Hacker News Reader

Some benefits of self-driving cars only come in when they hit certain market-shares. For highway driving there are big gains when you have enough for a dedicated lane, and when cars can "draft" one another, but the biggest gain will be in cities when (if) we get to 100%.

At that point we can do away with all traffic lights except for pedestrian crossings. Not only will stop signs be unnecessary, cars could approach intersections at speed, knowing that there was a gap in the cross-traffic for them to fit through. If you let a single human-driven car onto those streets you couldn't implement anything like that, though.

Yes and we will have flying cars.

The flying cars argument is silly. Yes it was a fanciful 1950s dream that never made it to reality... however autonomous flight is real.. a good portion of commercial air flight is done under autopilot (including landings).

We never even got close to flying car tech... but saying that truly self driving cars that require no dynamic input from humans at any time is something we'll get asymptotically closer to, but will never quite achieve is silly, we're already very close to this reality. I'd be very surprised if in 10 years we don't have a completely computer controlled car.

Why is it silly? Consider this: we've had autopilots for many, may years. But even very limited gadgetry such as automatic parking is a very recent development in cars. And that is an almost trivial scenario.

(The reason the flying car won't be a reality is that all of those (auto-piloted) air planes do in fact have a crew of 1-3 highly qualified professionals -- and the flying cars would be piloted by the sort of people who can't figure out roundabouts. Also, the density of planes is extremely low compared to what would be the situation if flying cars were commonplace).

Also, planes have maintainance regimes that are vastly different from your car. I bet that you did not go through a checklist when you got into your car this morning, nor when you parked it. You probably do not have the faintest idea how the fluids are doing. You probably don't even know if all the external lights work. It is highly unlikely that you even performed the most rudimentary tests of all before getting into your car: walking around it and kicking the tyres. There was no qualified and certified mechanic (for your specific model) to go over your car prior to setting off.

All of this is true.. however you categorically stated that we "will never quite achieve" totally autonomous cars. To me that sounds like a silly statement especially when compared to flying cars (a totally different class of problem).

The fact is that we're very quickly edging towards fully computer controlled cars. You have to have a truly limited imagination to think we will never get there.

The argument about airplane maintenance also doesn't hold water. Result of a malfunctioning plane: plunge 30,000ft and hundreds dead. Result of a malfunctioning car: coast to a gentle stop. See any difference?

I was being imprecise. Sorry about that. I don't think that we will see totally autonomous cars _in_our_lifetime. "Never" is not a useful timescale.

  You have to have a truly limited imagination to think we
  will never get there.

40 years ago people did the same sorts of extrapolations on air-travel and figured that 15-20 years later all commercial air travel would be supersonic. And why wouldn't they? Orville Wright was still alive when the first supersonic flight took place. We went from not being able to fly heavier-than-air machines to breaking the sound barrier in 43 years. But of course, that didn't happen.

And in just the past few years we've gone backwards. There is currently no supersonic passenger jet in service.

A lot of domains appear to exhibit the same asymptotic behavior. Space travel, medicine, AI etc. Domains where initial speed causes undue optimism and where we later suffer regressions. (No regressions in medicine you say? Well, how about the crazy anti-vaccination people and the re-emergence of diseases that were practically eradicated?)

I build stuff, I program stuff, I occasionally build and race cars. I don't think my skepticism stems from _lack_ of imagination -- I think it has more to do with an abundance of imagination.

  The argument about airplane maintenance also doesn't hold water. Result of a
  malfunctioning plane: plunge 30,000ft and hundreds dead. Result of a
  malfunctioning car: coast to a gentle stop. See any difference?

I'm sorry, but that is pure nonsense and many, many, many people pay the ultimate price to prove you wrong every day.

I don't think that we will see totally autonomous cars _in_our_lifetime

Now you're moving the goalposts to suit your argument.

Supersonic flight was abandoned because it is expensive and prohibitively noisy. Will some set of obstacles also arise to stymie autonomous vehicles? Maybe, but we wont know till we try.

The regression in medicine is due to pure stupidity. Will the same or similar type of stupidity raise its head to block self-driven cars? Again the answer is maybe. Maybe not. We wont know till we've tried.

many, many, many people pay the ultimate price to prove you wrong every day

You misunderstood what I said. My point was that if a computer controlled car's computer fails it can be made to fail-safe. Shut down and coast to a stop. The same cant be said for planes.

Of course we will have scenarios of computers being mis-programmed and causing death. The point being that once a bug is found and eliminated it wont reoccur. Not to mention that a human will never drive as well or recover from trouble (such as a fishtail) as well as a computer. Our interfaces are simply too slow, clumsy, prone to fatigue, prone to mis judgement due to adrenaline etc. the list goes on.

I get what you're saying about optimistic extrapolation being a trap... but we don't have to extrapolate far from where we are today. The google self driven cars have already driven tens of thousands of miles with the only recorded accident occurring when the vehicle was under human control.

Not only that but Nevada has already licensed the cars for testing on its roads. Now you can argue that the testing is not 100% authentic. From what I understand the routes are pre-programmed and 2 occupants must be in the car at all times. But to say that this tech wont go anywhere for the next 50-60 years (however long you intend to live :) ) ... seems to lack imagination.

This isn't everyone using personal jet-packs or interstellar travel we're talking about here.

> You misunderstood what I said. My point was that if a > computer controlled car's computer fails it can be made to > fail-safe. Shut down and coast to a stop. The same cant be > said for planes.

Just because you are at ground-level you are not safe if you lose control of the vehicle.

> The point being that once a bug is found and eliminated it > wont reoccur.

We already have a software industry and we know that this doesn't happen: bugs do not occur only once and then get eliminated forever. There are people making a living writing books about mistakes that people repeat over and over in software.

"There is currently no supersonic passenger jet in service."

The dream isn't dead yet:

http://www.aopa.org/aircraft/articles/2012/120426back-to-a-s...

I've come to realize that autonomous cars are a pre-req for flying cars. Obvious now that I see it, of course. Humans as a whole can't possibly handle flying machines when they have so much trouble with cars.

I'm skeptical about totally autonomous cars, but I think freeway driving will be increasingly automated. Thrun has said that he's let his car drive him from the bay area to Reno several times. I assume he's exaggerating a bit, but probably not by much. It's a solvable problem for good road conditions.

Right on!

There is no way a mass market for personal air vehicles can happen if we need traditional pilot's licences. Following the eventual success of driverless cars, one can imagine that all-weather PAVs will not require a license because we will just be passengers.

See: http://www.boeing.com/news/frontiers/archive/2004/july/ts_sf...

Your argument allows for certain driverless cars in which a human is present to provide dynamic input for situations that require it. This is currently the legal requirement in Nevada.

Admitting that driverless cars cannot be truely perfected, does not preclude mass adoption - because they merely need to be safer and cheaper to operate than the current human-only traffic situation. I think we are less than a decade away from that tipping point. Does anyone else agree?

I would be careful to conclude without data. For instance insurance companies have data that suggests that systems for warning the driver of lane drift make cars significantly more likely to have accidents. I can't remember if they had figured out why, but this is the sort of counter-intuitive results you have to expect when you are dealing with a field that isn't adequately understood.

And large populations of autonomous cars in obviously something we have zero experience with so we can't really know if there will be unexpected dominant factors.