Thread: Autonomous Vehicles

GPS co-ordinates+Google maps+mapped roads should find most people. If you call an Uber to your address, you will be expected to make sure the gate is open. After all, the pervasive spread of Big Brother means it's very hard for anyone to hide for long.

Originally Posted by bee utey

Of course, zero fatalities is a goal which will never be reached in practice. That doesn't mean a 90% reduction is impossible.



Because it has to be to be effective.

I am not clear where you get the "has to be effective" from! While self driving cars may remove some issues from driving, such as not knowing the intentions of the car in front (or more importantly the one up the side street, or about to pull out etc), it introduces new uncertainties - was this particular series of circumstances tested in the software development, and even more importantly, was it tested after the last software update? Look at the latest Windows 10 update - this, perhaps the most widely used and tested software in the world, unexpectedly started to delete files in some circumstances, for no apparent reason. If there is any reason to expect self driving software to be more free from this sort of undocumented feature, I have yet to see it explained!

Self driving tech isn't an iphone app.

Nowhere have I even hinted this. My concerns arise after having used software over the last 45+ years from mainframes to workstations and from programmable calculators to today's phones. A good example from personal experience was of a piece of code in navigation software (on a mainframe) that was used for about 25 years before the right circumstances produced anomalous results, leading to an investigation, which showed a basic error in coding (the coder failed to remember that the cosine of a negative number is negative, and explicitly changed the sign). This error only resulted in a small difference in results, and only applied in the southern hemisphere. This was in a subroutine that was perhaps 25 lines of code. Self driving software will involve millions of lines of code. If you think that it does not abound in similar errors you are kidding yourself. The same if you think that testing or even real life experience will ever find all of them.

And the difference is that while the navigation software or the Windows update only damages data, in self driving software, software errors can result in deaths or injury.

Where the real benefits come is the ability to communicate instantaneously with other vehicles around, much more effectively than humans can ever manage.

Neither is this an unmixed blessing. If the self driving car can communicate instantly with other cars, so can malicious software. The ability to maliciously take control of steering, accelerator and brakes remotely by hacking into the entertainment system of existing human driven cars has already been demonstrated. Having thousands of cars on the road inviting communication from thousands of others is a prospect that I find really problematic. In virtually every other field of endeavour, every effort is made to shield life critical systems from outside communication.

I'd give my eye teeth to know what the dope 2 cars in front of me is about to do.

I assume he/she is an idiot. It is called defensive driving! And I have seen nothing to suggest that self driving cars will be communicating with that pedestrian about to step off the footpath with eyes glued to their iphone, or the roo in the Canberra suburbs, or even the lycra clad idiot about to run a red light!

Today I passed the site of yet another fatal accident. A young woman died because she made a deadly decision that also affected two other people. Shouldn't young people be given the chance to be out and about without needing 20 years of safe driving behind them?

There is no evidence to date that this is an outcome we can expect from self driving. To expect it as an automatic outcome of the technology is wishful thinking of the worst kind.

I've said this many times before and I'll keep saying it: Self driving tech needs first to be better than inexperienced, tired, drunk, enraged or otherwise incapable drivers to be of real benefit. Then it can get better than that.


And of course reducing death isn't the only criteria, injury is 10x as prevalent.

Of course - but number of deaths is the most reliable measure - it is always measured by the same criterion (alive/dead), and is always taken seriously and actually recorded. And there is no reason to suppose self driving will significantly alter the death/injury ratio.

Self driving cars may reduce road accidents, and hence deaths and injuries, but there is no evidence to show that this is an inevitable result of self driving cars, and there are good reasons, outlined above, to be sceptical. While the results of humans making the wrong decisions lead to accidents, the results of humans making the correct decisions go unnoticed. And any calculation of the ratio of the wrong to right decisions by humans in driving gives a fraction that far exceeds the reliability that has been documented for any software in any application.

In civil aircraft using "fly by wire", where the basic controls are controlled through software, this software is triplicated, and majority rule applies. And the software is not allowed to communicate outside the vehicle. This in a far simpler environment.

In the meantime while statistical data clearly demonstrates that a large proportion of all road accidents are closely related to the use of alcohol and to a lesser extent other drugs, yet there is no will to deal with this, which has a much better prospect of real results.