Autonomous Cars

[Danny McG]

There's problem with them is they don't require someone to be behind the wheel. At the moment, if I am crossing a road where there are cars apparently parked across** the road from me, I can tell whether they may be about to move because they have someone sitting in the driver's seat.

 

[mosaix]

It is one thing to have a power failure for objects that are stationary, like buildings, but what happens for large masses of moving traffic when the internet burps, suffers from latency, a term makes self driving car manufacturers shudder, or outright power failure, what does that do to all those vehicles relying on electronic contact instead of good old fashioned line of sight with a brain that can adapt to whatever it is seeing in a moments notice. Doesn't even take into account hacking. How does such a system react, does it slam on the brakes for everything moving at that moment. Slowly slow everything down as it shifts everything moving over to sensors riding on every car. Everything slows down to a complete stop. Or do we let the manufacturers assume that is never going to happen.

If I remember correctly, Robert, the communication will be via a protocol like Bluetooth rather than the internet. But still, there has to be a way of controlling things when communication fails - as it will.

Edit: Although my satnav has an internet connection via a SIM card so, no doubt, all future systems will have some kind of internet involvement.

 

[CupofJoe]

Do you think that the car manufacturers will try and give their autonomous cars a personality that reflects their brand?
Will a Ferrari or Porsche give you a different drive experience than that from a Ford or Honda? I don't mean what is in the passenger cabin, but how the car feels as it moves on the road.

 

[Vladd67]

Another way for people in China who fail on their social score to have their movements restricted.

 

[REBerg]

Do you think that the car manufacturers will try and give their autonomous cars a personality that reflects their brand?
Will a Ferrari or Porsche give you a different drive experience than that from a Ford or Honda? I don't mean what is in the passenger cabin, but how the car feels as it moves on the road.

I see car makers temporarily trying to outdo each other with self-driving models, but I also see individual ownership giving way to manufacturing company fleets providing transportation on demand.
I suppose riders might derive some sort of status through the quality of the vehicle they summon, but the days of taunting your neighbor with your brand new Mercedes would be over.

 

[CupofJoe]

I see car makers temporarily trying to outdo each other with self-driving models, but I also see individual ownership giving way to manufacturing company fleets providing transportation on demand.
I suppose riders might derive some sort of status through the quality of the vehicle they summon, but the days of taunting your neighbor with your brand new Mercedes would be over.

I was reading an article if we have already bought our last cars. It made a lot of good arguments for urban and suburban car ownership coming to an end. Not so much for those without street lighting. The move to shared ownership or subscription is something that has the car companies quaking at what might happen over the next 10-20 years. I guess the bragging rights will be if you subscribe to Rolls-Royce-Rolls, Mercedes-Uber or Dacia-Lyft?

 

[Vladd67]

I see a think tank has suggested that all private cars should be banned from London by 2030 and that TfL should operate their own version of Uber along with making escooters available and offering free tube rides. Who pays for the free tube service wasn’t mentioned.

 

[thaddeus6th]

Think tanks come out with some tosh sometimes. I remember one from a couple of years ago suggesting abolishing real world money in the next decade or so.

 

[Foxbat]

On the subject of learning: as far as I understood it, the learning was not about alpha or beta testing on the road but more a build up of AI experience. The argument touted was that an individual who drives on the road can only improve by learning from their own experience but autonomous cars can share that experience with each other almost instantaneously. The example given was something like a car learns something in (say) London and minutes later that learning is put into practice in Miami.

P.S. Isn't the answer to the comms situation the much vaunted 5G?

 

[Foxbat]

Just had a thought. If 5G is the answer and Huawei decided to make cars, would they be allowed in the USA or would that be seen as a prelude to invasion?

 

[CupofJoe]

P.S. Isn't the answer to the comms situation the much vaunted 5G?

That is what a LOT of money is being spent on... 5G on your mobile phones is not the same as the 5G IoT [Internet of Things]

 

[Ursa major]

but autonomous cars can share that experience with each other almost instantaneously. The example given was something like a car learns something in (say) London and minutes later that learning is put into practice in Miami.

But that would be alpha and beta testing on the road.

How can something be tested elsewhere if, minutes after Car A learns something (Learnt Lesson L) in London (based on Incident X), Car B is putting it into practice in Miami? Where is the off-road testing? There isn't any. There's no involvement in the creation of the solution (that which is learnt and then passed onto other cars) by anything other than Car A.

There is no guarantee that the solution does not cause problems, either in a replay of Incident X or incidents that resemble Incident X. And if the solution is a variation on "I wouldn't start from here", it may involve doing something else in the run up to Incident X that avoids Incident X entirely... but then has other cars changing their behaviour in situations that resemble the run-up to Incident X that would not have ever led to a replay of Incident X, but might lead to another incident.

Now if that sounds complicated, just imagine what lesson learning means, both for the cars learning them and to those changing their decision trees based on that learning.

And then there is the issue that laws (or even the side of the road on which the cars drive) are not identical in different countries. Such things mean that the lessons learnt have to be transformed so that they take account of the different driving environments found around the world.

So the list of areas where problems might occur is legion: incorrect analysis of Incident X; incorrect determination of the solution; incorrect creation of the solution for Car A's decision tree; incorrect transformation of Car A's decision tree changes to match the driving environment experienced by Car B in Miami. And in all of this, there is no independent stress testing of the solution in Car A or Car B (or the hundreds of millions of other cars that are given the solution)... unless you count the stress of passengers in those cars that suddenly start performing strange, not to say inappropriate to the conditions, actions all over the world.

The only safe place to be might be behind vehicle-proof walls.... (And did you see those huge autonomous mining trucks in that episode of Click...? )

 

[Robert Zwilling]

If conditions stayed the same the cars would not need constant updating except for security updates. The weather is on the move due to the constantly increasing freshwater being added to the system and the slowly increasing temperature. It is literally changing every week and it is not heading for a calm plateau anytime soon. The automatic vehicles will need constant updating to respond to the changing physical conditions which is seat of the pants navigation never out of beta situation. The easiest option is to just have the vehicles sit it out once the local weather reaches a threshold point. Will the cars reverse directions and not even go into storms to start with. Are emergency vehicles better off always on manual drive with people able to drive them. Will a new class of pilots spring up, called land pilots who travel where there are no automated roads. Instead of a driver's license, the ordinary person gets a passenger license. Once people get used to automatic cars they will start losing their ability to rapidly respond to risky situations while driving a vehicle. Will bicycles/scooters/motorcycles have to ride in physically separated lanes, their own private roadways as cars that are not linked into the system would probably be banned from driving automated roads.

 

[REBerg]

If millions of distressed and distracted (as well as more than a few downright dumb) drivers can successfully travel the roadways of the world daily, it would not take an exceptionally advanced computer system to take their places.

Driving isn’t all that complicated. Vehicles are confined to designated areas. Within those areas, a finite set of rules must be followed. Unlike human drivers, autonomous vehicles would follow those rules to the letter.

Networking can keep driverless vehicles from colliding with each other. Sensors can prevent them from hitting people, pets, wildlife and brick walls.

Legal responsibility for accidents would shift from drivers to vehicle manufacturers and victims. Case law would decide how responsibility would be assigned to a manufacturer and how much to a victim.

The existence of autonomous cars would obviously be public knowledge. Bicyclists, pedestrians, parents and pet owners would be obligated to take reasonable actions to protect themselves and others as they interact with driverless cars -- just as they now do with vehicles controlled by humans.

I honestly believe that autonomous cars will save lives. Now, what about autonomous, passenger-carrying drones?

 

[Foxbat]

But that would be alpha and beta testing on the road.

How can something be tested elsewhere if, minutes after Car A learns something (Learnt Lesson L) in London (based on Incident X), Car B is putting it into practice in Miami? Where is the off-road testing? There isn't any. There's no involvement in the creation of the solution (that which is learnt and then passed onto other cars) by anything other than Car A.

There is no guarantee that the solution does not cause problems, either in a replay of Incident X or incidents that resemble Incident X. And if the solution is a variation on "I wouldn't start from here", it may involve doing something else in the run up to Incident X that avoids Incident X entirely... but then has other cars changing their behaviour in situations that resemble the run-up to Incident X that would not have ever led to a replay of Incident X, but might lead to another incident.

Now if that sounds complicated, just imagine what lesson learning means, both for the cars learning them and to those changing their decision trees based on that learning.

And then there is the issue that laws (or even the side of the road on which the cars drive) are not identical in different countries. Such things mean that the lessons learnt have to be transformed so that they take account of the different driving environments found around the world.

So the list of areas where problems might occur is legion: incorrect analysis of Incident X; incorrect determination of the solution; incorrect creation of the solution for Car A's decision tree; incorrect transformation of Car A's decision tree changes to match the driving environment experienced by Car B in Miami. And in all of this, there is no independent stress testing of the solution in Car A or Car B (or the hundreds of millions of other cars that are given the solution)... unless you count the stress of passengers in those cars that suddenly start performing strange, not to say inappropriate to the conditions, actions all over the world.

The only safe place to be might be behind vehicle-proof walls.... (And did you see those huge autonomous mining trucks in that episode of Click...? )

As I understood it, the piece on learning was after the widespread use of autonomous cars on the roads had been established and this was to be an ongoing process, with a much greater learning database than any one human could have. Perhaps I misunderstood.

 

[Ursa major]

As I understood it, the piece on learning was after the widespread use of autonomous cars on the roads had been established

The number of autonomous cars as a proportion of all the cars on the road has zero relevance to the issue of a lack of testing.

The lack of testing is an inevitable consequence of a car in, say, London learning something -- it isn't doing this in a lab, but on the road -- and what it has learnt being disseminated around the world in a few minutes, including to a car in, say, Miami.

Whether there would be only two autonomous cars in the world, or those two cars would be but two out of all the hundreds of millions of cars in the world (all of them autonomous), makes no difference at all...

...except that if the lesson learnt was not the right lesson:

• in the former case, only two cars would be in danger of following the wrong rule,
• in the latter case , every car in the world would be in danger of doing that,

and in both cases, not a single second would have been spent on independently analysing or testing the new rule, either in isolation or as but one rule in combination of all the other rules (an ever increasing number of them) that the cars are meant to follow.


Frankly, this is the sort of dubious stuff one might expect in a badly thought-out SF story but, unfortunately, it appears to be the norm for reporting on science and technology, where statements made by people working on the technology or science (or, worse, by the marketing departments of the companies involved) are repeated verbatim and without question.

 

[Toby Frost]

Think of the fun a Hostile Foreign Power could have when it hacks the controls for the driverless cars!

 

[Robert Zwilling]

At times we seem to be living the logic of Phillip K Dick science fiction stories, why would industry not be subject to the same conditions.

Being part of a network the system would need updates from time to time. How could those updates be tested, would a small number of cars be updated ahead of others so if it was a bad update only a few cars would experience problems. How many would be enough and if an update was instantly needed would the risk of mass updating be overlooked with fingers crossed. Are there any operating systems that don't need constant updates. Could a multitude of manufacturers be allowed to make the cars or would there be only one car company to make sure everything was always in step. Is this a case of one person's dream, another person's nightmare. Can we even make a network that would encompass all kinds of vehicles using the one size fits all scenario.

There would probably be a market for hacked cars, with fake registrations or more options normally allowed such as unlimited speed, done in such a way that they would be cruising under the radar. One aspect of automatic cars is the ability of authorities to shut the car down any time they wanted to. That could be extended to repossessions done by remote control. If it was possible to control a family members use of a car, would some kind of safeguard be needed so one person couldn't block another family member from legitimate use of a the car. Would it be possible to stop cars from being used for illicit purposes before they were used. The cars would probably be able to recognize legitimate users making it far easier to take one's license away or would the need for that punishment disappear.

 

[thaddeus6th]

Toby, indeed.

The Internet of Things (online kettles etc) have been hijacked to provide computing power for DDOS attacks and hackery. The NHS got hit by an attack a couple of years ago. Just because autonomous cars are possible doesn't make them a good idea.

 

[Foxbat]

Online kettles. Why? Just why?

I can just imagine the text message it might send to its owner: I need filling up but the tap is busy ransoming the NHS again. Tap says no! Repeat! TAP SAYS NO!