Excellent point. It’s not a bell curve.
So much in life is Pareto. 
So about 15 below the speed of traffic. That sounds safe. NOT!
It goes flow of traffic and doesn’t block anyone.
I just noticed a couple of things on my last drive. When a motorcycle comes up and passes you, it scooches over to give them more room. On city streets at a stopped light. It leaves space for people to go left in front of you into a parking lot and doesn’t pull up to the car in front
I’m in San Francisco right now and just got out of my first Waymo ride. I would judge it as significantly smoother than the human Lyft driver we had on the way to dinner.
I finally got the chance to test FSD with a stopped school bus and it worked well. The bus was facing me and stopped with the lights flashing and the little stop sign out. The car stopped and waited and then moved again when the bus turned off the warnings.
It also correctly navigates the few previous trouble spots.
Amazon’s Zoox is launching its robotaxi service in San Francisco, offering free rides through parts of the city as it accelerates its attempt to challenge Waymo’s early lead in the race to transport passengers in self-driving vehicles.
The expansion, announced on Tuesday, will be confined to a few major San Francisco neighborhoods and limited to people who signed up on a waiting list to ride in Zoox’s gondola-shaped robotaxis, which have no steering wheel. The San Francisco launch comes less than three months after the Amazon-owned robotaxi company launched its first ride-hailing service along the Las Vegas strip.
We had a lot of rain here (Phoenix) the last couple days, and there are reports of Waymo cars getting “stuck”. They probably can’t see the lines on the road. And I understand - I can’t see them either in the rain! We have terrible lane identifiers that, when it rains, become indistinguishable from the other marks like asphalt repair lines and, even worse, older lane lines that have been sandblasted off.
So, this thread was started in October, 2017. It’s not quite a decade old. Do we still think “decades”, or will a decade and change be enough?
Considering that Waymo (the company that seems most successful to me so far) still has problems with anything but clear weather, I’m still thinking decades.
My nephew used to work at Tesla, in the “writing self-driving code” group. I asked him if the software would work in a blizzard, and he replied that that was playing in hard mode, and they weren’t trying for that, yet.
After around 20k miles of driving FSD, I think we are a ways away from complete and full autonomy and as you know I am a major fanboy. I think things like full autonomy to the point that you can safely text (like a Waymo) on well travelled highways and city streets is pretty close. Rural roads and bad weather will be a while. Yeah, I am being vague because I truly have no idea. We are significantly better than I ever imagined it would be if you asked me five years ago and there has been a huge improvement in the two years that I have owned the vehicle.
And this is the $64 jillion question for all of us. Does progress get easier as our skill with sorta-AI gets better, or do we hit a brick wall where progress gets harder and harder and slower and slower.
Either position can be defended.
[Serious answer] I’ve given this reply before. Tesla’s is based on vision, so if you can’t see, it can’t see. It does remarkably well in poor visibility conditions, but it will turn off eventually. That is very close to the point when no human should be driving either. The worst part of using it when there is snow on the road (blizzard or clear) is that it can’t figure out the lanes.
I had one of my few encounters with another Tesla that was definitely using FSD this evening. I knew it was on FSD because earlier today I’d used FSD on that exact same stretch of road and intersections, and the other car had the same stupid turn signal behavior that mine did.
In the left lane and approaching an underpass with the intention of making a left turn, the car will repeatedly briefly signal to go into the right lane. I don’t know if that is something peculiar to this intersection, or that people on US-287 go 15 under in the left lane, so different aspects of the model are fighting between “pass” and “turn left soon”.
After the left turn, the other car, and mine earlier in the day, put on their right turn signal to continue on the same road that happens to make a curve to the right.
Well, obviously there are blizzard conditions where no one should be driving. But there are lesser blizzard conditions where I can drive. That’s what i intended to ask him about.
It’s been a while since I drove in a snow storm, because I live near Denver, which gets less snow than Hawaii.
As long as the roads are clear FSD does a pretty good job when it is snowing. We often get a 40F temperature drop and then snow, so the ground is warm enough to melt the snow, even though the air temperature is well below freezing. I’ll use FSD in those conditions, because the roads are no worse than from rain, and it’s much less snow than “blizzard”. FSD will put up warnings that conditions are bad, and to pay extra attention, but it keeps working.
My biggest concern with FSD is road conditions. The lane lines can be visible through both slush and ice. The traction and stability controls are pretty aggressive (and good) in the Tesla, but sliding straight into the back of someone because FSD should have slowed down 200 feet early is the likely failure mode.
Wet snow will stick and block the cameras, which can be a problem. The front cameras are heated and swept by the wipers, but not the side cameras. When Teslas used to use the radar, the bumper would get covered in snow and block the radar, which would break FSD.
No snow here and it very rarely rains very hard. The issue where I am asked to take over is very bright and direct sun in the front camera on the top of the windshield. Previous to the latest update you had to immediately take over. Now if the condition ends in a second or two, it takes back over without the driver having to reset.
I do periodically post updates about my prediction, it’d be interesting to gather them up and see if I’m at all consistent or if I’m trying to rewrite history. The big challenge of the last 8 years is trying to define what I (or we the dope, or we the dope + the engineers, or we the dope + the engineers + the hypemen, collectively) meant by “self driving cars” in the first place.
What I think I meant, and I still hold to this being a distant reality, is that we could pop on down to our local car dealer and buy a car that would drive us home while we read the owner’s manual in the back seat, and then that would be our car that did all the things our current cars can do, going to all the same places, but without us having to lift a finger to get them there other than telling them where to go. You know, sci-fi stuff.
I didn’t spend much time thinking about all of these intermediate steps we’re going through. And so we’re left marveling at how far we’ve come and acknowledging how far still we have to go.
8 years ago, the basic technology (massive neural networks being trained on real world driving data) that underpins our most hopeful efforts basically didn’t exist and was not seen as the path forward. That’s pretty damning of any predictions anyone made at the time, IMHO. But now that’s a real thing, and it’s given us self-driving, geo-fenced taxis that aren’t profitable and still need human intervention. Pretty amazing! But not really talked about as being the state in 2025. It’s also given us cars that can generally drive themselves 95% of the time but still need humans to remain alert. Again, nobody really talked about this as being some kind of a goal in 2025. But it’s still pretty amazing.
And cars themselves have changed in the last 8 years. They’re more expensive, people are hanging on to them longer and using them less. Will self-driving cars that we go buy even be thing ever? Or will you just buy a subscription to a car service that can take you wherever you want to go in their self-driving cars? Will there be some as-yet undiscovered business model that makes even more sense than that? Will the infrastructure to train self-driving cars need to be so unbelievably expensive that it will never be a thing for regular people?
All things I wonder about now that didn’t occur to me in 2017.
It was interesting to skim the first few hundred pages of this thread and see how optimistic people were. I’ll go back to my first post…
…and I think that while “at least 30 years” leaves a lot of room at the back, I think “not in a hundred years” might be on the table for me now.
Making a separate post for this: Waymo continues its expansion, getting permissions in a bunch of different places. Apologies for the breathless ecstasy of this article (was it AI-written?), it was the one that captured the most info on current and new locations incl. outside California: