
Originally Posted by
JDNSW
While active protection in modern cars undoubtedly have some effect, in my view this is overrated as a reason for the dropping road toll. Statistical data compiled by Monash shows no discernible relation between safety features of modern cars and safety outcomes when you look at overall statistics. If you look at, for example, survival in specific types of accidents, yes, but when you include the accident frequency, the correlation is lost.
In my view the major reasons for improvements (apart from seat belts and random breath tests, which did show up in statistics) are a steadily improving safety culture - for example, in 1970, someone booked for drink driving was considered unlucky - today they are considered a fool at best by most people. (Still a lot of room for improvement!), and vastly improved roads. You only have to look at the accident rates per vehicle - kilometre for divided roads compared to undivided roads, despite the generally higher speed limits on divided roads to see the vast difference road construction can make.
And even without these expensive changes, I look back and see a lot of minor improvements that add up - for example, edge marking, rumble strips, offsetting country crossroads, improving corners, better safety railings, replacing single lane bridges, that sort of thing.
I also had seat belts from that place in Chippendale - Light Aircraft Supplies, or something like that, from memory.
Driverless cars are, I suspect, a long way off, if ever. One almost insuperable problem is the legal framework, but another is that ordinary drivers are actually very good at decision making - drivers make tens of thousands of decisions every day, and the accident statistics show that a very, very, large proportion of these are made correctly. The "getting it right" decision rate for human drivers, it seems, is greater than the rate of fault-free performance achieved from any type of software.
For self driving software to be accepted as reaching the level of reliability of a human driver, it is going to have to be tested in a very large number of different circumstances. (Tesla in fact are doing this, as the main reason their cars call home!) But then the software (and the hardware supporting it) must be protected in some secure way from change or modification without a similar rigorous testing regime being applied to the modifications.
I am sure some will point out that this has been done successfully in airliners. Two points - aviation "self driving" is a far simpler problem than driving on the road, and airliners cost tens of thousands of times more than do cars, so have much more effort put into their design. And even then, think Boeing MCAS!
Bookmarks