David Lochrin wrote: It's interesting to see this topic surfacing again so soon, it obviously > excites passions. >
Isn't it? If cars are to be completely computer-controlled by law <snip> This is a possible endpoint at the moment. It's not a real issue. That would only happen if cars meet all the requirements you listed and a lot more. > at 3am Computers don't suffer from drowsiness like humans > over unlit rural roads which may be wet, unmarked, or unmade. Driverless cars can already do this. > kangaroos and other animals, washouts And this. > How is the computer to navigate the journey? > Cached maps, GPS, cameras, sensors, etc. > It would seem navigation data must be either (a) locally read from > information associated with the road such as road-markings, or (b) obtained > externally from a GPS or other system, or (c) obtained by an on-board AI > system Yes > which is at least the equal of the human driver who has traveled the same > road many times before. > Robotic driving systems are trained against millions of complex situations. > Case (a) requires _every_ single road & track in the country to be > equipped with the appropriate infrastructure. No, machine vision can already determine where the road is, its condition and any obstructions, stationary or moving, alive or dead. And read speed limit signs. > Case (b) requires a GPS or inertial-navigation system far ahead of > anything we have now and there would still remain the problem of mapping > inaccuracies & updates. Case (c) requires an AI system far ahead of > anything we have now together with some form of positional data per (a) or > (b). > Have you seen what autonomous vehicles can do? > And then there's cost... > If it is too expensive it won't happen. Computers and sensors are cheap once the product works and commodified. Cheaper than humans. Design and development are slow and expensive. As I have said before, robotic systems get better every year. They learn. A robotic driver system does not have to be smarter than a human, all it needs to do is *drive a car* - better, safer and more reliably that the average driver. It doesn't need to be able to chat in an lively manner about "the roo that jumped out." It just needs to know how to miss it if possible without compromising safely. This is more than a lot of human drivers can do. There is no reason to believe that robotic drivers won't exceed human drivers. If there is a limit to how good they can get we are nowhere near it yet. If at any time they aren't fit for any driving task they simply won't be used. If it is important enough to pay for it will be developed. Right now, humans can beat robots at a lot of driving tasks. The list will certainly get smaller and I expect it will disappear. I can programming a computer to play tic tac toe was an achievement. Then they mastered checkers but chess was considered too complex. Chess programs played like morons. Then weak players. Then they could beat any human chess player but Go was too hard. Now they are beating the best human Go players. We all know the story. Jim _______________________________________________ Link mailing list Link@mailman.anu.edu.au http://mailman.anu.edu.au/mailman/listinfo/link
