[Laarlf]

Do you work with software? Because you don't seem to be scared one bit. I don't see this becoming a thing anytime soon. Not just because of how horrible software is and how little it improves. In the last 20 years we barely had any improvements in what cars assist us with. The Mercedes Benz W220 from 1998 had active cruise control. And today modern active cruise control systems barely work better. Some cars can sometimes change lanes for you. But even that is far from good. I look at how bad lane keep assists (which also have been around for over 10 years) detect roads, especially country roads and i don't feel like we will get anything remotely good from "self driving" cars in the next... MANY years.

[jjav]

Agreed, I don't believe there will be any such autonomous coordinated cars in many decades. It is a seriously difficult problem for the general case. It's effectively AI and true AI has been just around the corner for many decades but never arrives.

For the happy path of interstates in good weather, sure, that's not too hard.

[jawns]

I think the closer we get to the point where all vehicles are autonomous, the easier it will get. The interim is the hard part. Right now autonomous vehicles have to be able to operate alongside human operators whose ability to communicate their intent is limited to brake lights and turn signals, and there is basically no way to cooperate. Now imagine a world where it's all autonomous vehicles who are constantly communicating intent and can cooperate in marvelous ways, like with orderly zipper merging or intersection management without the need for traffic signals. There will still be a need for computer vision to detect pedestrians, obstacles, and wildlife, but it will be simpler than having to deal with other human operators.

[Laarlf]

My issue is that even today "autonomous" vehicles are far from capable of doing what they are supposed to do. The technology is decades old and barely anything happens.

And no, i like driving. I would hate a future where driving would be illegal.

[jjav]

> I think the closer we get to the point where all vehicles are autonomous, the easier it will get.

The last 5% of anything is the hardest, and on this particular problem it's even harder.

[jawns]

That's often true when you're talking about a single project with a clearly delineated end state, such as, "Build a responsive website for my coffee shop that works in all major desktop and mobile browsers."

But it's generally not true when you're talking about advances in technology and market dynamics where there are multiple inflection points. If I told you that the products Apple is planning to build over the next 3 years are going to be harder to build than anything it's built in the previous 45, simply because the last 5% of anything is the hardest, you'd probably say, "That's not what I mean!"

I think it's pretty obvious that building autonomous vehicles that can understand human driving behavior without direct communication is harder than building autonomous vehicles that only have to worry about other autonomous vehicles, with which they can communicate in data-rich ways. The fact that the former needs to happen before the latter because of market dynamics doesn't change the fact that it's a harder technological problem.