now all we gotta do is add neutron laser beams in the headlights and we can get all sorts of military funding to make this real.

pisses me off that we spend ungodly amounts of money on wars of choice instead of stuff like this.

*quality is job #1.

Boosting this quality contribution up in the Hot Listing - declared quality by Lendl.

I've been wondering about autonomous vehicles... mostly about the cases that may come up. Like who would be liable if one causes damage, injury, or death? Because if the manufacturer is going to be liable, companies will likely be slow to bring these to market.

No footage of it in real world traffic.

O RLY?

*related=http://videosift.com/video/Self-Driving-Cars-Are-Getting-Smarter

Self Driving Cars Are Getting Smarter has been added as a related post - related requested by ChaosEngine.

I think the proper response is, an insurance pool, combined with extensive testing.

Let's be honest, these don't have to be amazingly reliable to be safer than humans. Not to mention how many wasted hours could be reclaimed, when commuters don't have to drive themselves.

Still not sold on this, I'm still feeling like it's an accident waiting to happen>
And Dammit, there are times, you want to slow down, and times you want to speed up. this would do neither...

There's going to be an interesting transition period from self driving cars that must have a competent driver that can take over, to self driving cars that are legally autonomous (i.e. doesn't need human oversight).

The ultimate end IMHO, is roads that only allowing AI vehicles. Think about it, computers are much faster than humans. if you had a road where all the cars were networked, you could have high speed traffic with perfect reaction times and no human error. At that point, allowing a human driver would be dangerous.

It's all fun until the hackers get at them

That's going to be the time the cops (or anyone else with access) can reroute or stop a car at will, without the passengers having any say in it. Of course, taking this thinking to the logical conclusion, humans wouldn't have much input in the whole system in the first place.

I think I'm going to go watch Colossus: the Forbin project now...

This just seems too far fetched at the moment. The number of variables and situations to consider, the speed of execution required, the level of reliability ... it just seems too much for today's processing power.

I mean, take something like speech recognition as an example. Computer processing continues to fail at even this. The visual spectrum is much more complex than the auditory, which is largely just wavelength and amplitude.

To jump straight into something like that, with life or death ramifications ... I just don't see them jumping to production any time in the next decade.

Taken to it's logical conclusion (where all road vehicles are automated) things like processing visuals becomes a lot simpler.

Most of the information will be preprocessed data - the road infrastructure, and knowledge of every other vehicle in the system. That's what it becomes - one homogeneous and optimisable system; not a single robot car making it's own decisions. Processing is done by servers in advance, not individual vehicles.

Real time visual processing then only has to be on the look out for things that are out of the ordinary, such as a deer running onto the road, and take caution. Something that is much easier to do when done with well coordinated cooperation with every other vehicle around you. The car behind isn't going to hit you; nor is the oncoming car in the opposite lane.

Even if every car in the situation doesn't have the stopping distance, they will all turn (where possible) in a coordinated manner to avoid or reduce collisions. The situation may be avoided in the first place, because the system knows about dangers specific to this stretch of road and will have limited speed and increase distance between vehicles appropriately.

If there is one thing that humans absolutely suck at, it's unspoken mass cooperation in an emergent system; our roads being the prime example.

People can't even keep the appropriate distance from the car in front to prevent collisions in an emergency stop; let alone leave enough distance to prevent traffic jams. Imagine a red light (real or virtual) turns green, and every car for a mile back accelerates at the same time. There's no queuing (the cause of jams).

Emergency services get a clear, safe and direct path to their destination.

Who knows, maybe the system even knows you have a flight reservation or a business meeting, or if you're just going to the mall or visiting friends and can prioritise intersections accordingly (though that is getting a little Orwell meets Skynet).

The hard part is going to be getting from our current fully manual and emergent system to a fully automated and coordinated system; but I'm really excited by the prospect it might happen in my lifetime.

one benefit of a fully computer controlled traffic system is that when the computer knows exactly where all the cars are at, it becomes safer to go MUCH faster

only reason we have speed limits is because you simply can't trust a human to drive safely and react in time to someone else's stupid decision.

not only do we not have to worry about reckless drivers, we can get to places much faster...which is ironic because most reckless drivers are trying to get to their destination faster than everyone else. it's because of their arrogance that we have to slow down so we don't kill each other.

Where's that video of all the cars going in a circle after being told to drive at a specific speed, and gradually slowdowns would appear, actually coming to an almost stop.

Betcha if they replaced them all with gCars they would continue forever.

A road system entirely populated by self-driving cars wouldn't need stop lights or speed limits. Intersections would just "Work" on a yield-to-demand basis. If you have 75% more vehicles going East-West than North South, the intersection timing would adjust to letting 3x as many cars through every switch one way than the other.

An interleaving intersection with 1:1 non stop traffic would be scary, exciting and fun to be a part of. Especially at high speed!

That's why we test. That's why this test is not in traffic.
Who is jumping straight in?

Perhaps not the best choice of words, but what i mean is the level of accuracy and reliability required seems unattainable.

Voice recognition which I think of very comparable, requires AI 'training' and a stable voice manner to be usable. Take Google's integration of this into YouTube and the garbage subtitles it spits out as an example. This is a technology which has been around for a decade and is still woeful unusable outside limited scenarios. The multitudes of increasing processing power over this period have apparently not made much difference.

If we can't master that in 10 years, how are we going to master automated driving to the perfection required to make millisecond life or death decisions?

But millisecond life or death decisions are what computers excel at. Unraveling the vagaries of human speech is a different problem. And the vagaries of human vision another.

Reaction times yes, but I think having a sufficient degree of certainty that the correct decision will be made is hard to conceive.

Imagine the legal liability of a clear software failure. Even if average accident rates were lower for automated cars, a clear incidence of failure would be a huge monetary legal risk. Whereas, if legal exceptions were carved out for the likes of Google, I doubt there would be very good consumer uptake.

I would suspect their automation algorithm are highly based on visual inputs. Pre-available GPS mapping data would get them only so far. These visual inputs are hugely variable. The number of different car makes, times of day, weather and road conditions among other things, would make for a incredible amount of scenarios to envisage.

I think voice recognition is very similar, if anything more constrained. The deciphering of combination of pitch, accent and pronunciation is a far simpler and smaller domain that we haven't mastered. That would seem to me to be demonstrable proof that automated cars to the level of reliability we would expect, are currently inconceivable.