Slashdot Mirror
Philosophical Differences In Autonomous Car Tech
An anonymous reader writes: The Guardian has an in-depth article on the status of self-driving car development at BMW. The technology can handle the autobahn just fine, for the most part. But the article highlights philosophical differences in how various companies are developing self-driving tech. European and Asian car manufacturers are fine working on it piece-by-piece. The car will drive itself when it can, but they expect drivers to always be monitoring the situation and ready to take control. Google's tests have taught it otherwise — even after being told it's a prototype, new drivers immediately place a lot more trust in the car than they should. They turn their attention away and stop looking at the road for much longer than is safe. This makes Google think autonomous cars need an all-or-nothing approach. Conversely, BMW feels that incremental progress is the only way to go. They also expect cars to start carrying "black boxes" that will help crash investigators figure out exactly what caused an accident.
In related news, Google is bringing on John Krafcik as the CEO of its self-driving car project. He has worked in product development for Ford, he was the CEO of Hyundai North America, and most recently he was president of Truecar.
Really? Hmm...just one MORE thing to have spying on me and my habits? I guess now I need to learn how to remove said black box, or if it can't be removed, to rig it to be "destroyed" after a wreck and make it plausible that the accident caused the problems?
Really I just don't want to be tracked any more!!! Fitbit to see how I live, black boxes to see my driving habits, using my cell phone to track my movements.
Enough...FUCKING ENOUGH!!
Light travels faster than sound. This is why some people appear bright until you hear them speak.........
Expecting a driver to take control in a failure scenario is not a solution.
I would never trust a car that could require me to take control in an emergency. At the very least, the autonomus driver should get the car to a safe stop before requiring a human to take over.
Expecting a driver to take control in a failure scenario is not a solution.
I would never trust a car that could require me to take control in an emergency. At the very least, the autonomus driver should get the car to a safe stop before requiring a human to take over.
I agree that "expecting a driver to take control in a failure scenario is not a solution" and stopping is an acceptable solution but it doesn't necessarily
need to be an all or nothing. A better piecemeal solution would be to have it only engage on known safe highways. It would still be extremely useful
in trucks, RVs, and regular cars if it only engaged on predesignated roads or interstates. The trucking industry already has depots at both ends of
Kansas where trucks double or triple up before taking the long straight stretch across Kansas to minimize drivers. I see no reason why driverless
cars couldn't do the same where you could only engage autopilot on certain known safe highways with good shoulders to do emergency stops.
You could also do the same with weather. If it detects rain starting then it gives a 60 second warning and pulls over to the side of the road.
Expecting a driver to take control in a failure scenario is not a solution.
Why not, that is the design philosophy for airliners made in the past 30 years and their pilots who operate them. Sit and babysit the machine for 99.99% of the time; then jump in ready to go for the 0.01% of the time the situation is beyond the programming of the software. (In which case the software 1) does wrong thing. 2) just shuts-down while displaying a message to the pilots to let them know that, suddenly, THEY are flying the plane.)
The problem with this is that when an airline pilot is forced to take control, they probably have MINUTES before any real issue will arise.
They are asking car drivers to take over when there are possible issues within SECONDS (possibly less).
People aren't good at driving and make bad decisions in emergency situations as it is. Now you want the person to have 99.99% less experience behind the wheel, and yet be capable of doing the right thing when a tricky situation is suddenly thrown at them with half a second notice? Are you sure you don't see the problem with this?
We hope your rules and wisdom choke you / Now we are one in everlasting peace
The airliner scenario is only superficially similar.
At cruising altitude, a plane typically has minutes before it crashes to the ground. For example, from the time its problem began, to the point in hit the ocean, Air France flight AF447's pilots had 3 minutes and 30 seconds to try and save the plane. There are typically few, if any other planes in its airspace to worry about, so pilots can do things like take our their operating manuals and run through operating prodecures to attempt to rescue the situation without worrying about hitting the kerb, another plane, etc. If my self driving car is going to give me 3 minutes before the actual crash, then fine. Otherwise, it is less than useless to give the control to a driver who likely doesn't have the correct situational awareness (who might even have fallen asleep).
Even if the driver had not been sleeping, a driver's awareness is reduced because he doesn't have to process what is happening around him all the time like one does when they are driving. So, for example, if the problem is that he is about to crash, unless he was hyper vigilant, he is the worst person in the world to drop into the driving seat so to speak.
If you want to see where the state of the art is at (at Google, anyway), have a look at this video. The first part is high-level info about Google's approach to the problem; the actual demonstration of the software starts at the 7m50s mark, and an example of the car dealing with something unexpected is at the 11m00 mark.
Note that in none of the examples has the roadway been pre-fitted with any kind of tracking or placement system, and that these aren't imaginary scenarios; rather these cars have been on the roadways, in real life, for years already. If the problems were really as unsolvable as you suggest, I'd expect we'd have read about a number of hilarious and/or tragic mishaps by now.
I don't care if it's 90,000 hectares. That lake was not my doing.
> better piecemeal solution would be to have it only engage on known safe highways.
Also, what we're already seeing is more and more driver assist. My 2012 Dodge has an option for "smart" cruise control where is slows down if you're getting too close to the car In front. Most cars these days have traction control, where the computer automatically brakes the wheels independently in order to turn the car in the direction the steering wheel is pointed. Do we already have systems that will nudge the steering a bit when you start to drift out of your lane? If not, that could be added. Not overriding a clear steering input from the driver, just a slight torque so that the existing self-centering action of the steering wheel follows the lines which mark the lanes. In other words, with today's cars, if you let go if the steering it'll tend to go straight ahead. Mayb with tomorrow's cars if you let go of the wheel they'll TEND to follow the lane.
On my 2012, the headlights automatically turn on and off as needed.
I could see more and more of that stuff being added, stuff where the computer insures that the car does what the driver wants/expects it to do. Eventually, you slowly get to the point where "what the driver expects" is defined by the destination they select in the gps.
And I'd absolutely disagree.
The first step for autonomous driving would make sense to be implemented ONLY on long-stretch highway drives, with strong signals many minutes before exit-destination arrivals and a "pull over and stop" system for drivers that don't respond/wake up.
To suggest that driverless cars have to be able to cope with every conceivable situation is totally unreasonable. Hell, HUMANS can't cope with "every conceivable situation", really.
-Styopa
The bigger issue is that pilots spend hundreds of hours practicing those emergency maneuvers. Car drivers not so much. Pilots have strict rules about how many hours they can fly. Automobile drivers, not so much.
If we held automobile and truck drivers to the standard we hold even private pilots, there would likely be many fewer accidents. But we don't.
Faster! Faster! Faster would be better!
That's not exactly correct... the Google cars have incredibly precise maps of the roads they're on, not just the route, but maps of the actual surface of the road (e.g., where the potholes are). That level of detail available to the onboard computers is pretty much the same as having sensors on the road. It requires an incredible amount of prep work. Of course, map updates could be handled by sensors on other cars constantly providing real time information. It's a cool approach, but only practical when you have that level of detail available.
Google, et al, are showing very controlled research projects. Even though they're testing in the real world, they're still highly controlled experiments.
Sure, many of the problems are resolvable using this approach, but what we don't know is what new problems will evolve once there are more than a handful of self driving cars on the road. More research will help identify these, but anyone who's done real science or engineering knows that what works at small scale rarely scales as you would hope/expect.
-Chris
There's nothing that could ever satisfy that test. A kid who runs into the road is going to get hit by a car in ordinary circumstances. If he's incredibly lucky or the driver is heroically responsive and saves the day, the kid won't get hit. No system, human or otherwise, can ever be created to solve this situation satisfactorily. The best hope is probably automatic braking.