Your thoughts on 'Driverless cars"

[Lil devils x_v1legacy]

UGH. I have to wonder if anyone who thinks this is safe in any way actually understands ANYTHING about programming.

snip

Examples aside, I think you are underestimating the complexity of the problem. A driverless car has been at the aim of very smart people for generations. That should give you an idea of the kind of effort here. It is the holy grail of AI. A machine that can interact with the real world, in a highly dynamic, highly complex continuous environment, without all the relevant information available, with split seconds decision making (as scary as a failure of a water or power station is, the systems running them have perfect information of all relevant variables at all times and very limited options available that they can outperform a human doing it).

I don't disagree with this, it's a huge problem, and we don't have the systems to deal with it properly yet[\i].Tech issues aside we are not going to wake up one day with driverless cars, what is far more likely to happen is that driver assists will increase over time, to the point where actually relinquishing control to the vehicle will be a minor point. There is already talk of vehicles which will automatically brake when a dangerous situation is detected, over time as the public accepts more and more of these assists, then public acceptance of driverless cars will become more palatable. sure the tech to safely have a driverless car on the roads is still a little off ( I'm not foolish enough to give a timescale), but if asked 10 years ago, we'd have thought they were farther off than they seem today.

You don't need to know all the variables of a system, but you do need to know enough, in our financial systems there is huge uncertainty, and generally that is a risk that is managed, a properly equipped driverless car can have far more information available to it than a human doing the driving. and there are simply loads of really really neat solutions to a lot of the problems. currently the biggie ( and yeah it is a biggie!), is the detection of unexpected objects on non-motorways ( children, animals, bikes etc ad nauseum ) but practical demonstrations of networked autonomous vehicles on motorways were first demonstrated a decade ago. Sure it's speculation, but it's not inconceivable. I could go into detail about specialist systems, object recognition systems & networked threat mapping as excellent examples of solutions to problems that seemed insurmountable only a few years ago.

@Lil Devils : I'm not disagreeing with you at all, there are aspects of our cyber security that are really REALLY scary, we could have nuclear explosions, planes dropping out the sky, cars blowing up, our entire economic system collapsing. With a backdrop of all that ( and with a car that can already be hacked and blown up ), I'm really not fussed if my car is slightly more autonomous, it's already at risk. You know that in a year or so's time, every single new car in Europe will be required to have a sim card and a dedicated network connection as standard, among other things it'll be able to automatically dial emergency services if it detects a crash. Imagine what a malicious agent could do with that.... the only way to not be worried about your car getting hacked is to either not drive, or have an old car.

Or you could have a custom car, as I do.

 

[Lil devils x_v1legacy]

It is a matter of availability. There are more cars then there are drones, and these cars are deployed at all times in every city. You are not talking about something that is of limited use, you are talking about what people rely on to get them to and from work, shopping, hospitals and every where else they go every day. BIG difference.
http://www.huffingtonpost.com/2012/06/29/drone-hacked-by-universit_n_1638100.html

THEY ARE HACKING CARS. Read the links above in the thread, however, you still have a driver currently. Since they have cars available for them to hack , they already do so, but luckily we still have drivers in control.

Although some doubt that even having a driver will help with the current vulnerabilities.
http://www.berkeleydailyplanet.com/issue/2013-08-09/article/41312?headline=On-the-Strange-Death-br-of-Michael-Hastings-br-Was-the-Reporter-Car-Hacked-or-Bombed---By-Gar-Smith
http://www.huffingtonpost.com/2013/06/24/michael-hastings-car-hacked_n_3492339.html

I dont doubt they are hacking cars, i just wonder if these people, apparently willing to cause death and trauma with their hacking, are too lazy to go down to the nearest military base and hijack a drone.

Sure its not massively easy, but we are discussing murderers, are they so lazy that they cant be assed to use a missile drone and would rather use a car? Are domestic terrorists so common but at the same time so incapable of actively looking for tools? You make a totally fair point and I agree with you, im just sorta confused by the attitudes of these hypothetical murderers.

It isn't even a matter of intent when it is most likely a 13 year old kid. We are discussing kids, teenagers who are bored and experimenting with life. They do not think about what can happen, they just think about " Can I do this?" This is due to how their brain forms during these years. The people most likely to do these things are the same ones who are most likely to hack into games and schools and bank computers.. They hack into what is around them simply because " They can". Oh.. The things I remember me and my friends doing during those years...

http://www.theguardian.com/science/2014/sep/05/teenage-brain-behaviour-prefrontal-cortex

 

[hermes]

UGH. I have to wonder if anyone who thinks this is safe in any way actually understands ANYTHING about programming.

snip

Examples aside, I think you are underestimating the complexity of the problem. A driverless car has been at the aim of very smart people for generations. That should give you an idea of the kind of effort here. It is the holy grail of AI. A machine that can interact with the real world, in a highly dynamic, highly complex continuous environment, without all the relevant information available, with split seconds decision making (as scary as a failure of a water or power station is, the systems running them have perfect information of all relevant variables at all times and very limited options available that they can outperform a human doing it).

I don't disagree with this, it's a huge problem, and we don't have the systems to deal with it properly yet[\i].Tech issues aside we are not going to wake up one day with driverless cars, what is far more likely to happen is that driver assists will increase over time, to the point where actually relinquishing control to the vehicle will be a minor point. There is already talk of vehicles which will automatically brake when a dangerous situation is detected, over time as the public accepts more and more of these assists, then public acceptance of driverless cars will become more palatable. sure the tech to safely have a driverless car on the roads is still a little off ( I'm not foolish enough to give a timescale), but if asked 10 years ago, we'd have thought they were farther off than they seem today.

You don't need to know all the variables of a system, but you do need to know enough, in our financial systems there is huge uncertainty, and generally that is a risk that is managed, a properly equipped driverless car can have far more information available to it than a human doing the driving. and there are simply loads of really really neat solutions to a lot of the problems. currently the biggie ( and yeah it is a biggie!), is the detection of unexpected objects on non-motorways ( children, animals, bikes etc ad nauseum ) but practical demonstrations of networked autonomous vehicles on motorways were first demonstrated a decade ago. Sure it's speculation, but it's not inconceivable. I could go into detail about specialist systems, object recognition systems & networked threat mapping as excellent examples of solutions to problems that seemed insurmountable only a few years ago...

Ok, that does sound a lot more likely than the picture the OP painted. That picture included a BBC note that seems to imply legislation was in place to allow 100% driveless cars (like the ones from Google) in the roads of UK in the near future.

The problems with driveless cars is not on the sensors, but the "thought process" behind it. Its called "Moravec's paradox"... Nowadays, a PC hooked to a camera can detect unexpected objects on the street, but ask them what they are and the most likely answer (after some processing) will be "a round object out of place". Ask a person and the immediate answer it will be "a basket ball"... Those thought processes that allowed our ancestors to distinguish a tree branch from a hunting lion are the kind of things our brain is extremely efficient at doing, and the kind of things a PC is not... we are able to do that for an entire room with just a glimpse. And those kind of split-second decision making and pattern recognition are fundamental to driving.

Networked autonomous vehicles cop-out of most of the problem because of one reason: they drive in a heavily controlled environment, which assumes everything there is controlled by the same system. Many don't even have obstacles unless they are programmed into. A real life application would have to run in an extremely dynamic, extremely continuous environment, where many (if not all) the other relevant objects can not be preloaded, networked, communicated with or even trust...

 

[Jesterscup]

snip

snip

I wouldn't make assumptions about who the hackers 'are' or 'why', they can be anything from a script kiddy, a disgruntled worker, organised electronic crime syndicates ( gotta love the Russians, man they do the most awesome electronic financial crimes), a commercially employed corporate espionage crew ( these really exist! ), to state sponsored teams. we've seen with lulzsec what the script kiddies can do, and from stuxnet what nations can do... it really is across the board. Why is just as complicated, from because you can, to money, fame, power, terrorism ( stuxnet can easily be classed as an act of either war or terrorism depending on how you view it).

 

[Killclaw Kilrathi]

It will make driving much safer, until it doesn't. There will inevitably be a flaw, something overlooked. We are imperfect creatures constantly trying to create perfect systems. When we find the flaw, when it manifests in the driver-less cars, it will kill a lot of people in a short period of time until we fix it. Then we'll be safe again until the next crisis. Overall, the cars will make driving safer than walking in the park. We'll just have to worry about the next big bug causing the occasional cluster of fatalities. Otherwise, fantastic.

Good point, there is always the possibility of design flaws. I would argue though that the current system is much more dangerous. A system designed by flawed humans in which "driver" behaviour is coded, tested and re-tested may yield an unexpected bug in a very unique situation in one or more car models, but that's still miles better than putting the flawed humans behind the wheels themselves for every single car journey. An error by a developer in the first model will likely be found and fixed before it becomes a problem, but every single error in the second model could cause a fatal situation. The latter also has a lot more humans, so a lot more potential for errors.

 

[Lil devils x_v1legacy]

snip

snip

I wouldn't make assumptions about who the hackers 'are' or 'why', they can be anything from a script kiddy, a disgruntled worker, organised electronic crime syndicates ( gotta love the Russians, man they do the most awesome electronic financial crimes), a commercially employed corporate espionage crew ( these really exist! ), to state sponsored teams. we've seen with lulzsec what the script kiddies can do, and from stuxnet what nations can do... it really is across the board. Why is just as complicated, from because you can, to money, fame, power, terrorism ( stuxnet can easily be classed as an act of either war or terrorism depending on how you view it).

I am talking about the hackers I know. I grew up with hackers that now work for our government and THIS is what they did as kids. MY best friend had the FBI come to his house at the age of 13 due to him hacking a bank and luckily instead of jailing him he received a scholarship to MIT and works for the Department of defense. This is not making an assumption, did you read the defcon kids article above? I could not even begin to tell you the number of kids I have known in games that were keylogging others and making bots. Of course there are MANY other hackers out there, but to assume that they all have malicious intentions is not even the case much of the time when they are just kids doing what kids do.

 

[Lil devils x_v1legacy]

It will make driving much safer, until it doesn't. There will inevitably be a flaw, something overlooked. We are imperfect creatures constantly trying to create perfect systems. When we find the flaw, when it manifests in the driver-less cars, it will kill a lot of people in a short period of time until we fix it. Then we'll be safe again until the next crisis. Overall, the cars will make driving safer than walking in the park. We'll just have to worry about the next big bug causing the occasional cluster of fatalities. Otherwise, fantastic.

Good point, there is always the possibility of design flaws. I would argue though that the current system is much more dangerous. A system designed by flawed humans in which "driver" behaviour is coded, tested and re-tested may yield an unexpected bug in a very unique situation in one or more car models, but that's still miles better than putting the flawed humans behind the wheels themselves for every single car journey. An error by a developer in the first model will likely be found and fixed before it becomes a problem, but every single error in the second model could cause a fatal situation. The latter also has a lot more humans, so a lot more potential for errors.

The difference with the current system though and an integrated system where the cars work together is they are all separate now using separate technology and if one car or manufacturer has a defect, it only takes those vehicles out of action, not all of them. In a system where they are forced to work together, you create a whole new level of vulnerability as that would affect ALL the cars in the system and not just one here or there. One error could cause ALL the cars to fail vs one error causing limited number to fail.

 

[Jesterscup]

The problems with driveless cars is not on the sensors, but the "thought process" behind it. Its called "Moravec's paradox"... Nowadays, a PC hooked to a camera can detect unexpected objects on the street, but ask them what they are and the most likely answer (after some processing) will be "a round object out of place". Ask a person and the immediate answer it will be "a basket ball"... Those thought processes that allowed our ancestors to distinguish a tree branch from a hunting lion are the kind of things our brain is extremely efficient at doing, and the kind of things a PC is not... we are able to do that for an entire room with just a glimpse. And those kind of split-second decision making and pattern recognition are fundamental to driving.

You are quite right, and this is THE big issue with autonomous cars. But you'd be surprised with what you can do in a split second, you've already got the objects size, speed and direction, and from that alone you've solved a huge subset of situations. yeah a few years ago this was near insurmountable, then the kinect came out, and we now can tell the outline of an object fairly easily.

A fallen tree blocking the road does not need to be differentiated from a stopped truck, just known that it's big, not moving and blocking the way. If we assume that these cars are networked and sharing sensor data, your car in theory can know enough about a huge bunch of objects in it's potential future, without sensor data of it's own. Take a cyclist ( or person on horse, it's pretty easy to make the assumption that it counts as a vehicle, it's big enough, and moving in a certain reasonably predictable way, again you don't need to know it's a bike ( motor or pedal ) or horse, you just need to know what potential actions/reactions should be considered.

Now the trickier part of this is detecting ( for example, there are others) a child/pet/ball from each other. How would the car behave differently in these situations. But lets look backwards, this is going to help some.

if ball = can I stop/react safely ( considering the other vehicles nearby)
if pet = considering the potential size, can I stop/react safely ( considering the potential impact to the passengers then to other vehicles nearby, then to the animal) -
if child = can I stop/react safely ( considering the safety of the child and potential impact to the passengers, then to other vehicles nearby)

it actually helps that the actual reactions in any situation are kinda limited ( and 'mostly' decided rather easily too ), we know the other vehicles nearby, and the state of our vehicle, and the potential options available, the question is in fact how to we prioritise our potential reactions. Ball and pet kinda fold into each other, and into a 'is it big enough" category. when it comes to humans ( particularly children ) it gets trickier, because our priorities change. Yet as soon as there is the chance of impact we've only got a small pool of possible actions,
A. Do I need to act ( change cars current velocity or direction)?
B. Can I react safely ( without any collisions, & within safe stopping/reacting parameters - this is helped with nearby networked vehicles too)?
C. what are the potential collision options ( obv. a list here..... basically what else if B has no valid results)?

So yeah, still big, but suddenly from our a huge mass of data and options we get down to "how big is it" and what can we do, sure sometimes it'll be a ball, but is it a beach ball or a bowling ball? but so much has been eliminated for what needs to be worked out. Once we start looking at it this way, we can already program autonomous cars that are safer ( in theory, still years of work though ).

I'm not saying this isn't a huge issue, but there is some really nice stuff being worked on, and I know the above is an oversimplification. But it's a pretty good example about how something incredibly complex the way we do it in the human brain, can be simplified down, often in really surprising ways. you'd be surprised how the release of the kinect changed things, before it edge sensing of objects was thought to be decades away, and yet it's solution had been sitting there awaiting being applied for years. Now we can make accurate decisions from a limited pool of actions, based on knowledge that we do have, and can gather quickly and process easily.
child-ball-pet is an issue ( actually THE big one ), because it's a case where the system CANNOT be wrong, car doesn't need to know if is bike or is horse, because we can confidently state that its a "big thing in road", but car does need to know what "small thing near/in road" is as the result changes the cars potential reaction.

Yeah we are not there yet, but personally I'm pretty optimistic that we are kinda far along.

Disclaimer - I don't actually think that we'll use a system that detects the shape can determines 'what a thing is' you're more likely to get 'fixed shape (ball) this size, estimated momentum, estimated mass, chance of damage on contact. for a ball, and for humans/animals "irregular object, probably squishy", living things are so brilliantly random we can easily guess that something is alive if it's shape/speed change and are irregular . So not fully determining what something is, just enough to make a good choice out of a limited pot.

 

[Lil devils x_v1legacy]

The problems with driveless cars is not on the sensors, but the "thought process" behind it. Its called "Moravec's paradox"... Nowadays, a PC hooked to a camera can detect unexpected objects on the street, but ask them what they are and the most likely answer (after some processing) will be "a round object out of place". Ask a person and the immediate answer it will be "a basket ball"... Those thought processes that allowed our ancestors to distinguish a tree branch from a hunting lion are the kind of things our brain is extremely efficient at doing, and the kind of things a PC is not... we are able to do that for an entire room with just a glimpse. And those kind of split-second decision making and pattern recognition are fundamental to driving.

You are quite right, and this is THE big issue with autonomous cars. But you'd be surprised with what you can do in a split second, you've already got the objects size, speed and direction, and from that alone you've solved a huge subset of situations. yeah a few years ago this was near insurmountable, then the kinect came out, and we now can tell the outline of an object fairly easily.

A fallen tree blocking the road does not need to be differentiated from a stopped truck, just known that it's big, not moving and blocking the way. If we assume that these cars are networked and sharing sensor data, your car in theory can know enough about a huge bunch of objects in it's potential future, without sensor data of it's own. Take a cyclist ( or person on horse, it's pretty easy to make the assumption that it counts as a vehicle, it's big enough, and moving in a certain reasonably predictable way, again you don't need to know it's a bike ( motor or pedal ) or horse, you just need to know what potential actions/reactions should be considered.

Now the trickier part of this is detecting ( for example, there are others) a child/pet/ball from each other. How would the car behave differently in these situations. But lets look backwards, this is going to help some.

if ball = can I stop/react safely ( considering the other vehicles nearby)
if pet = considering the potential size, can I stop/react safely ( considering the potential impact to the passengers then to other vehicles nearby, then to the animal) -
if child = can I stop/react safely ( considering the safety of the child and potential impact to the passengers, then to other vehicles nearby)

it actually helps that the actual reactions in any situation are kinda limited ( and 'mostly' decided rather easily too ), we know the other vehicles nearby, and the state of our vehicle, and the potential options available, the question is in fact how to we prioritise our potential reactions. Ball and pet kinda fold into each other, and into a 'is it big enough" category. when it comes to humans ( particularly children ) it gets trickier, because our priorities change. Yet as soon as there is the chance of impact we've only got a small pool of possible actions,
A. Do I need to act ( change cars current velocity or direction)?
B. Can I react safely ( without any collisions, & within safe stopping/reacting parameters - this is helped with nearby networked vehicles too)?
C. what are the potential collision options ( obv. a list here..... basically what else if B has no valid results)?

So yeah, still big, but suddenly from our a huge mass of data and options we get down to "how big is it" and what can we do, sure sometimes it'll be a ball, but is it a beach ball or a bowling ball? but so much has been eliminated for what needs to be worked out. Once we start looking at it this way, we can already program autonomous cars that are safer ( in theory, still years of work though ).

I'm not saying this isn't a huge issue, but there is some really nice stuff being worked on, and I know the above is an oversimplification. But it's a pretty good example about how something incredibly complex the way we do it in the human brain, can be simplified down, often in really surprising ways. you'd be surprised how the release of the kinect changed things, before it edge sensing of objects was thought to be decades away, and yet it's solution had been sitting there awaiting being applied for years. Now we can make accurate decisions from a limited pool of actions, based on knowledge that we do have, and can gather quickly and process easily.
child-ball-pet is an issue ( actually THE big one ), because it's a case where the system CANNOT be wrong, car doesn't need to know if is bike or is horse, because we can confidently state that its a "big thing in road", but car does need to know what "small thing near/in road" is as the result changes the cars potential reaction.

Yeah we are not there yet, but personally I'm pretty optimistic that we are kinda far along.

Disclaimer - I don't actually think that we'll use a system that detects the shape can determines 'what a thing is' you're more likely to get 'fixed shape (ball) this size, estimated momentum, estimated mass, chance of damage on contact. for a ball, and for humans/animals "irregular object, probably squishy", living things are so brilliantly random we can easily guess that something is alive if it's shape/speed change and are irregular . So not fully determining what something is, just enough to make a good choice out of a limited pot.

In addition, when we are discussing object detection and identification systems and making split second decisions, you have to account for the human decision of whether to sacrifice yourself, your car and other objects to save the life of the child in the street. This actually happened on the street near mine, they chose to hit the car in the driveway and the front of the house over hitting the child on the street. They chose to hit much larger and more dangerous objects to save the life of a toddler on a tricycle. No one besides the driver was injured and due to this decision both the driver and child are alive today.

I am not sure this will ever be possible with our current systems even if they do improve object detection and identification as they lack the human emotion and ability to self sacrifice in order to save others.

 

[Nukekitten]

In addition, when we are discussing object detection and identification systems and making split second decisions, you have to account for the human decision of whether to sacrifice yourself, your car and other objects to save the life of the child in the street. This actually happened on the street near mine, they chose to hit the car in the driveway and the front of the house over hitting the child on the street. They chose to hit much larger and more dangerous objects to save the life of a toddler on a tricycle. No one besides the driver was injured and due to this decision both the driver and child are alive today.

I am not sure this will ever be possible with our current systems even if they do improve object detection and identification as they lack the human emotion and ability to self sacrifice in order to save others.

You could specify that people beneath a certain size have a higher value if that's what you're concerned about. Computers certainly have the ability to sacrifice themselves for an objective, cruise missiles do it fairly regularly. The system doesn't consider its own survival unless you tell it to - and even then you can tell it to do so only up to a point.

 

[Lil devils x_v1legacy]

In addition, when we are discussing object detection and identification systems and making split second decisions, you have to account for the human decision of whether to sacrifice yourself, your car and other objects to save the life of the child in the street. This actually happened on the street near mine, they chose to hit the car in the driveway and the front of the house over hitting the child on the street. They chose to hit much larger and more dangerous objects to save the life of a toddler on a tricycle. No one besides the driver was injured and due to this decision both the driver and child are alive today.

I am not sure this will ever be possible with our current systems even if they do improve object detection and identification as they lack the human emotion and ability to self sacrifice in order to save others.

You could specify that people beneath a certain size have a higher value if that's what you're concerned about. Computers certainly have the ability to sacrifice themselves for an objective, cruise missiles do it fairly regularly. The system doesn't consider its own survival unless you tell it to - and even then you can tell it to do so only up to a point.

The issue of course is, we do not have a system that can do that nor, will we have one for the foreseeable future. We do not even have a system that can tell the difference between a child and an animal currently.

On Halloween here it is common to see all sorts of shapes in the road.

 

[Gitty101]

Well, it'd be awesome to be able to get myself to pubs and back again under my own robotic powered steam.

 

[Nukekitten]

The issue of course is, we do not have a system that can do that nor, will we have one for the foreseeable future. We do not even have a system that can tell the difference between a child and an animal currently.

Uhm. This seems like a different discussion to be having than the one in the post I responded to which, as far as I understood it, was about whether even if we improved object recognition computers would be capable of making those sorts of decisions. I took as one of your assumptions that a computer would be able to tell the difference between children and other things reasonably reliably.

If the question is whether it would be possible to make a computer tell the difference between children and different objects... well, I feel like it is. There's a lot of work going on with surveillance towards face recognition, gait recognition and the like. It seems like the question of 'Which person is this?' is a harder one to answer than 'is this a dog or a person?' and since the former question has had significant advances over the last twenty years I'm not convinced that it's a not in the foreseeable (timespan?) future type of hard problem.

edit:
If we're talking about the problem generally, I feel it's also worth mentioning that it's extremely rare for someone to just run out into the road without any warning. When you're driving you notice things on the pavement and if sight's restricted, space is restricted, or you see them moving towards you - you slow down in case they do step out without looking.

At 15-20mph, which I'd consider a reasonable speed for driving a short distance from a line of parked cars on the side of the road where you might miss seeing someone who's about to step out into traffic, your stopping distance is very small, and within the time that stopping distance allows I'm not sure that a human could adequately categorise a threat as someone in a Halloween costume vs a real dog and decide whether to hit it or not. We're talking 3 car lengths sort of distance, half of which the driving books say is just your reaction time. It seems to me that in that sort of time you'd just slam on the brakes and trust the ABS to stop you rather than spending the thinking time.

 

[michael87cn]

as a carless pedestrian (11 years running) I am in favor of people losing their abiliy to control the vehicles themselves. You know how many times I get nearly run over by people running red lights, not yielding to me when the light gives me 18 seconds to cross the street, honking at me when I walk across a pedestrian crossing section like I'm the one breaking the law...? You know how many people I see talking on their phone driving with one hand not even noticing me or driving without a seatbelt with 3 kids in the back seat playing?

People. Are. Stupid.

The most stressful part of my day is my 15 minute walk to and from work as I see a literal sea of cars rush around in a crazed, rushed manner. What's the fucking rush? You don't have to walk you lazy bastards.... Why are you in such a hurry? None of you deserve the luxury of that vehicle. Unfortunately life doesn't operate on what we deserve.... but fortunately we may see direct control of vehicles be taken away so that everyone can travel safely and the stupid hicks that rev their engines honk their horns, speed, or basically drive monster trucks on the road put to an end. Driving is not a game and a car is not a toy.

Grow up....

Also, since the worlds supply of fuel is projected to run out by like 2055 anyway we can all rest safely knowing that either the world is about to end or the 'fun' of driving a car 'ourselves' is about to end NO MATTER WHAT.

 

[Repomancer]

My concern doesn't include hackers trying to prove how 1337 they are, undetected sensor failure, software bugs, or shape misinterpretation -- those can all be addressed and certainly will be before driverless becomes the accepted norm. I'd worry more about some murderous teenage nutjob standing on the overpass with a cinderblock. A driverless car might be able to react to the falling object (sorry, passengers + whoever was in the adjacent lane), but an alert (human) driver could avoid any trouble far in advance.

Sure, that's not the sort of thing you see in the average million miles of driving, but such things do happen. I hate to use the term "AI", because intelligent is what computers aren't, but until a car's AI can divine the intent of that shape on the overpass I'll be happier with a human behind the wheel.

I was once on the 405 in Los Angeles heading from Santa Monica up the hill to the Valley, and saw a car hit by a golf ball -- it smashed the windshield and caused a huge wreck, which I mercifully avoided. Some loon in the yard of one of the Nice Houses? that line the mountainside was driving balls onto the freeway, for fun and to presumably see how many people he/she could kill. (Bad example, because a human couldn't deal with it any better than software could, but it got me to thinkin'. Ihopeyougetmydrift.) I guess my point is that I trust what driverless cars will be able to deal with on the road. Things happening off the road, I'm not so sure.

 

[Lil devils x_v1legacy]

The issue of course is, we do not have a system that can do that nor, will we have one for the foreseeable future. We do not even have a system that can tell the difference between a child and an animal currently.

Uhm. This seems like a different discussion to be having than the one in the post I responded to which, as far as I understood it, was about whether even if we improved object recognition computers would be capable of making those sorts of decisions. I took as one of your assumptions that a computer would be able to tell the difference between children and other things reasonably reliably.

If the question is whether it would be possible to make a computer tell the difference between children and different objects... well, I feel like it is. There's a lot of work going on with surveillance towards face recognition, gait recognition and the like. It seems like the question of 'Which person is this?' is a harder one to answer than 'is this a dog or a person?' and since the former question has had significant advances over the last twenty years I'm not convinced that it's a not in the foreseeable (timespan?) future type of hard problem.

edit:
If we're talking about the problem generally, I feel it's also worth mentioning that it's extremely rare for someone to just run out into the road without any warning. When you're driving you notice things on the pavement and if sight's restricted, space is restricted, or you see them moving towards you - you slow down in case they do step out without looking.

At 15-20mph, which I'd consider a reasonable speed for driving a short distance from a line of parked cars on the side of the road where you might miss seeing someone who's about to step out into traffic, your stopping distance is very small, and within the time that stopping distance allows I'm not sure that a human could adequately categorise a threat as someone in a Halloween costume vs a real dog and decide whether to hit it or not. We're talking 3 car lengths sort of distance, half of which the driving books say is just your reaction time. It seems to me that in that sort of time you'd just slam on the brakes and trust the ABS to stop you rather than spending the thinking time.

I think you are assuming that neighborhoods where children are likely to run out in the street also have lower speed limits, and in some places this is correct, however, the speed limit out in front of my parents home where I grew up is 55mph. Every home on our street had children, and kids were frequently in the road. The street is ALSO lined with trees along the fence line greatly reducing visibility on the sides. Most speed limits where I reside are 55- 70mph, and yes, humans have been able to avoid running over the children here.

In order to avoid hitting things in the road, you frequently have to maneuver off of the road, not just stop in order to avoid hitting them and you have to make a judgement call as to what it is you are hitting would be better than what you could have hit. Sometimes that is deciding between hitting a cow or a horse with a rider, or hitting child or a parked car or a house.

I am not sure why you think it is rare to have people run out into the street, it really is not. It happens quite frequently, and sadly one of my friends from school also died that way at the age of 20.

 

[Nukekitten]

I think you are assuming that neighborhoods where children are likely to run out in the street also have lower speed limits, and in some places this is correct, however, the speed limit out in front of my parents home where I grew up is 55mph. Every home on our street had children, and kids were frequently in the road. The street is ALSO lined with trees along the fence line greatly reducing visibility on the sides. Most speed limits where I reside are 55- 70mph, and yes, humans have been able to avoid running over the children here.

In order to avoid hitting things in the road, you frequently have to maneuver off of the road, not just stop in order to avoid hitting them and you have to make a judgement call as to what it is you are hitting would be better than what you could have hit. Sometimes that is deciding between hitting a cow or a horse with a rider, or hitting child or a parked car or a house.

I am not sure why you think it is rare to have people run out into the street, it really is not. It happens quite frequently, and sadly one of my friends from school also died that way at the age of 20.

Driver training where I am is probably quite different to driver training where you are judging from that. If I'd done 55mph through a place with bad sight lines or dense civilian traffic, regardless of the posted speed limit, I'd have failed my test. Seriously, there are areas around here where the posted limit is the national one - 70mph on that type of road - where people aren't likely to run out into the road, and I've not seen anyone do more than 40mph through them, and lower where they're approaching areas of limited visibility, because the road doesn't support speeds greater than that safely.

It was like one of the basic rules for driving when I was learning - one of the things that if nothing else I was repeatedly told I had to get hammered into my head: "Always drive in such a manner that you may stop within the space that you can see to be clear on your side of the road."

So, my response to that is basically, 'Regardless of whether you're a computer or a person: Don't drive at 55mph through areas people are liable to run out into the road! The speed limit is an absolute maximum, not a target.' I mean it sucks that driving is like that where you are, and maybe you're obliged to drive in that manner because that's what everyone around you is doing and it would be more dangerous not to, but I don't think that it reflects on the general feasibility of automated cars that drivers in your area behave badly.

Say someone does swerve to avoid a collision and hits something else instead: It's a good reaction, but it's dreadful planning. What happens if the thing they swerve and hit is another person? What if their option is to hit the child or run into oncoming traffic where the oncoming vehicle has a bunch of kids in it? What if...

They still hit something. They were no longer in total control of the course of the vehicle of they'd have elected not to. And you're promoting that sort of reaction as a good thing but I'm seeing that sort of driving as strongly contributing to needing to take that action in the first place. The reaction just manages to correct for the mess sufficiently well that, thank god, no-one necessarily dies. In the words of a court case I attended in the last few months, 'You were clearly driving far too fast to react safely to the unexpected.'

That's not good driving to my mind, and if we're going to make driverless cars behave like that, then yeah maybe don't bother with them. You shouldn't be using technology as a bad band aid on bad practice. Fix the bad practice instead, either in software or in law.

 

[Lil devils x_v1legacy]

I think you are assuming that neighborhoods where children are likely to run out in the street also have lower speed limits, and in some places this is correct, however, the speed limit out in front of my parents home where I grew up is 55mph. Every home on our street had children, and kids were frequently in the road. The street is ALSO lined with trees along the fence line greatly reducing visibility on the sides. Most speed limits where I reside are 55- 70mph, and yes, humans have been able to avoid running over the children here.

In order to avoid hitting things in the road, you frequently have to maneuver off of the road, not just stop in order to avoid hitting them and you have to make a judgement call as to what it is you are hitting would be better than what you could have hit. Sometimes that is deciding between hitting a cow or a horse with a rider, or hitting child or a parked car or a house.

I am not sure why you think it is rare to have people run out into the street, it really is not. It happens quite frequently, and sadly one of my friends from school also died that way at the age of 20.

Driver training where I am is probably quite different to driver training where you are judging from that. If I'd done 55mph through a place with bad sight lines or dense civilian traffic, regardless of the posted speed limit, I'd have failed my test. Seriously, there are areas around here where the posted limit is the national one - 70mph on that type of road - where people aren't likely to run out into the road, and I've not seen anyone do more than 40mph through them, and lower where they're approaching areas of limited visibility, because the road doesn't support speeds greater than that safely.

It was like one of the basic rules for driving when I was learning - one of the things that if nothing else I was repeatedly told I had to get hammered into my head: "Always drive in such a manner that you may stop within the space that you can see to be clear on your side of the road."

So, my response to that is basically, 'Regardless of whether you're a computer or a person: Don't drive at 55mph through areas people are liable to run out into the road! The speed limit is an absolute maximum, not a target.' I mean it sucks that driving is like that where you are, and maybe you're obliged to drive in that manner because that's what everyone around you is doing and it would be more dangerous not to, but I don't think that it reflects on the general feasibility of automated cars that drivers in your area behave badly.

Say someone does swerve to avoid a collision and hits something else instead: It's a good reaction, but it's dreadful planning. What happens if the thing they swerve and hit is another person? What if their option is to hit the child or run into oncoming traffic where the oncoming vehicle has a bunch of kids in it? What if...

They still hit something. They were no longer in total control of the course of the vehicle of they'd have elected not to. And you're promoting that sort of reaction as a good thing but I'm seeing that sort of driving as strongly contributing to needing to take that action in the first place. The reaction just manages to correct for the mess sufficiently well that, thank god, no-one necessarily dies. In the words of a court case I attended in the last few months, 'You were clearly driving far too fast to react safely to the unexpected.'

That's not good driving to my mind, and if we're going to make driverless cars behave like that, then yeah maybe don't bother with them. You shouldn't be using technology as a bad band aid on bad practice. Fix the bad practice instead, either in software or in law.

You are giving terribly dangerous driving advice, as you would get people killed here doing that. You are more likely to cause an accident if you drive too slow rather than too fast, and you will rightfully get a ticket here for doing so. Most people here drive 70+mph. It is less likely to cause an accident if you are keeping up with the flow of traffic. Even going 75-80 mph on highway 80 here and a ladder rolled across multiple lanes, all of the vehicles were able to maneuver around it without getting anyone harmed at high speeds. We have to expect vehicles to be able to do so here, or they are dangerous. Some speed limits in Texas are 85mph and we have to have vehicles that can operate at these speeds. The point of that reaction is to be able to identify what is on the side of the road and maneuver your vehicle accordingly. You DO need to be able to tell whether it is a bunch of kids is what I am saying, we as humans ALREADY do. I do not think anyone should be behind the wheel that cannot operate at those speeds. You seem to think this is fast, but I have operated vehicles with full control at much higher speeds( my Father designed, built and raced Pro stock cars).
You do not " plan" what is on the road, you have to be able to react accordingly REGARDLESS of what is on the road. Things fly out of vehicles all the time, and we have been driving at high speeds with this happening for many years now already. Of course we should expect all the vehicles on the road to be able to do the same, otherwise they should not be on the road in the first place.

http://www.carinsurance.com/Articles/driving-too-slow-tickets-insurance.aspx
http://www.dailymail.co.uk/news/article-2016721/Slow-drivers-dangerous-roads-cause-crashes.html
http://www.autoblog.com/2012/10/25/texas-85-mph-toll-road-opens-weeks-ahead-of-schedule/

 

[Nukekitten]

You are giving terribly dangerous driving advice, as you would get people killed here doing that. You are more likely to cause an accident if you drive too slow rather than too fast, and you will rightfully get a ticket here for doing so. Most people here drive 70+mph. It is less likely to cause an accident if you are keeping up with the flow of traffic.

"it sucks that driving is like that where you are, and maybe you're obliged to drive in that manner because that's what everyone around you is doing and it would be more dangerous not to, but I don't think that it reflects on the general feasibility of automated cars that drivers in your area behave badly."

Even going 75-80 mph on highway 80 here and a ladder rolled across multiple lanes, all of the vehicles were able to maneuver around it without getting anyone harmed at high speeds. We have to expect vehicles to be able to do so here, or they are dangerous. Some speed limits in Texas are 85mph and we have to have vehicles that can operate at these speeds. The point of that reaction is to be able to identify what is on the side of the road and maneuver your vehicle accordingly. You DO need to be able to tell whether it is a bunch of kids is what I am saying, we as humans ALREADY do. I do not think anyone should be behind the wheel that cannot operate at those speeds. You seem to think this is fast, but I have operated vehicles with full control at much higher speeds( my Father designed, built and raced Pro stock cars).
You do not " plan" what is on the road, you have to be able to react accordingly REGARDLESS of what is on the road. Things fly out of vehicles all the time, and we have been driving at high speeds with this happening for many years now already. Of course we should expect all the vehicles on the road to be able to do the same, otherwise they should not be on the road in the first place.

http://www.carinsurance.com/Articles/driving-too-slow-tickets-insurance.aspx
http://www.dailymail.co.uk/news/article-2016721/Slow-drivers-dangerous-roads-cause-crashes.html
http://www.autoblog.com/2012/10/25/texas-85-mph-toll-road-opens-weeks-ahead-of-schedule/

I think the definition of fast is defined by the context you're in. I've driven cars well in excess of a hundred miles an hour on public roads on blue light runs before without feeling that it was unsafe. The only way to do that reasonably safely is to plan your drive. Planning your drive is of course different than planning what's going to be on the road. No-one can plan what's going to be on the road, we don't have perfect knowledge. You plan so as to increase the time you have to react and options to change speed and direction. It's not going into a corner at silly speeds when there might very well be an accident on the other side or someone stepping into the road. It's moving so as to increase your sight lines and safety. It's maintaining an escape route. It's maintaining a reasonable separation from the car in front. It's noticing when someone's going to emerge from a junction and staying out so that they can do so safely. It's building enough space and time into your drive so that if something happens your options aren't 'Kill someone or slam into something.'

Maybe that's not viable where you are. That sucks for you. I don't see it as a general argument against automated cars but as an argument that your traffic laws and driver training are terrible. You don't need to be able to classify whether something's a dog or a child to know it's probably not going to be conductive to continuing your drive in a safe manner to slam into it at 55mph or take a poorly controlled swerve into a nearby object.

And frankly the statement that driving at those speeds in the context you do so is safe, considering the ridiculous number of road fatalities and injuries in America? I don't buy it. You may have no better options, I won't dispute that one way or the other. But I'm not convinced it's safe - and statements about kids stepping out in front of traffic and things flying off of cars that make you swerve at those sorts of speeds, and that you personally know one person who's crashed from that sort of thing and another who's died, does no favours in that consideration.

I've heard nothing on this point to convince me that driverless cars should fulfil the standards of Texan traffic as much as I've heard things that make me believe the standards of Texan traffic shouldn't be adopted. If that's the way it's going to be in Texas, just don't use driverless cars in Texas until the technology's more mature.

 

[Dirty Hipsters]

They aren't why would anyone think they were?
They already have been hacked, and people have been complaining about those as well. I am not sure what you are going on about.

http://rt.com/usa/texas-professor-drone-hacking-249/
http://www.cnn.com/2009/US/12/17/drone.video.hacked/

The sentiment seems to be that if cars were hackable people would commit atrocities with them, drive everyone into a wall, cut breaks ect, that this would be an inevitable consequence.

I ask that if we have pilotless flying missile launchers and THESE are hackable why isnt there mass terror and hysteria as the hackers do these things already with far more potent tools?

Hackers can already commit mass murder apparently, just turn a drone around and blow up cities. But they havnt, either UAV's cant be hacked by these back garden hackers or they dont want to commit domestic terrorism. Either way theres a demonstration that theres simply not been an abuse of drones to kill people by hacking them. Why would it be different for cars, arguably a far less efficient method?

Honest question, if you think there would be hacker car attacks why hasnt there been a hacker drone attack? Answer that one question tbh since thats the crux of it.

It's a lot easier to get a million automated cars to crash than it is to get a single drone to blow something up.

What do you have to do to get a drone to kill thousands of people? You have to hack the drone, then take direct control of its systems, take the time to fly it somewhere with a lot of people (which could take several hours during which you would need to be in control of all the drone's systems), be able to aim the missiles, and then fire them.

What does it take to get millions of driver less cars to crash? Tell them all to turn left.