Will driverless cars save lives? (more than 130 collisions)
Discussion
FiF said:
Probably a good job they don't use Teslas.
From a recent road test of the latest Model 3. "There’s also a lane-keeping assistant, which Tesla claims is a self-driving system (it isn’t) and in practice the camera-based system misses rather too much for comfort. It failed to spot a horse right in front of the bonnet, but did spot a wheelie bin beside it, so it’s a neigh from the equine population, then.
It also failed to detect a pedestrian in a dark coat and an unlit van parked half on the pavement and was preparing to drive straight into it. "
Yet they still want to charge £6k+ to prepare the "self driving system".
The lane-keeping assistant is part of Autopilot which all Teslas now comes with free of charge. Autopilot is basically the name they market their cruise control under. What Teslas call (future) full-self driving is a separate package which cost a lot and isn't worth it from my point of view.From a recent road test of the latest Model 3. "There’s also a lane-keeping assistant, which Tesla claims is a self-driving system (it isn’t) and in practice the camera-based system misses rather too much for comfort. It failed to spot a horse right in front of the bonnet, but did spot a wheelie bin beside it, so it’s a neigh from the equine population, then.
It also failed to detect a pedestrian in a dark coat and an unlit van parked half on the pavement and was preparing to drive straight into it. "
Yet they still want to charge £6k+ to prepare the "self driving system".
The idea that you can't write stable, predictable software is bilge. There's tons of it around keeping machines running. Aircraft are a pretty good example of what can be achieved. They are properly complicated systems running huge amounts of software and they are highly fault tolerant.
But that's got bugger all to do with the challenge of making a car drive itself down the road. The car's challenge is an AI problem; sensing and understanding what's going on around it and where the road goes.
At the moment this is a much more complicated problem than it need be because the car has to cope with a system designed entirely for humans. You could easily get cars to start broadcasting their intent to neighbouring vehicles so that traffic could become self-organising and far more cooperative than humans could achieve. You could get digital information from the road that is far more precise and rich than just a sign that says "bend ahead". And of course you can give cars sensors that see through fog, see in the infrared and "see around corners" by sharing data with nearby vehicles.
It's a classic AI problem and doesn't require a huge leap forward in computer science, just refinement of current engineering supported by the usual incremental reduction in the cost of number crunching.
But that's got bugger all to do with the challenge of making a car drive itself down the road. The car's challenge is an AI problem; sensing and understanding what's going on around it and where the road goes.
At the moment this is a much more complicated problem than it need be because the car has to cope with a system designed entirely for humans. You could easily get cars to start broadcasting their intent to neighbouring vehicles so that traffic could become self-organising and far more cooperative than humans could achieve. You could get digital information from the road that is far more precise and rich than just a sign that says "bend ahead". And of course you can give cars sensors that see through fog, see in the infrared and "see around corners" by sharing data with nearby vehicles.
It's a classic AI problem and doesn't require a huge leap forward in computer science, just refinement of current engineering supported by the usual incremental reduction in the cost of number crunching.
ATG said:
The idea that you can't write stable, predictable software is bilge. There's tons of it around keeping machines running. Aircraft are a pretty good example of what can be achieved. They are properly complicated systems running huge amounts of software and they are highly fault tolerant.
.
The key here is fault tolerant, the fact there are extra bits to make it fault tolerant is accepting the software isn't reliable and faults have to be mitigated with backup systems, etc..
Reliable software is a fallacy.
ATG said:
The idea that you can't write stable, predictable software is bilge. There's tons of it around keeping machines running. Aircraft are a pretty good example of what can be achieved. They are properly complicated systems running huge amounts of software and they are highly fault tolerant.
But that's got bugger all to do with the challenge of making a car drive itself down the road. The car's challenge is an AI problem; sensing and understanding what's going on around it and where the road goes.
At the moment this is a much more complicated problem than it need be because the car has to cope with a system designed entirely for humans. You could easily get cars to start broadcasting their intent to neighbouring vehicles so that traffic could become self-organising and far more cooperative than humans could achieve. You could get digital information from the road that is far more precise and rich than just a sign that says "bend ahead". And of course you can give cars sensors that see through fog, see in the infrared and "see around corners" by sharing data with nearby vehicles.
It's a classic AI problem and doesn't require a huge leap forward in computer science, just refinement of current engineering supported by the usual incremental reduction in the cost of number crunching.
You cant compare the two since aircraft don't fly 3 feet apart from each other and the other aircraft they're trying to avoid aren't flown by Doris who can barely see past the nose cone. Aircraft are predictable in their behaviour. Cars, pedestrians and roads are not.But that's got bugger all to do with the challenge of making a car drive itself down the road. The car's challenge is an AI problem; sensing and understanding what's going on around it and where the road goes.
At the moment this is a much more complicated problem than it need be because the car has to cope with a system designed entirely for humans. You could easily get cars to start broadcasting their intent to neighbouring vehicles so that traffic could become self-organising and far more cooperative than humans could achieve. You could get digital information from the road that is far more precise and rich than just a sign that says "bend ahead". And of course you can give cars sensors that see through fog, see in the infrared and "see around corners" by sharing data with nearby vehicles.
It's a classic AI problem and doesn't require a huge leap forward in computer science, just refinement of current engineering supported by the usual incremental reduction in the cost of number crunching.
CLK-GTR said:
You cant compare the two since aircraft don't fly 3 feet apart from each other and the other aircraft they're trying to avoid aren't flown by Doris who can barely see past the nose cone. Aircraft are predictable in their behaviour. Cars, pedestrians and roads are not.
Indeed. And despite the predictability of its flightpath and the inherent safety net of a huge buffer zone around it, every single passenger aeroplane in the sky has a couple of blokes sitting at the front to make sure it doesn’t crash. Odd that. You’d think super clever infallible software would be able to fly a plane pretty easily really, if it’s so capable. Autonomous vehicles are a load of b*ll*x that simply don’t need to exist and will never function reliably in any of our lifetimes. A shedload of ‘driver assist’ gimmicks but a human ultimately in charge is likely to be as good as it gets, and I think that’s plenty.
Automation works well in a clearly defined environment. Inside a cage in a factory.
In a factory, following lines on the floor.
On a battlefield, with no humans in the near vicinity.
If all cars were autonomous, and were able to communicate with each other, the problem.of pedestrians and other road users would come into play. Put a 6 foot fencea around every road, and you have the safety that trains have. Oh wait. People jump.in front of those.
In a factory, following lines on the floor.
On a battlefield, with no humans in the near vicinity.
If all cars were autonomous, and were able to communicate with each other, the problem.of pedestrians and other road users would come into play. Put a 6 foot fencea around every road, and you have the safety that trains have. Oh wait. People jump.in front of those.
Full self driving feels like a long way off at the moment, not least because cars have to interact with other road users (pedestrians, cyclists, human drivers) who are unpredictable.
But I would guess that full self driving in limited situations will be here much sooner - motorways being a good example. It would be relatively straightforward to create software that allowed drivers to join a motorway, set intelligent cruise control and let the car sit in lane 1 (or lane 3) at a steady 70mph while the driver does something other than driving, only taking over when the time comes to leave the motorway. You would get to your destination much more relaxed and refreshed, and accident rates would be lower.
The “last mile” is always going to be harder to solve.
But I would guess that full self driving in limited situations will be here much sooner - motorways being a good example. It would be relatively straightforward to create software that allowed drivers to join a motorway, set intelligent cruise control and let the car sit in lane 1 (or lane 3) at a steady 70mph while the driver does something other than driving, only taking over when the time comes to leave the motorway. You would get to your destination much more relaxed and refreshed, and accident rates would be lower.
The “last mile” is always going to be harder to solve.
My wife's Volvo's semi-autonomous lane keeping on the motorway is crap, every junction it tries to pull off always sits off centre in the lane towards the left and fights like hell when I try and correct it, I know you're supposed to keep hands on the wheel anyway, but I don't trust it to let go anyway.
My cheap electric Hyundai is brilliant by comparison.
As for fully autonomous unless it's on rails I don't want to try it.
My cheap electric Hyundai is brilliant by comparison.
As for fully autonomous unless it's on rails I don't want to try it.
Zigster said:
Full self driving feels like a long way off at the moment, not least because cars have to interact with other road users (pedestrians, cyclists, human drivers) who are unpredictable.
But I would guess that full self driving in limited situations will be here much sooner - motorways being a good example. It would be relatively straightforward to create software that allowed drivers to join a motorway, set intelligent cruise control and let the car sit in lane 1 (or lane 3) at a steady 70mph while the driver does something other than driving, only taking over when the time comes to leave the motorway. You would get to your destination much more relaxed and refreshed, and accident rates would be lower.
The “last mile” is always going to be harder to solve.
Except you wouldn’t, because “self driving” cars still need someone sitting in the drivers seat being ultimately responsible for everything the car does. What you would end up with is two camps; one being the drivers who play by the rules and are bored rigid, and the other being those who sit there playing on their phones hoping the thing doesn’t crash itself, ending up losing their licenses as being on your phone would be as illegal in that situation as it is now. Again; it just doesn’t work and it just isn’t needed. Lane assist, adaptive cruise and and auto braking are pretty effective safety kit on the motorways.But I would guess that full self driving in limited situations will be here much sooner - motorways being a good example. It would be relatively straightforward to create software that allowed drivers to join a motorway, set intelligent cruise control and let the car sit in lane 1 (or lane 3) at a steady 70mph while the driver does something other than driving, only taking over when the time comes to leave the motorway. You would get to your destination much more relaxed and refreshed, and accident rates would be lower.
The “last mile” is always going to be harder to solve.
ATG said:
The idea that you can't write stable, predictable software is bilge. There's tons of it around keeping machines running. Aircraft are a pretty good example of what can be achieved. They are properly complicated systems running huge amounts of software and they are highly fault tolerant.
But that's got bugger all to do with the challenge of making a car drive itself down the road. The car's challenge is an AI problem; sensing and understanding what's going on around it and where the road goes.
At the moment this is a much more complicated problem than it need be because the car has to cope with a system designed entirely for humans. You could easily get cars to start broadcasting their intent to neighbouring vehicles so that traffic could become self-organising and far more cooperative than humans could achieve. You could get digital information from the road that is far more precise and rich than just a sign that says "bend ahead". And of course you can give cars sensors that see through fog, see in the infrared and "see around corners" by sharing data with nearby vehicles.
It's a classic AI problem and doesn't require a huge leap forward in computer science, just refinement of current engineering supported by the usual incremental reduction in the cost of number crunching.
Hmm. Unless you ban humans from interacting with the road completely then you have no choice but to design a system to cater for them. There will always be pedestrians, cyclists, animals etc so the idea that the problem can be solved by dismissing them is a total non-starter. Interactions with other vehicles is a small part of the problem, and the part that is most readily solved, at least technically although not necessarily commercially and from a global regulatory perspective. But that's got bugger all to do with the challenge of making a car drive itself down the road. The car's challenge is an AI problem; sensing and understanding what's going on around it and where the road goes.
At the moment this is a much more complicated problem than it need be because the car has to cope with a system designed entirely for humans. You could easily get cars to start broadcasting their intent to neighbouring vehicles so that traffic could become self-organising and far more cooperative than humans could achieve. You could get digital information from the road that is far more precise and rich than just a sign that says "bend ahead". And of course you can give cars sensors that see through fog, see in the infrared and "see around corners" by sharing data with nearby vehicles.
It's a classic AI problem and doesn't require a huge leap forward in computer science, just refinement of current engineering supported by the usual incremental reduction in the cost of number crunching.
Even if we accept your grossly over simplistic and wholly unrealistic premise, how long will it be before standards for inter-car safety co-operation are developed and rolled out to sufficient new vehicles to actually make a difference? Decades is the answer. Put the human interaction problem back in, as we must because it is there and is not going away, then it's even longer.
Pretending the hardest part of a problem doesn't exist because you don't have the technology to solve it is beyond disingenuous.
Southerner said:
Except you wouldn’t, because “self driving” cars still need someone sitting in the drivers seat being ultimately responsible for everything the car does. What you would end up with is two camps; one being the drivers who play by the rules and are bored rigid, and the other being those who sit there playing on their phones hoping the thing doesn’t crash itself, ending up losing their licenses as being on your phone would be as illegal in that situation as it is now. Again; it just doesn’t work and it just isn’t needed. Lane assist, adaptive cruise and and auto braking are pretty effective safety kit on the motorways.
Surely if it’s actually self driving the car, or manufacturer, is responsible while it is engaged?Assuming it works to a decent standard, it'll avoid many collisions. It won't get emotional, frustrated or distracted like humans do. It remains cautious all of the time. It can't be perfect or avoid all collisions and perhaps it'll cause a few that wouldn't have happened but the net result will be fewer collisions and KSIs.
Forester1965 said:
Assuming it works to a decent standard, it'll avoid many collisions. It won't get emotional, frustrated or distracted like humans do. It remains cautious all of the time. It can't be perfect or avoid all collisions and perhaps it'll cause a few that wouldn't have happened but the net result will be fewer collisions and KSIs.
The result will be gridlock, as computers famously lack the common sense aspect which makes humans a far more capable beast. Two cars come nose to nose in a tight urban street, which robotwagon is going to inch up the kerb to make room to pass? Neither of them are, because they’re not allowed to. And so on and so on… There is a reason they test these gimmicks on big, open, straight and wide roads in the US and not around central London!FlabbyMidgets said:
Surely if it’s actually self driving the car, or manufacturer, is responsible while it is engaged?
Do you think? Ho ho ho; no they have no intention of soaking up that risk, it’ll be your fault when it crashes and your insurance footing the bill. Another reason they’ll not be genuinely “self driving” any time soon.Southerner said:
Except you wouldn’t, because “self driving” cars still need someone sitting in the drivers seat being ultimately responsible for everything the car does.
Exactly this.Can someone explain the me the logic of having a driverless car that has to have a driver. What is the point?
Southerner said:
The result will be gridlock, as computers famously lack the common sense aspect which makes humans a far more capable beast. Two cars come nose to nose in a tight urban street, which robotwagon is going to inch up the kerb to make room to pass? Neither of them are, because they’re not allowed to. And so on and so on… There is a reason they test these gimmicks on big, open, straight and wide roads in the US and not around central London!
There's a risk of that, but software design is clever enough to work around it as systems are made more sophisticated.I had a Volvo with the autonomous driving. In stop start through town it'd creep up to the car in front and ignore the junction I was now blocking to my nearside. This would cause all sorts of problems if every car on the road did it. However it's not beyond the wit of man to write some software that recognises most junctions and uses a set of parameters to leave that space clear. It's just not how it works today (on that car, at least).
Forester1965 said:
Southerner said:
The result will be gridlock, as computers famously lack the common sense aspect which makes humans a far more capable beast. Two cars come nose to nose in a tight urban street, which robotwagon is going to inch up the kerb to make room to pass? Neither of them are, because they’re not allowed to. And so on and so on… There is a reason they test these gimmicks on big, open, straight and wide roads in the US and not around central London!
There's a risk of that, but software design is clever enough to work around it as systems are made more sophisticated.I had a Volvo with the autonomous driving. In stop start through town it'd creep up to the car in front and ignore the junction I was now blocking to my nearside. This would cause all sorts of problems if every car on the road did it. However it's not beyond the wit of man to write some software that recognises most junctions and uses a set of parameters to leave that space clear. It's just not how it works today (on that car, at least).
Edited by Southerner on Saturday 17th February 10:06
Southerner said:
Do you think? Ho ho ho; no they have no intention of soaking up that risk, it’ll be your fault when it crashes and your insurance footing the bill. Another reason they’ll not be genuinely “self driving” any time soon.
They have no choice, there is a Bill going through Parliament now that changes that. It will also stop misleading marketing calling things ‘self driving’ when it is just ADAS
Also there is a developer that does testing in London, see them on the road quite often
FlabbyMidgets said:
Southerner said:
Do you think? Ho ho ho; no they have no intention of soaking up that risk, it’ll be your fault when it crashes and your insurance footing the bill. Another reason they’ll not be genuinely “self driving” any time soon.
They have no choice, there is a Bill going through Parliament now that changes that. Gassing Station | Speed, Plod & the Law | Top of Page | What's New | My Stuff