[ingenter]

There is a significant difference between self-driving cars in 1960, flying cars and self-driving cars today.

Flying cars are doable, and there are several prototypes of a flying car. However, flying cars aren't safe and aren't cost-effective.

Self-driving cars in 1960 require a significant infrastructure to operate, and can't operate with human drivers. (Edit: Not to mention they don't exist)

Self-driving cars today work, and are cost-effective, and it's likely that they will be safer and create a much better traffic ecosystem than humans ever will.

[Hydraulix989]

"Self-driving cars today work, and are cost-effective, and it's likely that they will be safer and create a much better traffic ecosystem than humans ever will."

Self-driving cars ONLY work on the immaculate freeways of California that they are almost always (unfairly) tested on. Any precipitation renders the LADAR sensors all but useless, and many everyday driving scenarios like "left turn onto traffic" and "no lane markers" are still are far from being solved yet.

Any real improvements in the technology will be the result of fundamentally different techniques than what the state-of-the-art is currently using (since this is Hacker News: state-of-the-art really is just an "ad hoc" pipeline that looks for things like "lane markers" using things like Canny edges along with a PID controller for the steering wheel actuator, with some other "cheats" like an over-reliance on human-compiled map data provided "a priori").

An end-to-end deep learning approach seems promising, but the current results aren't even usable at this point.

Five years from now, every car (even my Jetta, not just luxury cars) will have the equivalent of what Tesla's "auto pilot" looks like now, but a human in the front seat will still very much be a necessity.

[Animats]

The first step to good self-driving is full profiling of the terrain and road ahead, along with any obstacles. We did that in the DARPA Grand Challenge, and so did almost all the other teams. That's essential to off-road driving. That tells you where you can physically go. That was figured out by 1985. It's mostly a sensor problem.

Next comes recognizing where the road wants you to go (lane markings and such), which works reasonably well now, especially if you have mapping information. That's automatic lane keeping. All the major manufacturers have that working.

Then comes dealing with other road users. Google is putting a lot of effort into this, with some success. See Urmson's video from SXSW.

Looking ahead, we need somewhat better and much cheaper sensors. The rotating Velodyne thing is still too clunky and too expensive. LIDARs that deal with rain and fog can be built; you need to get back more info than just the first return. "First and last" return is helpful; you'll get a solid "last" return from a hard obstacle in the rain, while "first" looks like noise. That technique is used in aerial LIDAR scans to get both the top of vegetation (the first return) and the ground surface (the last return). It's also possible to range gate through fog. Here's some range-gated imagery.[1]

Solid-state flash LIDAR may be the way to go. Units today are still about $60K, but that's a consequence of low production quantity. The custom imaging ICs aren't inherently expensive. There's a startup claiming to do this, but their web site is all hype, no shipping products.[2] (Pro tip: Calling yourself "The leader in 3D sensing" when you haven't shipped makes you look fake.)

Automotive is now using mostly 77 GHz radars. That's almost good enough if you have scanning in both elevation and azimuth. Even at 77GHz, you can see bicycles and people. Really good 3D radar plus vision might be good enough for serious automatic driving. Existing low-end 2D narrow vertical angle radar just keeps you from rear-ending the car in front.

[1] http://www.obzerv.com/webfolder_download/bb0e026522747b003d9... [2] http://www.quanergy.com/

[Hydraulix989]

I am very thankful that you wrote this.

[gfodor]

You're making a lot of claims here about the current state of the art of self driving tech. Presumably google has the best right now, so you must be talking about them. Do you have insider information or are you just guessing? Because it seems to me that the google self driving cars are capable of doing much more than working on "the immaculate freeways of California."

edit: Also to be clear the driver assist features seen on cars like the model s are not what most people are referring to when they talk about self driving cars here.

[Hydraulix989]

"Also to be clear the driver assist features seen on cars like the model s are not what most people are referring to when they talk about self driving cars here."

Just saw your edit/addition, so I'm going to add more as well:

When did I even conflate them -- even if the state of the art is actually "driver assist" as you've described?

Google’s self-driving cars aren’t “fully autonomous”. Without human drivers, they’d lose control approximately every 1,500 miles, according to Google’s own reports.

Can you guess how many driverless miles their "self-driving cars" have driven?

Hint: It's zero.

If you're curious, here is an example of a much more grounded academic talk (read: not a TED talk) from 2015 by a renowned (in the autonomous vehicle research community) Boston (not California) researcher who worked with Thrun that corroborates the issues I've mentioned earlier:

https://www.youtube.com/watch?v=x5CZmlaMNCs

[Hydraulix989]

"Do you have insider information or are you just guessing?"

That's a loaded question.

I don't feel comfortable sharing my credentials here (even though they would very much lend credibility to what I am saying), but I can also assure you that I am not divulging any insider information.

However, I claim that my argument still stands without the support of "insider information" -- given that self-driving vehicle research's domain is primarily academia; all of Sebastian Thrun's work (THE founder of the Google self-driving car project and Google X itself), for example, is more or less out in the wild (even if behind walled garden academic journals).

"Presumably google has the best right now, so you must be talking about them."

That's quite an assumption.

[seanmcdirmid]

California roads and freeways are actually not that great, at least compared to other western countries. But regardless, we will see self driving cars adopted first where they are needed the most, cities like Beijing and Shanghai that have lots of cars, no where to put new roads, and a need to optimize the roads they have. The killer app is not road rich America, but the rest of the world that is dying in traffic.

[Hydraulix989]

Which western countries?

Also have you ever been to the east coast of the US? The roads here are much worse, and we have potholes.

[seanmcdirmid]

Switzerland, France, Germany, ....

[ocdtrekkie]

No, they don't work and aren't cost effective. But they sure are marketed well.

[Hydraulix989]

If you noticed, one thing I didn't dispute at all was the "cost effective" part -- LADAR sensors are becoming cheaper and cheaper every year. I remember back in 2008, they were prohibitively expensive (around the same price as a Model S is now).

Now, you can get ones anywhere from $200-$5000. That's why I am sticking with my 5 year prediction for an auto pilot Jetta.