how that supports your "no" answer to the OP.

My reply was a guarded no because I want to be optimistic. I want to be optimistic because ... well ... robo-cars would be kind of cool. I reserve the Zeus given right to optimistic. :D

Of course, if we're being optimistic about future technology, I want my robot butler, immortality serum, and side-effect-free hair loss cure. <sigh>

Your comment about driving distractions is the subject of at least one chapter in the book. I resolved to limit my use of distractions while driving after reading the book. I'm still working on that.

Jason

[SK.]

I can strongly recommend Sebastian Thrun´s recent lecture about the topic from the recent "25 years of field robotics at CMU" symposium. He´s one of the leading figures in probabilistic robotics as well as research into autonomous driving. There´s quite some interesting points he makes, about many cars just sitting around nearly 100% of the time of their existence, about the possible savings in energy, road construction, money, etc.
And obviously, also about the many problems that still have to be solved till this vision can become a reality.
Seeing how I can´t post URLs yet, just google "robotics cars: are we done yet?"
You´ll find a ~45 minute IIRC video.

[Jungle Jim]

The future will get here eventually. In my previous post I suggested that in 50-100 years that cars that would not be driven, they would be programmed. Maybe, 50 years is too soon. We need to step back and look at the changes that can occur over a 100 year period.
Here is photo of a 1909 Model R Ford: http://shelfunk.files.wordpress.com/2009/04/1909-ford-model-r.jpg


Here is a photo of a 2009 Ford Fusion: http://batonrougefordfusion.com/wp-content/uploads/2009/03/2009-ford-fusion.jpg


A hundred years from now you will not recognize, nor be able to find your way around the city or town you live in (e.g., the house you grew up in won’t be there). The changes that will occur in transportation, architecture, urban planning, technology are virtually certain to make 2109 a very different world. As has been pointed out in this thread, the technology for a self driving car already exists (albeit in its early stages). Moreover, in 100 years people will not even own cars—they will be museum pieces. Personal transportation (aside from mass transit) will be in the form of essentially programmed taxis. Here are a couple of sites to fuel your imagination: http://faculty.washington.edu/jbs/itrans/reynoldsfuturist.htm; http://library.thinkquest.org/04oct/01249/

[makaya325]

That's total nonsense. You can buy GPS navigators for a few hundred bucks. Not only do they have all the maps installed that you need, they also have a radio receiver built in that receives a TMC signal. With that it "detects" traffic jams, blocked roads, etc, and recalculates the route to go around them.

What about different levels of GPS for different car brands? Will a Hyundai GPS be as accurate as a Mercedes GPS? I guess what i am trying to say in relation to this topic is: Different strokes for different cars:p

[Christian Klippel]

What about different levels of GPS for different car brands? Will a Hyundai GPS be as accurate as a Mercedes GPS? I guess what i am trying to say in relation to this topic is: Different strokes for different cars:p

Really, mak, you should stop making stupid post in whatever threads you want to participate. Posting something just for the sake of posting doesn't help you at all. Quite the opposite: you are working towards everyone finally giving up any shred of hope in you that might be left.

Go and educate yourself about GPS. Get a grasp on how GPS receivers are built nowdays. Your keywords here are "gps chipset" and "cooking with water". Use the former in Google, get a grasp on what the latter means (hint: it's a phrase). Now try to combine what you hopefully just learned by that and re-read what nonsense you have posted.

Counter-check that newly gained knowledge by looking for technical specs of GPS receivers and navigation system that use them. Again, your keywords are "gps chipset" and the "cooking with water" phrase.

If you still can't grasp it, simply try to refuse from posting about things you have no clue about.

Greetings,

Chris

[GlennB]

To put it simple: It doesn't need to know these details at all. The only thing it really needs to know that to drive slower when the street is wet, i.e. when it is raining, and even slower when it approaches a bend/curve in the road. All the other, additional details could be given by an external system. That way it could go faster at places where it is not so critical (i.e. a straight, even road) or slower at more critical places (straight, even road but where horses could appear).


Will it be able to read the hand-painted sign that says "Caution - horses on road" ? Or will every hazard - permanent or temporary - need electronic flagging?
Will it be able to tell the difference (as mentioned earlier) between a dog and a child who has fallen over? Or will every person carry a transmitter?

The cost in infrastructure to guide such robo-cars will be monstrous and utterly inappropriate in a world that should be aiming to tighten its belt rather than indulging its techno-addiction in this manner. And to what end? What will the significant benefit be?

I clearly recall articles in the 1970's confidently predicting an end to traffic jams when flying cars became freely available in , er, about 9 years ago in fact.

[Richard Masters]

Will it be able to read the hand-painted sign that says "Caution - horses on road" ? Or will every hazard - permanent or temporary - need electronic flagging?
Will it be able to tell the difference (as mentioned earlier) between a dog and a child who has fallen over? Or will every person carry a transmitter?

There probably isn't a silver bullet for each worst case scenario, but an easy solution is to designate certain highways as "hands-free highways" (HFH), such that there are one or two lanes dedicated to "robot-cars".

A car near an HFH will alert the driver of his/her piloting options, or alternatively, when selecting a travel route, it can try to find the most hands-free route.

If there has been an accident, local authorities should and will know, and they can broadcast that information via a simple text messages through local cell-phone towers. A car receiving the message will slow-down appropriately, or request that the driver take over. In the case that the driver isn't available (is asleep), the car can pull over until the highway status returns to normal.

On such a highway, animal crossings will be rare, and hands-free driving would be limited to carefully designated lanes, so that lane changes are minimal and other drivers are inconsequential.

The cost in infrastructure to guide such robo-cars will be monstrous ...

A deep infrastructure isn't really needed at first.

... and utterly inappropriate in a world that should be aiming to tighten its belt rather than indulging its techno-addiction in this manner. And to what end? What will the significant benefit be?

Fewer drunk drivers, fewer sleepy drivers, one or two extra hours for socializing, learning, or entertainment. Have your car park itself. Lend someone your car without driving it (Two or three people with erratic schedules could share a car and leverage it as if it were their own, without relying on each other or being limited by the others' schedules). Send it for a car-wash on its own. Pick you up after a night clubbing. Higher fuel efficiency.

I clearly recall articles in the 1970's confidently predicting an end to traffic jams when flying cars became freely available in , er, about 9 years ago in fact.

The problem with flying cars seems to be the hardware and the extra dimension, not the software. The problem with self-driving cars seems to be the software, not the hardware.

[SK.]

Will it be able to read the hand-painted sign that says "Caution - horses on road" ? Or will every hazard - permanent or temporary - need electronic flagging?
Will it be able to tell the difference (as mentioned earlier) between a dog and a child who has fallen over? Or will every person carry a transmitter?
Depends on the approach adapted of course, but a fully autonomous car will be able and has to cope with random obstacles. There´s already very capable sensors out there, and they will only further grow in capability, while shrinking in size and cost. 10 years ago state of the art laser scanners would be able to get 10000 distance measurements max per second, today a Velodyne HDL-64E scans 1.3 million points per seconds and basically gives you an instant 3D model of your surroundings out to 120m at 15Hz.
It´s of course still highly impractical for everyday use, but an example of the progress that is being made.
Other than that, there´s of course the computer vision community working towards people detection, scene understanding etc. A lot of stuff already works under controlled or semi-controlled conditions, but there´s still a long way to go, no question.


The cost in infrastructure to guide such robo-cars will be monstrous and utterly inappropriate in a world that should be aiming to tighten its belt rather than indulging its techno-addiction in this manner. And to what end? What will the significant benefit be?
As alluded to earlier, it sure depends on the approach. Approaches range from completely dumb cars that get steered by the environment to completely autonomous cars that require no change in the environment at all, with various stages of inbetweens.
In Sebastian Thrun´s lecture I talked about in my earlier post, he makes some very good points ( from memory and commented by me, so not 100% accurate citation wise ):
-30% of today´s car weight is dedicated to safety measures. A lot of that could get saved, if we get reliable autonomous driving.
-only 8% of the area of an highway is covered by cars even in traffic jams. By automomous driving, utilization could be vastly improved, resulting in more than doubling the capacity of roads - without doing any construction work.
-today´s cars sit around doing nothing nearly all of their "life". With autonomous cars, one could drive to work, but instead of parking the car, you could just send it to pick up the wife or share it with other people. A vast reduction in the number of cars would be possible this way.

All that saves a vast amount of money, resources, lives and pollution.
Of course, the technology isn´t there yet, but the advantages are pretty obvious.


I clearly recall articles in the 1970's confidently predicting an end to traffic jams when flying cars became freely available in , er, about 9 years ago in fact.
Well that´s the reason why most people here don´t make such dumb predictions. It´s obvious that there´s still a lot of work to be done, so predicting when we´re there isn´t possible with any amount of accuracy yet.
Nonetheless, the advantages are pretty clear and there is nothing in principle that will prevent we eventually get there.

[Rasmus]

Will it be able to read the hand-painted sign that says "Caution - horses on road" ? Or will every hazard - permanent or temporary - need electronic flagging?

Neither.

Not every human driver will be able to read those signs, either. You *can* drive in a foreign country without knowing what the local word for "horse" is, after all. And afaik the ability to even read in your own language is not a requirement to be getting a driver's licence, either.

Why should an automated car be required to know at least as much or even more about a local stretch of road than any random driver who doesn't happen to live around the corner?

Will it be able to tell the difference (as mentioned earlier) between a dog and a child who has fallen over? Or will every person carry a transmitter?

Not ervery human driver will always be able to tell the difference, either. Both dogs and children get killed in traffic after all, and then, even some dogs live.

The cost in infrastructure to guide such robo-cars will be monstrous and utterly inappropriate in a world that should be aiming to tighten its belt rather than indulging its techno-addiction in this manner. And to what end? What will the significant benefit be?

Fewer dogs and children would get killed, for one. A lot, I'd assume. I had to drive ~800km (500 miles) last week in a single trip. It took me over 10 hours and i was driving for well over 9 of those. Never mind the other 2 hours of work I had to put in earlier on that same day.

I would have much preferred sitting in the back of the car, being productive, reading a book or getting some sleep.

I clearly recall articles in the 1970's confidently predicting an end to traffic jams when flying cars became freely available in , er, about 9 years ago in fact.

So? We have flying cars. They are called air planes. They are entirely usable and can be bought and sold. Granted, they haven't replaced cars, but they sure seem to work to me.

[TjW]

The GPS that would be used would have to come installed in the car, adding another 2000-3000 bucks.$500 for the installed GPS unit that could have come with my 2006 Toyota Prius, but wasn't.
$59.95 for a GPS module at Fry's.


GPS technology is no longer expensive. The bells and whistles surrounding it in consumer apps may increase the cost.

[GlennB]

Quite a few replies above, so forgive me if I make a few points here rather than address every post ---

1. The absolute need for the possibility of immediate human intervention will make being "incapable in charge of a motor vehicle" (through drugs or exhaustion) just as illegal as it is today. Sitting in the back seat pissed or sleeping will not be an option.
2. "Pulling over" while the driver gets their act together on motorway/freeway, dual carriageways, urban freeways and a whole host of urban roads that have parking restrictions is currently illegal. I can't see much hope of that changing unless we rebuild our cities.
3. 90%* of the world's driving population that might have a use for this technology is literate in the home language. "Take care, horses on road" is useful information.
4. Assessing whether to hit that rock and wreck your suspension or swerve and plough into the line of cyclists heading your way is an easy one for a human driver, i.e. screw the suspension. If the obstruction is actually a cardboard box we hit it. Robo-car will need to assess whether the obstruction is a rock or a relatively harmless cardboard box.
5. Life in the future will not be like Star Wars.

* totally made up. It could be 9% and it would still be useful information.

[Richard Masters]

Quite a few replies above, so forgive me if I make a few points here rather than address every post ---

1. The absolute need for the possibility of immediate human intervention will make being "incapable in charge of a motor vehicle" (through drugs or exhaustion) just as illegal as it is today. Sitting in the back seat pissed or sleeping will not be an option.

Sounds like you can tell the future. I've been on boats that are set on autopilot overnight while traveling long distances. There is always someone on watch, but this technology is from the 50s. There are plenty of recognition systems today that can account for many scenarios, especially on a highway with lanes designated for these cars.

Also, look up compressed sensing and genetic programming.

2. "Pulling over" while the driver gets their act together on motorway/freeway, dual carriageways, urban freeways and a whole host of urban roads that have parking restrictions is currently illegal. I can't see much hope of that changing unless we rebuild our cities.

You wouldn't set your car to drive for you on any urban road. For that matter, the artificial intelligence (today) wouldn't be able to drive in most countries or states like Hawaii and Oklahoma where drivers are extremely erratic; that however, doesn't preclude the ability to have designated lanes at first. As the technology improved, these cars could do more.

3. 90%* of the world's driving population that might have a use for this technology is literate in the home language. "Take care, horses on road" is useful information.

See above.

4. Assessing whether to hit that rock and wreck your suspension or swerve and plough into the line of cyclists heading your way is an easy one for a human driver, i.e. screw the suspension. If the obstruction is actually a cardboard box we hit it. Robo-car will need to assess whether the obstruction is a rock or a relatively harmless cardboard box.

http://web.mit.edu/mcgovern/html/News_and_Publications/0702_poggio.shtml

5. Life in the future will not be like Star Wars.

It doesn't have to be. Cars can already park themselves; biological entities can already be distinguished from non-biological entities in object recognition systems, sure there are obstacles, but most of your objections can be addressed or are not relevant to some autonomy some of the time.

[Christian Klippel]

5. Life in the future will not be like Star Wars.

Sorry, but i think that statement is more or less nonsense. While it is true that we do not know for sure how it will be in the future, we still can see trends. Making such a statement is like predicting "X will happen at year Y".

As an example: A bit over half a century ago computer scientists and engineers said it was impossible that people will ever have their own computers. Back then, a computer was a room full of stuff, or more than one room. The computing power was laughable, compared to today. No one would have said with a straight face that "personal computers" are even a possibility in the future, except for Sci-Fi writers.

However, today even the average cellphone has multitudes more computing power than these early machines. When i got my first PC, hard disk sizes were 10 and 20 megabytes, RAM was a few hundred kilobytes. Today, a PC with 1 gigabyte of RAM installed is considered low on RAM.

Speaking of hard disk drives, how often have we heard that "now the end of capacity expansion is reached. more is not possible." And what happened? Today we have 1 terabyte or more in a single hard disk.

Really, to say that something won't happen in the future, just because today it looks like Sci-Fi, doesn't make much sense in many cases. Laser weapons, holographic displays and memory, war robots. All that Star-Wars stuff has come true already. Except for stuff like beaming, time- or FTL-travel and a holodeck, i am pretty sure almost everything is possible in the future. And in the case of the holodeck i am not really sure that it is not possible.

Greetings,

Chris

[Rasmus]

Quite a few replies above, so forgive me if I make a few points here rather than address every post ---

1. The absolute need for the possibility of immediate human intervention will make being "incapable in charge of a motor vehicle" (through drugs or exhaustion) just as illegal as it is today. Sitting in the back seat pissed or sleeping will not be an option.

I am not convinced that this need exists. A big red "get out of the way and let the ambulance pass"-button would be quite sufficient.

There will always be cases where one system is inferior to another. Some people are dad because they have been wearing seatbelts. That doesn't mean seatbelts are a bad idea, or that they are dangerous or that people shouldn't wear them. Far from it, in fact

2. "Pulling over" while the driver gets their act together on motorway/freeway, dual carriageways, urban freeways and a whole host of urban roads that have parking restrictions is currently illegal. I can't see much hope of that changing unless we rebuild our cities.


So don't let the car do that.

3. 90%* of the world's driving population that might have a use for this technology is literate in the home language. "Take care, horses on road" is useful information.

Yes, it is useful. But it is not necessary information to have if you want to operate a car. It is not required, either.

4. Assessing whether to hit that rock and wreck your suspension or swerve and plough into the line of cyclists heading your way is an easy one for a human driver, i.e. screw the suspension. If the obstruction is actually a cardboard box we hit it. Robo-car will need to assess whether the obstruction is a rock or a relatively harmless cardboard box.

See above: Some systems are going ot better than others in some situations. That doesn't mean they are the better systems.

[sheik yerbouti]

Let's say you have self-driving car.

Now the first person dies in an accident involving one.

And the lawyer proved to 12 morons, good and true, that it was due to a programming error.

Where do you think the tiny shards of the entire industry are going to be stored after the award?

For this reason, I think a fully self-driving car is most likely to happen in a non_US country. Possibly Japan, they seem to be far more comfortable with robots than Americans.

[GlennB]

You wouldn't set your car to drive for you on any urban road. For that matter, the artificial intelligence (today) wouldn't be able to drive in most countries or states like Hawaii and Oklahoma where drivers are extremely erratic; that however, doesn't preclude the ability to have designated lanes at first. As the technology improved, these cars could do more.


OK. You are drunk and sleeping in the back of the robo-car, on a motorway, in the designated robo-lane. An electronic failure occurs, requiring you to take over the driving. The car manages to pull over to allow you to wake up (this is the good scenario, we can talk about the bad one if you wish).

Already 'you' are dicing with the law, by obstructing the emergency lane (or hard-shoulder as it's also known in the UK). What happens next ???

You send your car away to 'self-park'. It suffers a flat tyre with you 1km away. What happens next ???

[Richard Masters]

OK, you are drunk and sleeping in the back of the robo-car, on a motorway, in the designated robo-lane. An unpredictable event occurs, requiring your human intervention. The car pulls over to allow you to wake up.

Already 'you' have broken the law, by obstructing the emergency lane (or hard-shoulder as it's also known in the UK). What happens next ???

A) You pay the fine.
B) You change the law.
C) You are redirected to a rest area.
D) You make use of the previously discussed lanes.
E) <insert solution here>
F) One or some of the above.

[PixyMisa]

Imagine a dozen or more trucks running in a "trucks only" lane at constant speed with only inches between bumpers, in continuous contact to maintain spacing, across the US plains.
We've actually had those for some time (http://en.wikipedia.org/wiki/Train).

[Richard Masters]

...
You send your car away to 'self-park'. It suffers a flat tyre with you 1km away. What happens next ???

The self-parking at a distance is much too difficult, so it's not something I would worry about for another 15 to 20 years, and I wouldn't consider it an obstacle to hands-free driving.

To address your scenario, you call your insurance company, (or your car does it for you), and you receive a text message or a phone call while the insurance company replaces your tire, just like you might order a pizza; without the hassle of ordering it.

Or, you accept that flat tyres happen and you choose to park your car yourself.

[Earthborn]

Makaya's objections that it would be difficult for the car to know exactly where it is and where it should be going are invalid. Every smart phone can do that now. Knowing where it is and where to go is easy. Yes, it does require huge computing power, monstrous infrastructure and humongous databases, but that is actually the reason why it is so easy; the problem can be solved by throwing huge computing power, monstrous infrastructure and humongous databases at the problem. In other words, we know how to do it. And much of is already done.

Reading handwritten signs and getting information from them? If that was the biggest obstacle for robot cars, they would be just around the corner. As long as the person who wrote it doesn't try to make a CAPTCHA out of it, it would be easy. I do think that when robot cars hit the market, there will be many objects electronically tagged. Not that robot cars would otherwise be unable to avoid them, but there probably is a market for cheap transmitters horse riders can buy that would cause robotic cars to be a little extra careful around horses to not to startle them. Just an extra precaution just in case the obstacle recognition doesn't recognise it immediately.

Highway driving could be automated tomorrow. I already linked to a site that shows that this can be done with 1995 technology.

These things are not the biggest obstacles towards a self-driving future. One of the biggest obstacles is the fact that while there are aspects of driving that are fairly easy to automate, those are also the aspects that we have the least to gain from automating them, while the hardest aspects remain fiendishly difficult. Highway driving is easy, every other car drives in the same direction, more or less the same speed, lanes are clearly marked and there are no pedestrians dashing through left and right. Easy for a robot, but also easy for a human, so there is not much incentive to automate.

City driving is much more difficult; there are much greater differences in speed between traffic participants. The streets are littered with different markings or have no markings at all, there are signs everywhere, some of which may look like traffic signs without being traffic signs. It is difficult for a human driver, but we haven't a clue how to program a computer to extract the necessary traffic information, and not be overwhelmed by the sheer amount of data and cause it to behave erratically. We probably need a few paradigm shifts in artificial intelligence to even begin to comprehend how to solve the problem. It is a pity, because it is just the environment where we need the automation the most.

The biggest obstacle is to get the robot cars to have behaviour robust enough so we can trust them to do all our driving for us. People in this thread have already mentioned ships, planes, trains and cars that are automatic. No matter how sophisticated they all share a common feature: they are supervised by experts who can be presumed to understand the limitations of these machines and have manual overrides or kill switches in case something goes wrong. But a family car that in the middle of a emergency just gives up and tells a human (who may be sleeping or watching a movie) to take over is the last thing you want. The car itself should come up with an appropriate action no matter what happens, even if it isn't exactly what a human would do in the same situation. The solution to that still seems very far away.

It is not something that seems impossible however, nor would it require a solution to many of the hard problems in artificial intelligence. Just consistently following rules will result in pretty safe driving as long as the car can recognise which rules to follow. Not only are the situations in which it preferable to drive through a stoplight or over the sidewalk extremely rare, those situations are probably encodable into rules as well.

No doubt we are decades away from driverless cars, meaning they'll come decades too late. No doubt the problems towards them are huge. But they're not insurmountable.

[Earthborn]

You send your car away to 'self-park'. It suffers a flat tyre with you 1km away. What happens next ???Scenarios where the robot car breaks down to the point that it can't continue don't show much appreciation for how powerful this technology would be. Cars get flat tyres because its owners don't check their tyres often enough. Cars break down because its owners forget about proper maintenance. But a robot car would probably be packed with diagnostic equipment so it knows when there is something wrong. And every [however many] miles/kilometers it can automatically sneak out of the garage at night to go a repairshop for a routine checkup. The very moment it leaves, a loan car will arrive automatically because it was ordered earlier by your car.

[William Parcher]

The NES and SNES consoles are not public domain and they were reverse-engineered to provide the public with NES and SNES emulators. Today, you can download and play any SNES video game on your computer.



http://i179.photobucket.com/albums/w310/william_parcher/8649c067.jpg

[Richard Masters]

http://i179.photobucket.com/albums/w310/william_parcher/8649c067.jpg

:D

http://www.pcauthority.com.au/News/144367,you-wouldnt-download-a-car-would-you.aspx

[GlennB]

OK. You are drunk and sleeping in the back of the robo-car, on a motorway, in the designated robo-lane. An electronic failure occurs, requiring you to take over the driving. The car manages to pull over to allow you to wake up (this is the good scenario, we can talk about the bad one if you wish).

Already 'you' are dicing with the law, by obstructing the emergency lane (or hard-shoulder as it's also known in the UK). What happens next ???

You send your car away to 'self-park'. It suffers a flat tyre with you 1km away. What happens next ???


A) You pay the fine.
And are - at best - one step closer to losing your licence. Or, if sufficiently drunk, you lose it immediately and maybe get a jail sentence thrown in. Not a solution.

B) You change the law.
To allow ratfaced drunk driving? I don't think so.

C) You are redirected to a rest area.
"Next motorway services 28 miles" and you're ratfaced drunk? Not a solution.

D) You make use of the previously discussed lanes.
See above.

E) <insert solution here>
That's the problem, there isn't one. Robo-cars would require the driver to be every bit as capable as an airline pilot who has his plane on auto-pilot. And there is frequently a spare pilot who is also capable.

F) One or some of the above.

None of them work in the situation under discussion. And, as pointed out elsewhere, such technology is of the least use on multi-lane highways.

[Richard Masters]

GlennB, I don't think I made my position on your objections clear enough. If you happen to be drunk when your robo-car requests your assistance, it should pull over (as this is clearly an emergency), or it should look for the nearest parking lot or rest area.

Now, here's a really neat aside: if your car learns your driving habits, it will recognize that you are drunk, or driving badly, and can prevent you from swerving into oncoming traffic. If you make too many unjustified lane changes, or can't keep the car steady, the car will pull over and prevent you from driving.

ETA: I see you added an "electronic failure" in your later edits, that's moving the goalposts as far as I'm concerned.

[makaya325]

If a self-driving car was involved in an accident, which was out of your reach, would you still be responsible for the accident, even though you only pushed a button and the car did the rest on its own?

[Pulvinar]

If a self-driving car was involved in an accident, which was out of your reach, would you still be responsible for the accident, even though you only pushed a button and the car did the rest on its own?

Compare it to the case today: say your wheel comes off, causing an accident. It might be the fault of the car maker (the bearing broke), or you (you forgot to replace the lug nuts). Same sort of possibilities would be examined for the self-driving car, though one thing would certainly make that easier: a "black box" feature keeping a record of all sensor data. If the algorithm was determined to be at fault it could be fixed and downloaded into all cars so that case theoretically wouldn't ever happen again.

[SK.]

That's the problem, there isn't one. Robo-cars would require the driver to be every bit as capable as an airline pilot who has his plane on auto-pilot.
There are fundamental differences between robo-cars and airliners. The most important is that you can bring a car to a safe state in mere seconds, while you obviously can´t just park an airliner.
Even in case of an electrical failure or software bug, a cascade of redundant systems that can bring the car to a safe halt could be made relatively cheap, simple and reliable.

[Andrew Wiggin]

In Robert Heinlein's 'Job', while the protagonists are hitchhiking naked across a future version of texas (god keeps popping them from time stream to time stream to make life extra fun, and popped them into this one while they happened to be bathing), they come across a roadway surrounded by an electric fence. They find a streambed and crawl under the fence in hopes of flagging down one of the cars on the road, and eventually are picked up by the Devil himself, who happens to be travelling home (he lives in texas and commutes to hell). The first thing he does is to caution them, as the roadway was designated for automated cars only, and pedestrians can get in serious trouble for being caught on the wrong side of the electric fence. Good descriptions of what it would be like to travel by automated car follow, including the fact that he needs to over-ride the computer to pull off the road and back on, and if I recall, the way it causes the computer to have to reschedule and compensate for the time lost on the journey. I don't recall if the car was described as being internally or externally guided. I don't specifically recall mention of sensors in the roadway.

A

[Rasmus]

There are fundamental differences between robo-cars and airliners. The most important is that you can bring a car to a safe state in mere seconds, while you obviously can´t just park an airliner.
Even in case of an electrical failure or software bug, a cascade of redundant systems that can bring the car to a safe halt could be made relatively cheap, simple and reliable.

And if all else really failed, the car would presumably just stop. So that might make it more difficult for other drivers and it might be somewhat dangerous but that already happens now as well.

[Rasmus]

And are - at best - one step closer to losing your licence. Or, if sufficiently drunk, you lose it immediately and maybe get a jail sentence thrown in. Not a solution.

To allow ratfaced drunk driving? I don't think so.

You are assuming that whoever told the car where to go would be legally "driving" it. That doesn't necessarily have to be so. I still doubt that you need someone to be able to operate the vehicle in an emergency.

Again, we already have situations today where cars simply break down and cannot be moved anymore.

In fact, I wouldn't want a system to be used that required a capable driver to be at the ready at any time: People would drive drunk, or fall asleep, or drive without a licence or simply forget how to react in certain situations. They do all these things *now* when they actually have to drive the cars themselves.

"Next motorway services 28 miles" and you're ratfaced drunk? Not a solution.

So there will be situations where the automated car breaks and just grinds to a halt. So what? How is that new, how is it different from what cars are already doing?

[PixyMisa]

In Robert Heinlein's 'Job', while the protagonists are hitchhiking naked across a future version of texas (god keeps popping them from time stream to time stream to make life extra fun, and popped them into this one while they happened to be bathing), they come across a roadway surrounded by an electric fence. They find a streambed and crawl under the fence in hopes of flagging down one of the cars on the road, and eventually are picked up by the Devil himself, who happens to be travelling home (he lives in texas and commutes to hell).
Sure, that's only, what, twenty minutes on a good day?

[GlennB]

And if all else really failed, the car would presumably just stop. So that might make it more difficult for other drivers and it might be somewhat dangerous but that already happens now as well.

Remeber the Windows 3.1 "blue screen of death" ? Robo-car will not have the luxury of going through this kind of developmental stage. It will kill and it will cause traffic chaos. Anyone here full of confidence that such a car can be fail-safe from the outset? Anybody want to volunteer their city for field-testing?

Working some years ago on the software for a relatively simple sensoring system (entry and exit gates for trains on the new Hong Kong airport line) we were required to develop a system that could detect two passengers who were trying to shuffle together - up close and synchronised :) - through an entry gate on one ticket. The best we could manage was issue a "Warning. Wide passenger" alarm, or "Warning. Passenger splitting" (really!) alarm. At which point station staff would attend and investigate.

Of course technology advances, and hugely so. But robo-car has zero room for error, vast expense up front, and extremely doubtful benefits. It's the kind of thing R+D departments love, because they get to play with expensive toys. It's the kind of thing journalists love, because it's easy and exciting copy. But it's the kind of thing people hate, won't buy if they have the choice, and will deeply resent if they have to subsidise it while living in a location where its use isn't available or applicable.

[Darat]

I find it very interesting is that the "anti- the idea" side seem to be wanting robotic cars to do things and cope with situations that humans either do very badly or not at all. (I'm just making that comment as a "sociological" comment not as an ad hom of any kind.)

I think there are two strands to whether such vehicles are possible. The first is the purely technological and the second is the social. We seem to accept that with human drivers there will be deaths caused by the drivers being bad drivers, or being unable cope with novel and unusual circumstances but would we accept the same level of deaths caused by computer-controlled cars?

I suspect not, which I think is a pity because I reckon even if we had computer controlled cars that were only as good drivers as most of us are, then we would have much reduced fatalities and serious inures because at least speed limits would be obeyed.

[GlennB]

GlennB, I don't think I made my position on your objections clear enough. If you happen to be drunk when your robo-car requests your assistance, it should pull over (as this is clearly an emergency), or it should look for the nearest parking lot or rest area.

Now, here's a really neat aside: if your car learns your driving habits, it will recognize that you are drunk, or driving badly, and can prevent you from swerving into oncoming traffic. If you make too many unjustified lane changes, or can't keep the car steady, the car will pull over and prevent you from driving.

ETA: I see you added an "electronic failure" in your later edits, that's moving the goalposts as far as I'm concerned.

I was only hammering this "drunk" business because a few people mentioned it early on as a possible benefit, and it's an easy one to defantasise.

And raising the subject of electronic failure is not 'shifting the goalposts'. It's a serious issue. The malfunction of (say) a distance sensor on robo-car might make it non-functional in the robo sense, but still perfectly functional to the human driver. Causing local gridlock or motorway tailbacks because you were, for whatever reason, incapable of dealing with such a failure will not be an option. Pulling over because the car can't continue (a flat tyre, say) is normal and acceptable. Pulling over because the driver can't drive is not and never will be.

[SK.]

Pulling over because the car can't continue (a flat tyre, say) is normal and acceptable. Pulling over because the driver can't drive is not and never will be.
Well as far as I´m concerned, if I had the choice, I´d rather have some robocars pulling over when they have a sensor failure than people falling asleep at the wheel and causing hundreds of deaths each year (which is "state of the art" today).

[Rasmus]

Remeber the Windows 3.1 "blue screen of death" ? Robo-car will not have the luxury of going through this kind of developmental stage. It will kill and it will cause traffic chaos. Anyone here full of confidence that such a car can be fail-safe from the outset? Anybody want to volunteer their city for field-testing?

I don't expect the car to be fail-safe or perfect. Why would I? Nothing else is fail-safe or perfect. Least of all, humans that are now allowed to drive cars.

Yes, automated cars will be the cause of damage, death and injuries. So what? As long as there are fewer of them as what we have now it's a good thing.

Again, seatbelts have caused and will continue to cause injuries and deaths, too. They are not perfect. There are situations where it would be better to not wear one.

Working some years ago on the software for a relatively simple sensoring system (entry and exit gates for trains on the new Hong Kong airport line) we were required to develop a system that could detect two passengers who were trying to shuffle together - up close and synchronised :) - through an entry gate on one ticket. The best we could manage was issue a "Warning. Wide passenger" alarm, or "Warning. Passenger splitting" (really!) alarm. At which point station staff would attend and investigate.How many people got killed prior to the installation of that system, and for how many deaths would you estimate it was responsible afterwards?

Of course technology advances, and hugely so. But robo-car has zero room for error,********. [Edit: The auto-censor doesn't seem to like the name of a sceptical TV show. Oh well ...] Plain and simple. Everything else has room for error. A lot of people die every day because things are far less than perfect.


vast expense up front, and extremely doubtful benefits.Doubtful benefits? I know I'd pay for it, if I could afford it. Just like I Paid for my GPS now.

It's the kind of thing R+D departments love, because they get to play with expensive toys. It's the kind of thing journalists love, because it's easy and exciting copy. But it's the kind of thing people hate, won't buy if they have the choice, and will deeply resent if they have to subsidise it while living in a location where its use isn't available or applicable.I would love to buy something like this, and I doubt such a system would have to be subsidised. It would have to be if it relied on specific road-modifications - but I don't see those coming. I wouldn't buy a self-driving system that could only drive me from a very specific set of As to an equally specific set of Bs. (We have something like that already. It's called a bus.)

[Rasmus]

Pulling over because the driver can't drive is not and never will be.

But it is perfectly acceptable to do that.

If I suddenly find I can no longer safely operate my car I will pull over.

You have not justified why a car would need a human stand-by driver in case the automated one would fail. We let humans drive all on their own, even though it would be better to have a second driver on board at all times as well.

[Horatius]

Highway driving could be automated tomorrow. I already linked to a site that shows that this can be done with 1995 technology.

These things are not the biggest obstacles towards a self-driving future. One of the biggest obstacles is the fact that while there are aspects of driving that are fairly easy to automate, those are also the aspects that we have the least to gain from automating them, while the hardest aspects remain fiendishly difficult. Highway driving is easy, every other car drives in the same direction, more or less the same speed, lanes are clearly marked and there are no pedestrians dashing through left and right. Easy for a robot, but also easy for a human, so there is not much incentive to automate.

...

The biggest obstacle is to get the robot cars to have behaviour robust enough so we can trust them to do all our driving for us.


Here's the thing: We don't need the cars to do all the driving for it to still be useful. You've dismissed highway driving as being so easy, there's no incentive to automate it. I'd disagree. I hate highway driving, particularly for long distances. It's boring. I'd buy a car that could drive itself on the highway, even if I had to do all the driving to get it on and off the highway.

I also think you underestimate the benefits of automating just this part of driving. Highway accidents may be less common than accidents in city driving, but, due to higher velocities, they're also much more deadly. Reducing the number of just these accidents would probably have a disproportionate affect on the number of accident injuries and fatalities.

I imagine any such auto-driving system will be implemented in stages - highway driving today, then gradually expand the capabilities.



If a self-driving car was involved in an accident, which was out of your reach, would you still be responsible for the accident, even though you only pushed a button and the car did the rest on its own?


Any such objections are irrelevant. Fault for such situations will be decided the same way they were first decided when cars were invented - we'll take it to court, and let the judges sort it out.

[progressquest]

snip
Some people are dad because they have been wearing seatbelts.
snip

I don't want to get involved with you personal life, but I think you are using the seatbelt wrong.

And now back to your regularly scheduled discussion.

[Richard Masters]

I was only hammering this "drunk" business because a few people mentioned it early on as a possible benefit, and it's an easy one to defantasise.

And raising the subject of electronic failure is not 'shifting the goalposts'. It's a serious issue.

It's a serious issue, but it's still shifting the goalposts, since I addressed your previous objections.

The malfunction of (say) a distance sensor on robo-car might make it non-functional in the robo sense, but still perfectly functional to the human driver. Causing local gridlock or motorway tailbacks because you were, for whatever reason, incapable of dealing with such a failure will not be an option.

What is it with "...is not an option"?; I don't think it means what you think it means.

Pulling over because the car can't continue (a flat tyre, say) is normal and acceptable. Pulling over because the driver can't drive is not and never will be.

OK, what's your reasoning behind such absolutism?

[arthwollipot]

Remeber the Windows 3.1 "blue screen of death" ? Robo-car will not have the luxury of going through this kind of developmental stage. It will kill and it will cause traffic chaos. Anyone here full of confidence that such a car can be fail-safe from the outset? Anybody want to volunteer their city for field-testing?We already have zero-downtime computing systems - they're not that hard. The computers that run air traffic control at major airports, for example. The Stock Exchange. Not everything runs on Windows. :)

[GlennB]

We already have zero-downtime computing systems - they're not that hard. The computers that run air traffic control at major airports, for example. The Stock Exchange.

Your faith is touching.

In 2008 (http://www.telegraph.co.uk/finance/newsbysector/banksandfinance/2795834/London-Stock-Exchange-trading-derailed-by-computer-crash-on-frenzied-day.html):

"Trading on the London Stock Exchange has been halted after a computer system failed on one of the most frantic days of trading so far this year.
In an embarrassment for the LSE, the exchange said that no orders can be entered or executions of those trades occur. The LSE plans to bring back trading in a “controlled way’’, but couldn't say how long that will take"

In 2005 (http://www.independent.co.uk/news/uk/home-news/flight-chaos-as-computer-crash-hits-britains-air-traffic-control-504160.html):

"Thousands of flights out of Britain were cancelled or delayed after air traffic control systems crashed yesterday. The chaos was made worse by a simultaneous computer failure at Maastricht in the Netherlands which covers airspace over much of north-western Europe."

And, of course, errors and crashes aren't quite the same thing.

"On June 5, Moody’s Corp. will host an “Investor Day,” where top executives will be trotted out to talk about their outlook for the credit-rating agency.
They will have some serious explaining to do now that Moody’s is suffering through one of the biggest embarrassments in its 99-year history. The Manhattan-based company said Wednesday that it is conducting a “thorough review” after a report in the Financial Times said a computer error led Moody’s to incorrectly assign “AAA” ratings to securities worth billions of dollars. Shares in Moody’s plunged 15% on the news."

Here (http://www.crainsnewyork.com/article/20080521/FREE/238387364), in 2008

[SK.]

Your faith is touching.
Well building a robocar with faith in it´s one computer system not failing certainly wouldn´t be the way to go. But as explained earlier, redundant systems for bringing the vehicle to a safe state if something goes awry are realizable much cheaper and simpler than for example in airplanes.
I´d expect that electric and mechanical reliability won´t be the factors that slow the introduction of robocars, but reliability in cognition and decision making software (as in: getting the algorithms right so they react correctly to basically 100% of situations).

[Rasmus]

Your faith is touching.

In 2008 (http://www.telegraph.co.uk/finance/newsbysector/banksandfinance/2795834/London-Stock-Exchange-trading-derailed-by-computer-crash-on-frenzied-day.html):

"Trading on the London Stock Exchange has been halted after a computer system failed on one of the most frantic days of trading so far this year.
In an embarrassment for the LSE, the exchange said that no orders can be entered or executions of those trades occur. The LSE plans to bring back trading in a “controlled way’’, but couldn't say how long that will take"

In 2005 (http://www.independent.co.uk/news/uk/home-news/flight-chaos-as-computer-crash-hits-britains-air-traffic-control-504160.html):

"Thousands of flights out of Britain were cancelled or delayed after air traffic control systems crashed yesterday. The chaos was made worse by a simultaneous computer failure at Maastricht in the Netherlands which covers airspace over much of north-western Europe."

And, of course, errors and crashes aren't quite the same thing.

"On June 5, Moody’s Corp. will host an “Investor Day,” where top executives will be trotted out to talk about their outlook for the credit-rating agency.
They will have some serious explaining to do now that Moody’s is suffering through one of the biggest embarrassments in its 99-year history. The Manhattan-based company said Wednesday that it is conducting a “thorough review” after a report in the Financial Times said a computer error led Moody’s to incorrectly assign “AAA” ratings to securities worth billions of dollars. Shares in Moody’s plunged 15% on the news."

Here (http://www.crainsnewyork.com/article/20080521/FREE/238387364), in 2008

Do you want me to go and dig out every report of traffic accidents where the drivers were to blame for wreckage, death and injury?

You still haven't explained why an automated car should be expected to not only be much, much better than human drivers but perfect in any situation.

[arthwollipot]

Your faith is touching.Uh-huh. You found three examples that made news going back five years. Want to take bets on how frequently my laptop crashed in five years? How many air traffic control systems and stock exchanges are there in the world again? Three in five years is a damn fine record and beyond the capacity of most average computer systems.

"Zero-downtime" is of course an ideal, and I will take my licks for suggesting that it was an absolute. In hindsight that was a little optimistic. Regardless, systems that are marketed as "zero-downtime" exist, and they crash a whole lot less than regular computers.

[GlennB]

"Zero-downtime" is of course an ideal, and I will take my licks for suggesting that it was an absolute. In hindsight that was a little optimistic. Regardless, systems that are marketed as "zero-downtime" exist, and they crash a whole lot less than regular computers.

Zero downtime is a small issue. Robo-cars would effectively need zero failures from the outset.
Normal software and hardware can usually be tested to death in non-critical environments. Who will volunteer their city or motorway for beta testing of robo-car? This is why space missions sometimes fail catastrophically (confusion over units for example. And I believe one failed when crossing the international date line). They cannot be live-tested. The variables in a live robocar environment are truly staggering. And, of course, that's just a single issue out of thousands.

And while any hardware or software failure can be a real hassle, how would any of it compare to having the control of a car travelling - on robocontrol - at 70mph in the fast lane of a busy motorway unexpectedly thrown at you while you're reading a book or snoozing?

I have asked a few times how any sensor system will ever tell the difference between a fallen child (stop definitely) and a dog (don't stop, maybe). Not a single adequate reply. Until that one, randomly chosen, isolated issue alone is resolved robo-car doesn't get off the drawing board.

[Rasmus]

Zero downtime is a small issue. Robo-cars would effectively need zero failures from the outset.

No. Why would it?

Normal software and hardware can usually be tested to death in non-critical environments. Who will volunteer their city or motorway for beta testing of robo-car?

Bweta testing would hardly happen in a live environment. I wouldn't mind sharing my roads with the first models that were deployed, though. I cannot say this often enough: I expect them to be much better company than the average driver.

This is why space missions sometimes fail catastrophically (confusion over units for example. And I believe one failed when crossing the international date line). They cannot be live-tested. The variables in a live robocar environment are truly staggering. And, of course, that's just a single issue out of thousands.

Yes. And there will be accidents involving automated cars where people will die. So what?

And while any hardware or software failure can be a real hassle, how would any of it compare to having the control of a car travelling - on robocontrol - at 70mph in the fast lane of a busy motorway unexpectedly thrown at you while you're reading a book or snoozing?

I doubt this would happen. Much better to rik a regualr crash.

I have asked a few times how any sensor system will ever tell the difference between a fallen child (stop definitely) and a dog (don't stop, maybe). Not a single adequate reply. Until that one, randomly chosen, isolated issue alone is resolved robo-car doesn't get off the drawing board.

How does any normal driver tell the difference? Reliably? Then how come that both dogs and children die in traffic? How come we still permit people to drive cars?

[SK.]

And while any hardware or software failure can be a real hassle, how would any of it compare to having the control of a car travelling - on robocontrol - at 70mph in the fast lane of a busy motorway unexpectedly thrown at you while you're reading a book or snoozing?
Probably not too bad, considering that lower level systems keeping watch of high level control will likely detect that something´s amiss and go into emergency mode and bring the car into a static safe state ASAP.
A triply redundant system of microcontrollers fitted with inertial sensors would likely be sufficient for short term emergency control and bringing the car to a safe standstill, even if basically everything else at higher level crashed. Also needs redundant means for manipulating the car´s mechanical interface of course (mainly brake, steering).

I have asked a few times how any sensor system will ever tell the difference between a fallen child (stop definitely) and a dog (don't stop, maybe). Not a single adequate reply. Until that one, randomly chosen, isolated issue alone is resolved robo-car doesn't get off the drawing board.
How? MMW-Radar, Thermal Cameras, Daylight Cameras, LIDARs etc. One can only speculate how good results of multi sensor fusion from those will get in the end.

But the other question of course is, do you hate some dogs so much that it´s a necessity for you to run them over? Stopping for both dogs and children seems like a pretty reasonable option to me. It´s highly likely that a working robocar will be able to react much faster and safer to unforeseen obstacles than a human driver and thus prevent many accidents. I fail to see how a robocar stopping in front of obstacles (be it dog or child) is somehow inferior to a human driver stopping for a child and (maybe) running a dog over.
So the solution is: Just let robocar stop if there is an obstacle of sufficient size.

[Christian Klippel]

Zero downtime is a small issue. Robo-cars would effectively need zero failures from the outset.
Normal software and hardware can usually be tested to death in non-critical environments.

Sorry, but that's simply wrong on both counts.

Why should automated cars have to have zero failures? First, humans are not zero failure. And they are driving cars today. Why should X be less error-prone than Y? Are you aware of how many failures, in regular cars driven by regular humans, happen every day? Oh, and take into account the quite complex electronic systems that are already used in cars. You know, there have been cases where the OBU had shut down a whole car, while driving rather fast on highways, making it coming to a full brake?

Those things are nothing new, they simply happen. There is no way to make a complex system 100% error/fault free under every circumstance. That just wont happen happen, not today and not tomorrow.

About the "testing to death" for hard- and software. That won't happen either. Simply because a "non-critical" environment does not have the same issues as a "critical" environment. Yes, you can test a lot of things in a lab, but in the real-world there will always be unexpected things. Testing a hard-/software combination for every possible input/state/failure simply takes too long. The more complex the system, the longer it would take. Prohibitively long, that is.

There is a good saying in software development: giving a newly written program to the average Jane Doe house wife, her grandmother and her childs, without them knowing how to use it, is the best beta-test you can have. If it does not crash then, the chances for crashes in general are low. Simply because these people _will_ do things in a way you never ever would have thought about.

As for "robot cars", the only way to really test them is to put them out "in the wild" and letting them cope with the real world. Nothing will ever replace that.

Greetings,

Chris

Edit: Heck, even the computer-system that bring people to outer space and back do have failures sometimes. That is a known fact. And the systems used there _are_ the best and most tested ones available. In fact, they use quite out-dated hardware platforms because after so many years it is safe to assume that the used parts are developed and debugged, unlike the always newest and shiniest toys that we buy for our computers. But still there are failures.

[Richard Masters]

Zero downtime is a small issue. Robo-cars would effectively need zero failures from the outset.
Normal software and hardware can usually be tested to death in non-critical environments. Who will volunteer their city or motorway for beta testing of robo-car?

You forget, you don't need to have people inside the car to do most of the testing. In fact, most of the initial testing can at first be done inside a computer.

For more sophisticated testing, manufacturers would probably build a highway complex and introduce random variables like an erratic robo-car, and see if the other normal robo-cars react appropriately.

Then they can add obstacles like a dead dog in the middle of the road or mannequins resembling humans.

This is why space missions sometimes fail catastrophically (confusion over units for example. And I believe one failed when crossing the international date line). They cannot be live-tested. The variables in a live robocar environment are truly staggering. And, of course, that's just a single issue out of thousands.

Satellites and spacecraft are often moving at impressive free-fall speeds; and there are several hardware components that need to be automated. In the case of the early robo-cars, there are the gas, brake and wheel direction.

Why compare heavy spacecraft with cars?

And while any hardware or software failure can be a real hassle, how would any of it compare to having the control of a car traveling - on robocontrol - at 70mph in the fast lane of a busy motorway unexpectedly thrown at you while you're reading a book or snoozing?

If you don't want a robo-car, you don't have to get one; though I suspect that if you have bought a car in the last 10 years, fuel-injection and gear-shifting are to some extent controlled by software already. But that's not an option, to you.

I have asked a few times how any sensor system will ever tell the difference between a fallen child (stop definitely) and a dog (don't stop, maybe). Not a single adequate reply. Until that one, randomly chosen, isolated issue alone is resolved robo-car doesn't get off the drawing board.

You didn't ask how. You stated it could never happen and it would not be an option.

I provided you with a link to ongoing research on that very subject. If you want the specifics:

It involves neural network training to identify and classify objects. You only need a few such as living and non-living, human and non-human.

It doesn't have to be perfect. Humans can't deal with these situations effectively anyway, so expecting a robo-car to do so is hypocritical.

If standard neural networks are not good enough, we can use Restricted Boltzmann Machines (neural networks on steroids), which are excellent at recognition and categorization.

Additionally, each unidentified object encountered can be sent to to a service where a human being can categorize the object for future use. In one day of driving hundreds to thousands of new objects can be identified. The RBM can recategorize the objects into 'avoid' or 'drive through' or thousands of other useful categories, like metallic or non-metallic. The advantage of using the Restricted Boltzmann Machine, is that you can work with sparse data. You will only need a few examples of a dog and a few examples of a child to correctly identify them frequently. But, the more you have, the better.

But that's only the recognition part of the visual system. Deciding what to do will be easy, as insurance companies already have excellent advice on when to avoid an object (such as deer) and when to drive through it.

A car that has been keeping track of road conditions (friction coefficient, brake efficiency, etc.) will be able to better estimate whether a car can slow down enough in time to avoid an object or whether swerving could cause a larger impact elsewhere, or a greater loss of control over the vehicle.

If the car chose to swerve, it could also send a message to other cars indicating it's intent to do so, and could also map the hazardous object and send that information to the cars behind it, so that they have more time to react.

For security reasons, the cars behind may choose to partly ignore the message, so as to verify that there truly is a dangerous object in the way.

And yet, this system would achieve safer driving than the average human.

Of course, I'm only speculating, and I spent a few minutes of my time on the adequate solution. Imagine what a greedy car manufacturer could come up with given a few thousand human researcher years.

As for the fear of electronic failures, those can be addressed with redundancy. You can have cheap back up systems which learn to signal an emergency and pull the car over. On top of that you can have a system that can locate rest areas and do the same. And so on, depending on your paranoia, or real concern for electronic failures. An electronic failure can be monitored and addressed with such redundant systems.

There are lots of things that can go wrong, but if you introduce new features one at a time, and use fuzzy systems or machine learning ensembles, you can reduce the chance of software failures to achieve super-human driving, at least in terms of safety.

[Horatius]

If the car chose to swerve, it could also send a message to other cars indicating it's intent to do so, and could also map the hazardous object and send that information to the cars behind it, so that they have more time to react.

For security reasons, the cars behind may choose to partly ignore the message, so as to verify that there truly is a dangerous object in the way.

And yet, this system would achieve safer driving than the average human.




This is probably the most important factor. While the first robocars may not be as good as we'd like in distinguishing children from dogs, it's pretty certain they'd be a lot better than humans at reacting to what other cars are doing around them.

If I were to slam on my brakes to avoid running over a kid, I would not be at all surprised to be rear-ended by the fellow behind me, who hasn't seen the kid, and had no reason to expect me to come to a sudden stop in the middle of the street. A robocar would be far better at avoiding such secondary accidents than a human.

[Darat]

...snip...

If I were to slam on my brakes to avoid running over a kid, I would not be at all surprised to be rear-ended by the fellow behind me, who hasn't seen the kid, and had no reason to expect me to come to a sudden stop in the middle of the street. A robocar would be far better at avoiding such secondary accidents than a human.

The only car accident I have been in (when I was driving) was on the slip-road to a motorway when it was raining. The car two in front of me slammed their brakes on, the car immediately in front of me took a bit longer to respond and ended up rear-ending the car, I braked and managed to stop with my front bumper about an inch from her back bumper. Well it was a stop, and an inch to spare before the car behind me slammed into me, pushing me into the car in front of me!

It will be very good news when we have "collision avoidance" systems in all cars rather than just being in expensive cars or an expensive option.

[Horatius]

Well it was a stop, and an inch to spare before the car behind me slammed into me, pushing me into the car in front of me!

It will be very good news when we have "collision avoidance" systems in all cars rather than just being in expensive cars or an expensive option.



I was a passenger in a car a few years back. There was a sudden slowdown on the highway, and we ended up just barely hitting the car in front of us. However, about 10 cars behind us, some other guy totally creamed the car in front of him. This was long enough after our minor hit that I was actually outside the car, checking the damage when this second collision occurred.

No way would a robot have been that oblivious.

[Delvo]

Don't get me wrong, I love to laugh at the ridiculous self-parking system that can't actually park itself in a real-life setting, and at "drivers" so pathetic as to need one, BUT...

A few years ago, I had an idea for a system to replace roads with something like monorails, scaled down to fit something the size of cars on them for private ownership & use by anyone, the same way we use cars on flat roads today. I called them "unirails" and the vehicles designed for them "unicars" and even "unitrucks" and "unibuses" and "univans". (The vehicle would need to be able to "switch" itself when the beam splits or two beams merge instead of having the beams move as they do for trains, but I had a solution for that.) I even started researching accident rates, infrastructure costs, hidden costs due to traffic, vehicle weight capacities, and so on, for both roads and monorail train systems, and started working on files to put on a website I would have created to advocate for switching from flat roads to unirails, presenting not only the above data (summarized) but also 2D and 3D images, some of them animated, of basic system design and function and what a world with unirails instead of roads might look like. I also had plans for a physical model in a spare room with remote-control-car-sized robots running around on hardened clay beams, to be able to shoot video of the merging & splitting and traffic flow at work.

A lot of the advantages I had in mind came from the fact that when the vehicle is only capable of going in the direction of the rail and that's the same direction as every other vehicle sharing that rail, driving is simplified so much that it could be handled by a control system only about as complex and failure-prone as a digital watch or 4/5-function calculator, plus some sensors that would only need to be the equivalent of the ones in LEGO(TM) toy-robot building kits, for detecting other vehicles and speed markers in the unirails. This would allow transportation to be safer, faster, more energy efficient, more consistent and smooth and reliable, and cheaper, using lighter, simpler, cheaper vehicles, all while allowing occupants to use the time sleeping, working, enjoying the view, and so on, without compromising safety because they're not driving. We'd also no longer have traffic jams negatively affecting the economy's overall productivity, in addition to the increase of productivity from work getting done by some people during the ride, and the fact that people weren't driving would take away the psychological stresses that some people experience from driving. The same effects would result from a better driver safety record, but I knew that would be impossible with human drivers and didn't see it coming any time soon for automatic ones.

Then I started to find out about recent advancements in electronic driving systems which are bringing us closer to the point that they could handle flat-road driving as we now know it, without needing the environment to be simplified for them as it would be on unirails. So the same goals would be achieved without my unirails. :):(

Of course, whenever these systems are inevitably ready to go, they will be much more complex and expensive than the LEGO toy robot parts that I had in mind for unirails, but people won't have a problem with that compared to other costs they're familiar with for electronic gadgets and the fancy features in cars, and the improved safety once enough auto-automobiles are out there will mean they can be made cheaper by removing, shrinking, or simplifying other parts. And a unirail network to replace roads would have some startup costs of its own and feel like a bigger adjustment for people's minds to make. So I decided that a world of auto-automobiles is bound to come, just a matter of time, and with that already covering some of the main incentives for unirails, the unirails were a lost cause that will never happen. I abandoned the idea of ever making a website about them and deleted the files. (I still think unirails instead of roads would be good for other reasons, but those will never be compelling enough to get it to actually happen.)

[arthwollipot]

...confusion over units for example....Just a minor nitpick, as your other points have been adequately covered, but if I recall correctly this particular catastrophic failure was a human error, not a hardware or software failure.

[GlennB]

But the other question of course is, do you hate some dogs so much that it´s a necessity for you to run them over? Stopping for both dogs and children seems like a pretty reasonable option to me. .... So the solution is: Just let robocar stop if there is an obstacle of sufficient size.

I love dogs, have had dogs for years and would love to avoid running one over, ever.
But I recall that recommended practice (in the UK at least) is not to risk an accident by braking suddenly for small animals. So - you (or robocar) know you have a motorbike up close behind you. Small mammal appears on the road. If it's a child you stop as best you can and risk injury to the biker. If it's a dog you don't, you favour the biker. That was my point.

[Stacy Head]

That takes all the fun out of driving. Why would anyone want a corvette that could drive itself?

[Darat]

I love dogs, have had dogs for years and would love to avoid running one over, ever.
But I recall that recommended practice (in the UK at least) is not to risk an accident by braking suddenly for small animals. So - you (or robocar) know you have a motorbike up close behind you. Small mammal appears on the road. If it's a child you stop as best you can and risk injury to the biker. If it's a dog you don't, you favour the biker. That was my point.







Seriously GlennB do yo you really think people see such an obstacle, recall what the highway code recommendation is and then act in such a way?

I have to say I know I don't, I would merely react.

[Rasmus]

Seriously GlennB do yo you really think people see such an obstacle, recall what the highway code recommendation is and then act in such a way?

I have to say I know I don't, I would merely react.

Yes, and you would simply slam the brakes.*

Whereas an automated car would

a) remember the motorcycle in the back and be aware of said motorcycles distance, speed and basic options
b) not slam the brakes as hard as possible, but just hard enough to not hit the obstacle - be it a child, dog, or cardboard box (I am usually afraid to hit something like the latter, too. You don't know what's in the box, after all!) - and possibly even just drive around it to leave more breaking space fr the trailing vehicle
c) warn any other nearby automated cars that there's something on the road you might not want to hit. So even if the motorcycle behind you isn't one of them, some other car just might not crash into the motor cycle.

* I know I would. Because when you are about to hit something, if you are thinking at all, it is not about the stupid motorcyclist right behind you who should keep a safe distance anyway. It's about whatever damage you or your car might take even if its just a cardboard box or some animal. Never mind the dreadful thought that you could have been wrong in your assessment of what exactly the obstacle is.

[SK.]

I love dogs, have had dogs for years and would love to avoid running one over, ever.
But I recall that recommended practice (in the UK at least) is not to risk an accident by braking suddenly for small animals. So - you (or robocar) know you have a motorbike up close behind you. Small mammal appears on the road. If it's a child you stop as best you can and risk injury to the biker. If it's a dog you don't, you favour the biker. That was my point.
A dog ran in front of my car travelling ~80km/h a few years ago. While I´m aware of the recommendations on what to do in that situation, I did an evasion maneuver into the opposing lane (no incoming traffic though). Unfortunately I still hit the dog. I´m not totally sure that what I did was the result of a careful consideration of all facts, although it of course is convenient to think it was.

I think it´s pure theory that a human will be able to tell dog from child and react correctly. In practice this might be true in some cases, in others the driver might do evasion maneuvers and lose control of their car. Stuff like that happens every day. The variance of the results of such an encounter is pretty large with humans. Cherrypicking the potential best result is not realistic IMHO.

While a robocar might be worse at classification (and that´s not a given) in some cases, it won´t be startled and will react much faster to obstacles. It will also be able to warn other vehicles via VANET ( http://en.wikipedia.org/wiki/VANET ) , so these get to know the planned evasion/braking trajectory and react accordingly (given good enough hardware, the whole evasion/braking process could even be optimized among all participating cars). Having non-robotic vehicles take part of course makes the problem harder. Still, even these will either be part of the VANET or (in the worst case) their state is at least known to robo-car and can be taken into account.
All this is likely to cut severly into the number of traffic victims. Yes, one can construct freak situations where a human might react better than the robocar. The key word to watch out for is "might". A robocar will be much more reliable and consistent in it´s behavior.

[GlennB]

Seriously GlennB do yo you really think people see such an obstacle, recall what the highway code recommendation is and then act in such a way?


No. Not in the slightest. In fact there's a fair chance I'd be unaware of the motorbike in the first place, then be found liable for damage to the biker/bike when it turned out I'd driven dangerously to save a dog. That's OK, even though I lose no-claims discount on my insurance. It was my fault.

But robo-car will have to make this judgement. Because it can (or so I'm told). Because I - as its owner - will say to the manufacturer "it was only a dog. Why did the car scream to a halt and hurt that biker? Now my insurance premiums rise 70% and it's your fault. I know my highway code and would not have driven like that". And there's not a thing the manufacturer can do to prove me a liar.

And - as mentioned a few times now - this is just one small example of judgement calls required by a robotic system that happened to spring to mind.

[Rasmus]

But robo-car will have to make this judgement. Because it can (or so I'm told). Because I - as its owner - will say to the manufacturer "it was only a dog. Why did the car scream to a halt and hurt that biker? Now my insurance premiums rise 70% and it's your fault. I know my highway code and would not have driven like that". And there's not a thing the manufacturer can do to prove me a liar.

Motorcycle crashes into your back because they were too fast to break on time: Not your fault. Never.

That aside, why do you think you will be held responsible as a driver for a car that was driven by an automated system?

[GlennB]

Motorcycle crashes into your back because they were too fast to break on time: Not your fault. Never.

Really? I believe it's a fallacy that the rearmost driver is always held to be at fault. And I'm told robocar will have much greater and safer braking capabilities than a car under human control. How does the biker know? Flashing lights that say "Warning ! This car may perform irrational manoeuvres" ? A day-glo sticker on the back?

That aside, why do you think you will be held responsible as a driver for a car that was driven by an automated system?

Because it was my choice to buy it, get into it and set it running.

Who else might be held responsible and pay the insurance claim? BMW? Toyota?

[Delvo]

Once all vehicles have this, the one behind you would hit the brakes practically simultaneously with yours (much faster than a human operator could do so), and you'd both slow down together...

[Pulvinar]

Really? I believe it's a fallacy that the rearmost driver is always held to be at fault. And I'm told robocar will have much greater and safer braking capabilities than a car under human control. How does the biker know? Flashing lights that say "Warning ! This car may perform irrational manoeuvres" ? A day-glo sticker on the back?

Smart drivers don't tailgate. Other human drivers today are more likely to perform irrational maneuvers.


Because it was my choice to buy it, get into it and set it running.

Who else might be held responsible and pay the insurance claim? BMW? Toyota?
Same as who would pay today if your brakes fail: it depends. It would be the same fallacy as you mention above to assume that the owner is always held to be at fault.

[arthwollipot]

Clearly we need to put the roads in tubes so that there are no unwanted pedestrians, human or otherwise.