If those high targets are the only things remaining then I'd say the race is pretty close to finishing. I'd be more pessimistic if features on the todo list still included 'identifying stop signs', 'recognizing people', 'not killing people', and 'driving more smoothly than a teenager'.

You do realize there are countries that have a significant chance of snow for 4+ months per year, right?

And that there are many places with stop signs that are almost completely obscured by foliage, or simply by the angle of roads on the crossroads.

Not everything is California. And not every city has a grid.

When autonomous cars can survive in a place like Rome or Paris[1], or even downtown Manhattan. Then we can talk about fully replacing humans.

[1] Rome and Paris are notorious for people having special city cars because they get banged up so much in the course of normal life

Here's an example of traffic in a modern European city: https://www.youtube.com/watch?v=HfJ-VokP3EQ

And another: https://www.youtube.com/watch?v=lay8aZlsbB0

How long until autonomous cars can do that? Hell, I'd shit my pants trying to do that.

PS: in the real world there are also many roads that flat out don't have lanes drawn on them.

Google's self driving cars only drive on roads that have been comprehensively mapped by google at great expense. You couldn't just take it to a new country and expect it to work.

So yeah, there's a very long way to go until self-driving cars are ready to replace human drivers.

I believe it'll happen quicker than you think. Even if we can't move beyond having to create and maintain detailed road maps, once the concept has been proven to work, you can expect many companies simultaneously jumping to map their local regions. That's one thing capitalism is really good at - distributed, simultaneous deployment of solutions that have money-making potential.

I hope you're right. I could see that happening in warmer climates. But what about when landmarks and points of reference are covered in snow? GPS is not accurate enough to be used as a sole reference, you need a visual/LIDAR reference to back it up.

The system will have to work in 100% of circumstances before laws will be changed to allow for driverless cars. I could see that happening in 20 years, but I really doubt it'll be on the streets in 5 years like Elon Musk and others have suggested.

Construction alone is going to make it extremely difficult, unless there is a system such that the cars know where construction is taking place and actively avoid those routes, but that is not always known.

There are just too many edge cases. I believe, that we'll get some kind of hybrid solution: computer drives on the highway during daytime otherwise drivers takes over.

In Switzerland we have those wonderful pass roads. On one side you have the mountain, on the other side there's nothing. Never in my life would I let a computer drive down that road with my family in it:

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

Sometimes roads are so narrow, that drivers naturally honk while going round corners. If there's a honk back you gotta slow down. Now add some snow fall and fog and drive down in the night. Good luck!

> There are just too many edge cases. I believe, that we'll get some kind of hybrid solution: computer drives on the highway during daytime otherwise drivers takes over.

Most likely, it'll start out as driving assist systems (like an advanced cruise control, similar to what tesla recently deployed). The computer does the driving, but the car needs a driver in the drivers seat, picking his nose and waiting for the very rare instance when intervention is needed.

To be honest, I think this would be a step back. Risk compensation[0] will ensure humans will put way too much faith in the system, and not be alert enough to intervene when needed. Plus, driving is boring enough when you're actually driving the car. Just sitting still on a 4+ hour journey, not able to actually do anything else (because you need to keep your eyes on the road) sounds like a special kind of hell to me.

Self-driving cars would be a huge leap forward. More driving assist systems will just make people less careful.

We can already see a lot of manufacturers rebranding their driving assist systems as "self-driving cars" because it's the cool thing to do.

[0] https://en.wikipedia.org/wiki/Risk_compensation

What's more, people are forgetting that normal traffic does not go through wilderness, and roads are not just slabs of concrete in a line. We draw markings on the road. We put signs next to them. It stands to reason that as self-driving cars will be slowly deployed (I don't see them rolling out simultaneously in sunny California and in snowy Switzerland), we'll be updating road infrastructure with signs and other objects designed for AI drivers - ones that could help them determine road conditions better.

Considering how poorly current roads are maintained I don't hold out hope for anything that requires more maintenance. Many roads I drive on are not properly marked for human drivers as it is.

I'm sorry but I think this is a grossly over-optimistic view. If vehicle automation requires any upgrades to road infrastructure I see it being a total non-starter.

do you know if they do well at seeing through heavy rain or running water on the road? Coming down from Snoqualamie resort near Seattle the snow turns to water with about a good inch over the road and the lines are invisible. Traffic suddenly becomes a negotiation from 3 lanes (actual stripes) to 2 (comfortable) and in between, mostly determined by what the trucks decide.

I wonder what the various dreams even are for what to do about snow -- embedding something in the lane marker paint or who knows what.

I rather hope that individual vehicles for everyone just turns out no longer to make sense eventually and goes away.