Hacker News Reader

Comment by chrismealy | Hacker News Reader

The main problem is totally technical! Robot cars can barely deal with following white lines on the road. How about rain, construction, poorly marked roads, fallen trees, badly parked delivery trucks, emergency vehicles, pedestrians, cyclists, etc? Can cars read roads signs and understand them? We'll have full-on AI before self-driving cars are for real.

View on HN

I've been parroting this on every self driving thread. I finally found one person who agrees with my thinking that driving is as hard as general AI.

Why is this idea so opposed by everyone?

At least we have each other.

Honestly, I am completely baffled that people who write software think self-driving cars are just around the corner. At this point I think it's basically a futurist cult.

There are some of us lurkers who agree with you wholeheartedly.

This the general consensus in the academy too.

99% of situations are the "easy" part. It is the last 1% that will take a long time. Snow. Rain. Missing headlight. Etc.

Can't we just accept that the car won't drive itself in the snow or rain or with a missing headlight?

I'd find that totally reasonable, and get back behind the wheel to take my own risks in a snowstorm (or ... you know... stay at home like the advisories always tell you to).

I would absolutely not find it reasonable if my "self-driving" car won't drive itself in the rain or snow, given that in my climate, it is raining or snowing >270 out of every 360 days of the year.

Such a product would still have a large market value since there are nice climates with dense populations in the world too, but not everyone lives in California.

What you said -- plus: the last thing you want is a car that 'self drives' until the conditions suddenly change at high speeds and the half-aware driver is forced to take over unprepared and unalert and convinced that his/her car is supposed to be driving for them.

Exactly, if we are forcing drivers to stay focused on the road with hands on the wheel. It's not "self driving", as in it's not level 5 autonomy

I actually live in Virginia. We get snow and rain just as much as any where else in the N.E. USA.

However our yearly precipitation is ~132 days, where each of those isn't a straight 24 hours of rain or snowfall.

I wouldn't find it terrible to buy a self-driving car that runs 2/3rds of the year without any problems, and I need to drive personally if and only if I must go out during heavy rain or snowfall instead of waiting 1 hour for the rain to stop.

Right now, many times I just wait for the rain to stop before going out because there isn't anything pressing or time sensitive about needing groceries on a Tuesday.

Yeah, and you can easily handle driver negligence in the easy part with always-on driver assistance. Plus it's unacceptable to allow humans away from the steering wheel when they are needed 1% of the time.

This is assuming that "self-driving cars" isn't just another name for driver assistance packages and is actually level 5 autonomy (which a few of my friends believe will be on the road this year with Tesla model 3)

FWIW I agree with both of you and say the same thing.

It's in particular a California cult, because they don't actually have weather there.

Anybody who has dealt with driving in a full blown northern winter is likely more skeptical.

General AI can drive but a self driving car doesn't have to do everything a general AI can do therefor driving is easier than general AI.

A fully autonomous self-driving car (if that is what you mean by self-driving) will have to be capable of reasoning it's way out of every possible situation, which requires general AI.

The sheer number of cars used and the variety of places they're used in means that thousands of cars get caught up in bizarre situations every day. Cars will be shutting down inconveniently all over the place unless they can apply general reasoning skill to navigate through the innumerable trials that drivers currently have to cope with.

I can think of a ton of everyday problems that will require actual AI: Following detour signs. Understanding a cop's hand signals. Understanding crossing guards in school zones. Making room for emergency vehicles on crowded streets. Driving on dirt and gravel. Navigating parking garages (ever notice that signage in garages isn't standardized? Probably not, because you're not a robot).

Most of the self-driving crash videos I've seen involve a car that lost it when it couldn't follow the white line. Following the white is not good enough.

Because we have self driving elevators, self driving trains (the DLR), driverless busses (http://www.telegraph.co.uk/news/2016/09/03/worlds-first-driv...)

and shortly self driving Volvos https://www.youtube.com/watch?v=2q00jIBhkq4

None of those have human level AI. There's a lot you can do in the real world without that.

It's interesting that Volvo seems, rather than out AIing their competitors to have gone to town on the sensors with several radars, cameras and lidar.

Elevators and trains are not even in the same ballpark as cars, and the busses you linked to can't function on public roads. They don't have the ability to interact with other vehicles.

It's entirely valid to state that we can do lots with our current tech, because we can, but when the topic is self driving cars on public roads, it makes no sense to claim they're possible without general AI based on the fact that an existing bus can drive around a closed course.

Giving those busses the ability to navigate arbitrary situations on public roads is not simply an incremental improvement, it requires that you solve a fundamentally different problem.

The Volvos drive themselves on public roads. OK level 4 not 5 and not quite launched on the public yet but not bad.

Yeah, but you agree the problem is technical, not regulatory.

I have multiple friends with years of experience in academia who think that Tesla is launching "Tesla Network" - driverless taxis this year.

Non-sense. Full-on AI is completely another level compared to self driving cars. It is like saying that we'll have full blown image recognition before pong.

If by self driving you mean driving in a single lane on a highway, then sure, a lot of cars have "automated cruise control" with "lane keeping assists" and "collision avoidance".

But if you mean truly driverless cars on public roads, with noone in the driver seat, then sorry, driving isn't just riding in a lane and reading a handful of road signs. It includes responding to anything in your environment. For example, if you see a ball bounce in front of your car, you will instinctively slow down since there might be a kid running after the ball. A driverless car making that inference? No chance.

Reply on news.ycombinator.com