In another thread (I don't remember which) we discussed autonomous drones and their safety in normal air traffic. I wanted to get back to that thread but I forgot. So here is a video illustrating why I believe that - in general - the problem of autonomous vehicles has been solved:




Aerial vehicles would need a pretty good natural language processor to interprete air traffic controllers but that is a problem we (computer scientists, not me personally) are solving right now.

---

http://www.politicalcompass.org/test
Economic Left/Right: -7.12
Social Libertarian/Authoritarian: -7.74

This is pre-WWII. You can sort of tell from the sketch style, from thee way it refers to Japan (Japan in the 1950s was still rebuilding from WWII), the spelling of Tokyo, lots of details. Nothing obvious... except that the upper right hand corner of the page reads "November 1931." --- Simon_Jester

Communications between air traffic control and pilots is not natural language, it's actually extremely formulaic. Provided you can get the human element (if any) in the system to cooperate with sticking to the formula, and you have a robust voice recognition program, we're probably there already.

I do like the driverless car concept, and I'm quite happy it's undergoing thorough testing at this point. I'm curious as to how it handles extreme weather, mechanical failures, and accidents involving other vehicles in it's path. Not that most humans are fantastic at those tasks, either, but the system needs to be at least as good as a competent human driver.

There are some significant differences between flight and driving. I don't think they're insurmountable, but what's suited to one may or may not be applicable to the other. The acceleration/turning/etc. instrumentation mentioned as being in the boot of the car is actually from aviation, it's been around for decades in the airlines so that's a fit. Likewise, GPS was used in aviation before automobiles. On the other hand, driving is primarily a two dimensional navigational problem and flying occurs in three dimensions. Will the computer system be able to keep up with the additional demands imposed by adding the third dimension? (Well, if not now then presumably in the near future as computing power/speed still seems to be rising). What is the range of those lasers? A "close" encounter between aircraft is roughly a half kilometer which, under some circumstances, might even be called a "near miss" at that distance which is a different scale of "close" and "far" than dealt with by automobiles. Flight speeds are also (once you get past the small airplanes I fly) significantly faster than those used by automobiles. I am assuming the system will be able to handle that but prior to real-world application we would need to be sure of that.

Road infrastructure also would have to be upgraded to acomodate driverless cars. Although it would be possible to make a system that can visually recognize road signs it would be much more error proof to transmit the info road sign contains wirelessly directly to car.
Another big issue could be making the system to recognize various situations a good human driver can spot well in advance that could potentially lead to accidents. For example a responsible human driver will slow down/be extra careful if he sees kids playing near the road, another driver or pedestrian behaving erratically, a deep puddle of water on the road, someone backing out from parking place and tons of other unpredictable random events happening on and near the road.
Bad weather also could be a problem, even such simple situation like ice forming on some laser sensor causing faulty data to be fed into computer.

Hm, wouldn't all three problems be effectively solved by some sort of datalink between vehicles? That way, cars in front could notify ones in rear about road signs, bad traction, roadside activity, and all - without major infrastructure upgrade and using tech we actually developed (say, WiFi). Of course, there would still be issue about cars driving alone - but then, maybe some sort of repeater stations holding info from past cars could be fitted into roadside lamps? In any case, cars sharing data would probably be good idea anyway, such as giving cars behind you a warning you might be forced to brake suddenly instead of having them react to you doing it.

As for other situations... Maybe just have 'caution' button operated by passenger forcing the car to slow down, along with 'panic' button for actual emergencies forcing the car to stop? Of course, some care and forethought would be needed to prevent traffic trolling/abuse.

That would certainly help alot. I expect most problems will be when driverless cars are first introduced commercially and when they have to drive along with human drivers. If you could magically at once replace all regular cars with driverless cars able to share data with each other it would certainly be much easier because they wouldn't have to deal with hard to predict behavior of human drivers driving "dumb" cars.

No, it wouldn't. It wouldn't warn of sudden appearance of pedestrians running into traffic, or animals doing the same.

That's one reason I was curious about the range of the sensors. Can it see a kilometer or two? If so, it's going to be detecting things about as well as a human driver, except that it has fewer/no blind spots. But can it anticipate? If I see kids playing right next to the road I might slow down considerably, but not so much if it's adults standing around talking. Can this software make that sort of judgement call?

If the car ahead of the driverless car spins out on, say, a snowy road how well does it react to that emergency?


That only works where there are roadside lamps.

It may be that initially driverless cars are an urban feature so that would be a minor point, but traveling long distances in rural areas means roads that are pretty minimal in many ways, including handy infrastructure to hang repeaters off of.


That kind of gets away from the idea of the car not needing a driver. And if you think people are distracted NOW, wait until they don't need to steer anymore.

The panic button makes sense - in a way, we have the start of that with services like OnStar. The "slow-down" button less so, because very shortly the passenger behind the wheel of a driverless vehicle is going to stop paying much attention to the road - after all, that's the car's job, right? You need a system with some judgement ability.

It'd also be vulnerable to tomfoolery and we wouldn't want that. At best, you could use a data link to broadcast intent like a turn signal, but as the video demonstrates, it already recognizes turn signals.



There is another video where someone steps out in front of the car. The car stops - but the engineer points out that it already saw the person by its planned course, so it slowed down to prepare for that possibility. In this video you see it handling pedestrians, a deer darting out in front of the car, and rather clearly detect a car spinning out behind it.

---

Give fire to a man, and he will be warm for a day.
Set him on fire, and he will be warm for life.

I guess one would go with radar for the longer distances. You just don't need centimeter accurate meassurements.

---

http://www.politicalcompass.org/test
Economic Left/Right: -7.12
Social Libertarian/Authoritarian: -7.74

This is pre-WWII. You can sort of tell from the sketch style, from thee way it refers to Japan (Japan in the 1950s was still rebuilding from WWII), the spelling of Tokyo, lots of details. Nothing obvious... except that the upper right hand corner of the page reads "November 1931." --- Simon_Jester

On that note; How to hack into cars and disable the breaks.

I wonder if there might be a stage between fully driver-operated cars as we know them today and fully driverless ones. I can imagine a car that won't let you drive through a red light, or exceed speed limits except as necessary for overtaking, or park illegally. Such a car might have a limited driverless capability if you took your hands off the steering wheel, perhaps just enough to find the next safe place to park. Even if a fully driverless car was developed, it might still have this kind of 'driver assistance' mode. People unable or unwilling to drive would let the car drive itself, while others might only use this capability to have the car let them off in town and then park itself, or if they were intoxicated.

Yeah, that's what I see as a reasonable prediction, as well.

---

http://www.politicalcompass.org/test
Economic Left/Right: -7.12
Social Libertarian/Authoritarian: -7.74

This is pre-WWII. You can sort of tell from the sketch style, from thee way it refers to Japan (Japan in the 1950s was still rebuilding from WWII), the spelling of Tokyo, lots of details. Nothing obvious... except that the upper right hand corner of the page reads "November 1931." --- Simon_Jester

[youtube]https://www.youtube.com/watch?v=_sUeGC-8dyk[/youtube]

---

http://www.politicalcompass.org/test
Economic Left/Right: -7.12
Social Libertarian/Authoritarian: -7.74

This is pre-WWII. You can sort of tell from the sketch style, from thee way it refers to Japan (Japan in the 1950s was still rebuilding from WWII), the spelling of Tokyo, lots of details. Nothing obvious... except that the upper right hand corner of the page reads "November 1931." --- Simon_Jester

god I hate this aspect of the board software... could a mod delete the other post please?

---

http://www.politicalcompass.org/test
Economic Left/Right: -7.12
Social Libertarian/Authoritarian: -7.74

This is pre-WWII. You can sort of tell from the sketch style, from thee way it refers to Japan (Japan in the 1950s was still rebuilding from WWII), the spelling of Tokyo, lots of details. Nothing obvious... except that the upper right hand corner of the page reads "November 1931." --- Simon_Jester