Who things they are a good thing or a bad idea on the roads?

We are crazy if we don't go down this route. Less pollution, less traffic jams, less roads to be built, less accidents.

We generally don't have automatics or cruise control here so I can imagine its going to be a fight.

I have some classic cars but I guess they will eventually be relegated to non-public roads.

Do you have confidence DD when you see how many computor problems we have one way and another ?

Aeroplanes have used electronics an computers for years. Lifts have, trains have, ships have etc. medicine and food is produced using computers. Our utilities use computers.

It's not difficult nor expensive to build in autonomous redundancy checks above and beyond.

I think the difference in my eyes ,causing my doubt, is the fact that roads are so cluttered , in a very random way - so many variables. Human beings all with their own agenda, changing their minds by the minute on what direction they want and resetting, random decisions to move by bicycles, peds, animals large and small etc. Would all the cars talk to each other electronically ? What would happen if Mr X driving along , turned back to manual to suddenly turn right, and knocked off course an autonomous car that couldnt adjust its direction in time ? Wouldnt there be a risk of the latter 'going randomly awry ' if damaged , and causing mayhem?
Also with the things you cited, there is generally a human watching over the system who can see the whole picture , which you wouldnt get if you were an individual in a car.
I can see it working with trains on rails, with lorries on those long boring little used transport routes in the world, and with planes where there is less unpredictability and a manual screen overseer, but ordinary domestic cars on local roads -- mmmm I dont know.


Computors regularly allow too much fluoride/ chlorine into water supplies, factory computors tell you sometimes stock remains which doesnt, and so on. The whole thing makes me nervous

Excessive fluoride is not a computer error in the far greater number of cases.

Yes, the cars should talk to each other but there will be an interim period where they may not.

There should not be manual override, just a stop button.

I think its a mad science fiction idea. Venice is right. Although ships and planes use software control, they are only used there when there is no traffic. The claims that they will lead to no traffic jams and safer motoring are based on nothing but optimism. My bet is that in order to do it the car would need to be at least as intelligent as a human, and we are a long way from that.

There are also moral questions that are not even satisfactorily solved by human beings which would need to be set out in detail. For instance, a child runs out, and the car has the choice of hitting it or a group of people on the pavement.

How on earth anyone expects a computer program to handle such a case baffles everyone. I doubt there is a workable program that is even capable of recognising the age of a person in poor driving conditions even!

Finally, as Venice points out, software is untestable under all possible conditions, and when it goes wrong, it can do almost anything. The designers of the systems used to guide aircraft are well aware of this and use multiple computers and take a majority vote. They even go so far as to ensure that the development teams are kept completely separate and have not been educated together or worked together to ensure as far as is possible the computers will not go wrong in the same way at the same time. Even then, human flight traffic controllers are used to keep aircraft well apart on landing and take off.

And who pays for the damage if your car makes a wrong decision? Your insurance? The software developer? I guess its great news for m'learned friends anyway!

There are many planes that are under continuous computer control, they are un-flyable without the computer doing most of the work, the pilot says what he wants the aircraft to do, the computer decides what the aircraft will do and how it will do it, the pilot cannot override the computer easily.

Some planes can land themselves though I don't think this is allowed as yet - but it does happen.

The child running out and deadlock scenario:- the computer has at least half a second and realistically more like 2 seconds advantage over a human driver. Unlike a human driver it will know what limits the car can achieve, many humans slam the brakes on and attempt a 90 degree turn - the car will become unpredictable, whatever the human decided to do morally just won't happen, iirc at least 30% of humans close their eyes, over 20% let go of the steering wheel and over 10% totally freeze. Most humans haven't got a clue in a real emergency, they are not used to it and the two main instincts of fight or flight are not conducive with control.

I would imagine that in a deadlock situation like that the computer would be programmed to stop, honk the horn and stay going in the direction of travel - that way it is predictable to 3rd parties and gives them the best chance.

The computer would see the situations arising and be programmed to slow down before hand - humans assess the probability and often decide to take the risk and do not slow down "its ok, they are bound to see me" attitude.

The computer has the advantage that it is 100% aware all the time, its anticipation is much greater, there is less risk factor allowed, it has 360 vision continuously, its reaction time is hugely faster.

If I had a shielding device on a starship, I would prefer it to be operated automatically by computer than someone with a button waiting to press it.

Yes there will still be obscure accidents but there will be far less of them.

I don't like the totally independent development that is happening, the government should produce an outline - eg the autonomous cars must talk to others, types and amounts of redundancy, under "normal braking", the brake lights must be lit 2 seconds before slowing etc.

As I said previously, redundancy systems add very little to the price, the electronics is cheap, basic sensors for redundancy are cheap.

There needs to be an arbitration system, if two autonomous cars are meeting in opposite directions and the road narrows to a single lane between them, who goes first must be negotiated between the two vehicles. At the moment the cars will hand control back to the human.

Some degree of centralised control would be best (whilst maintaining total autonomous capability), however the security aspects of that are immense at the moment, but not insurmountable. Problems are solved faster when there is a need to solve them or the rewards are greater.

The "need to be as intelligent as a human" doesn't hold much weight, there are loads of things computers can do that humans can't, some because of speed, some because of quantity/size/scale and some because of complexity. Memorizing pi to 70,000 digits is brilliant for a human and chickenfeed for a computer that can even calculate it much faster than a human can recall it.

But dont the computor systems have to be set and programmed and maintained by humans who are prone to errors? (And a few rogue nutcases thrown in now and again)

looks like its all ploughing on regardless anyway. Wont be long now, as you can see [url=http://www.driverless-future.com/?page_id=384]HERE[url]

Yes, but its relatively easy to test computers in these scenarios as you are testing for zero errors.

When you test humans you get loads of errors.

On your link you missed the "/" off the ending [/url

Ah, the slash before the last address -

so adding it, lets see if its [http://www.freshfieldsrescue.org.uk/blogHERE[/http://www.freshfieldsrescue.org.uk/blog]

No, I still havent got it right and I added the slash before I put the url after the word HERE

If most cars have cruise control not many use it. When following cars on highways in the states it's much easier driving because most cars are doing a steady speed.

The facts of air transport are that you are unbelievably unlikely to hit another aircraft in free flight. Thus a computer - or rather computers (plural) can be used for guidance. Even so, human are required to avoid unpredictable events such as clear-air turbulence etc, and decide whether to go higher or lower or go round it. Such decisions are NOT entrusted to computers, neither are radar detection of nearby aircraft.

Roads are far more crowded, and a dead fly or ice smeared over the lens of a camera could render the vehicle blind to a child running off a pavement.

Such events are easily overcome once they have been thought of, but how on earth can the designers of these systems think of every possible event when there are probably billions of them?

I can remember not long ago driving through Birkenhead and seeing a mother remonstrating with a child having a tantrum on the pavement ahead of me. I slowed down, and thank god I did. I was able to stop easily when the little lad escaped his mothers clutches and ran out in front of me. Could your computer manage that? Or even your computer(s).

You seem to think computers have a huge speed advantage. Generally they do, but digital signal processing takes significant time. You can hear the delay with DAB radios, and it takes seconds. The problem is worse for image processing of course and I doubt your cheery assumption that microprocessors in a car could do the job faster than a human. The evidence is that it can't (There are no systems in existence that can beat a half decent clay pigeon shooter for instance)

I doubt whether self driving cars without steering wheels will ever take over. They are neither necessary nor particularly desirable. Most of us would be bored silly by sitting and might as well do the driving, and many of us quite enjoy it.

Just because a thing is possible shouldn't mean we NEED to do it, and safe self-driving cars are not even possible. Power steering systems sometimes fail, by the way. It's happened to me whilst driving. I was able to drive the car quite safely however, but how would a computer manage such an event?

It is an idiotic thing to even attempt. that several wealthy companies have embarked on this is no great surprise. Many did the same with battery powered cars and they seem to be flopping too, despite huge attempts to foist them on the public.

That is a wild guess, compare the number of mid-air collisions with the total number of flights to the number of car-to-car crashes to the total number of car journeys and you may be shocked. If you also take into account the number of deaths then you may find that you are less likely to die in a single car journey than a single flight in an aircraft. You'd also be amazed how often a pilot has to take avoidance actions.

The reason that the number of mid-air collisions has fallen over the years is because of computers, not the quality of pilots. The regulations have been put in place to compel the pilot to use computer systems father than fly by the seat of their pants.



I'd love to see a pigeon shooter fire at a super-sonic missile or aircraft, these days systems can even fire down shells. Many systems are way faster than humans and that is why they are computer controlled.



And highlights the problem with humans, they over-optimistically assess risk based on other factors eg I'm in a rush to get somewhere, what is more a computer would probably detect it was going faulty long before a human would. You think you were driving safely with a known broken power steering system that potentially could have yanked the wheel sideways (which I have had happen). Fairly obviously a computer would be programmed to stop if a critical system is detected to be faulty.

Most electric power steering units fail on the sensor which detects what the driver is doing, its a lot less common for the actuator side to fail - but these aren't designed as autonomous fail-safe systems because when the actuator fails the driver still has control.

Autonomous cars will never be safe, but neither is anything else, they will become safer than humans. Humans don't need a fly on the windscreen or a child running out, they choose to have accidents of their own free will eg speeding, letting themselves be distracted, over-estimating their capability, over-estimating the car's capability or just being stupid.

There are many drivers out there that think they have a right of passage from wherever they are to wherever they want to be. A good thing that could come out autonomous vehicles is that they may be able to report other's ridiculous driving.

I used to drive stupidly for many years, because it appeared to be the norm. It is only by incredible luck that I never killed anyone else. I don't claim to be a perfect driver now but I am very conscientious about other people's lives which selfishly I didn't do in the past - computers don't do selfish, impatience etc, they are human traits.

Look at formula 1 drivers, a lot of time is spent comparing what the driver did against what the computer works out would have been optimum, a lot of time is spent training the driver to try and match the computer. Obviously the regulations restrict what is allowed to be done by computer during a race but its virtually being done by proxy. Even so, a computer is limiting the driver as to what he is allowed to do, very few F1 drivers would not destroy their cars if the the limits were switched off, humans gamble with luck.

Transportation and driving should be segregated, if you want to drive do it on a track, if you want transportation, let a machine take you there.

Fairly obviously a computer would be programmed to stop if a critical system is detected to be faulty.

There is such a thing as 'progressive failure' which involves the mechanism that detects failure failing before the main mechanism fails. Building systems with this sort of thinking in mind is HARD, and software makes it MORE difficult not less.

'Fairly obviously' you declare, the vehicle would stop as soon as a failure was detected. But what if this were on a motorway in the fast lane with traffic all around you? I suggest that once again your glib solution is a lot more complicated than you think.

And once again you have 'solved' the problem once it has been pointed out to you. This is a lot easier to happen than thinking up the possibility of it happening, and there are billions of possible events, many of which you would not dream of. The problem is not one of finding a solution. It is finding ALL POSSIBLE EVENTS and establishing a database of what should be done. I don't believe this is possible. The alternative is to have sufficient intelligence to cater for unpredictability, and we are a million miles from this. If you read Penrose's "The Emperor's New Mind", he presents a very cogent and plausible argument that we never will.

Shooting down a shell is not difficult. They are predictable, have long transit times and don't go very fast. Proximity fuses ease the problem considerably.

A reasonable clay pigeon player will manage over 90% hits when he has under second to aim and pull the trigger.

As I've never been a petrol head, I have little interest in cars or F1 drivers and dislike speed, but it seems to me that examining a computer analysis to 'improve' their performance is a very different kettle of fish from designing a robot that could beat them on the track. Footballers examine their tactics for the same reason and might well use a computer to help them in this, but does it help them in designing a robot striker? I've seen robot football matches, and pathetic is a wholly inadequate description of it.

By your own argument, humans do things like getting drunk or suicidal or irrational and do crazy things. Fortunately other drivers can recognise this and take appropriate action, but I wonder how you propose handling all this in software?

The idea of autonomous vehicles mixed with impatient or deranged drivers is ludicrous. The only way would be to trash all our cars and redesign all our roads so a computer could handle the problem. It is not going to happen. Dream on my friend.

I don't need to dream its happened.

One statistic that has cropped up is the difficulty of the handover from autonomous mode to driver control, mainly to do with foot controls.

On computer games, you don't bother analysing whether the randomly moving computer opponent is drunk, irrational, young old or suicidal, you react to the stimulus - adding emotional analysis unnecessarily complicates issues and is yet another self-distraction where human drivers fail.

I think that's about my lot - look forward to the next subject, cheers!

http://www.driverless-future.com/?page_id=384

There were similar forecasts about battery cars which have been wrong too.

Statistically, you would need billions of road miles to be able to decide whether autonomous cars are safer than human drivers or not. So far, there are none. We have a few prototype vehicles with human drivers ready to take over if they screw up, and some sensors have been installed such as radars which apply brakes or enable parking in optimal circumstances, but thats about it.

Just don't get a Tesla. Look at the way it swerves into oncoming traffic!

https://www.youtube.com/watch?v=MrwxEX8qOxA