Self driving cars are still decades away

In My Humble Opinion
2728

As echoreply said, FSD actually does quite well in glare/rain conditions. The only times I’ve seen it struggle were when I myself struggled. One should expect it to perform pretty well actually, since the side/rear cameras still have excellent visibility when the car is driving directly into the sun. Humans of course do very poorly driving into the sun.

Lol. Barely half the US population lives in a place that gets snow. Maybe you live somewhere where white shit falls from the sky on a regular basis, but a significant chunk do not. And of the rest, a large portion are in places where snow is rare enough that you can just not drive in it. Or drive in it, but not use FSD, assuming your car has a steering wheel.

Not that I expect it to be a problem even in the medium-term. A system that relies on Lidar will struggle but vision will do at least as well as a human.

2729

I’ve had the direct glare be so bad (very briefly) where it alarms and has me take over. It will also alarm sometimes when I spray wiper fluid and it’s sunny.

2730

Yeah, it’s not really ready for unsupervised driving without operating in a carefully chosen area, and poor weather makes the performance of pretty much every self-driving producer bad enough that they don’t even try.

43 US states have at least one city that averages more than a foot of snow per year. Fact is, most states get some snow (and all of them get rain). As far as I know, no one with a self-driving car allows them to run in rain or snow.

And really, if it can’t handle what I normally handle comfortably, in what way is it really self-driving?

2731

Yeah, lots of states have a mountain or two where it snows. Irrelevant to most driving.

It doesn’t rain much here in CA but FSD works fine in moderate rain. Last year we had an absurdly heavy rainstorm (flooding in many intersections, etc.) and I decided to take FSD out at night. It did get confused in an already confusing curved intersection, ending up in a different lane than it started, but aside from that it did well. Probably would have done fine if there were other cars around since it could have followed them.

Normal, moderate rain without road flooding? Never had a problem, even at night.

Why would anyone besides you care about what you can handle as a driver? If a self-driving system can handle 99% of population-wide driving needs, then it is not “pretty useless.”

2732

It’s basically the same dumb argument that EV haters use. Oh, an EV can’t haul my horse trailer 800 miles between fill-ups like my F-150 with extended tanks can. USELESS!

2733

But it currently can’t, and it’s not even close.

I’m not one of the fools that thinks the current Lightning is useless. If I needed or wanted a ginormous pickup and didn’t need to haul a trailer long distances, I’d be looking at it first. But I don’t need one, and if I get an electric truck, it will probably be the smallest one I can haul a sheet of plywood in. Because that’s the thing with the most challenging dimensions I expect to carry besides my 2x15 bass guitar cab, and the cab fits in my hatchback.

But if I was going to stick my wife in a driverless cab of some sort, it better be able to handle fucking rain. Hell, it’d better be able to pass the Finnish driving test. I’m no super driver, but I can get around in bad weather. Any self driving product that actually is self driving should be able to do that. Anything else is just pretending.

2734

FSD as a technology handles rain. The current Robotaxi launch, which no one will tell you is not extremely limited, does not. There’s no reason they can’t just flip a switch once they’re confident enough. It may not even be turned off because they have any reason to believe it’ll fail; just because they want to minimize the number of variables in their first release and work from there.

Again, if FSD and/or Robotaxis work in a large fraction of driving situations, then it will not be “pretty useless.” I have zero interest or knowledge of what a Finnish driving test looks like.

2735

This is what I’m pushing back against, though. It entirely depends on how you define “poor”. It’s easy to move that line to a point where self driving won’t work at all, and declare that you’re correct. FSD seems to behave fine in light, moderate, and heavy rain. At some point on the light to moderate end it warns of being “degraded.”

Eventually the rain can be so heavy, or perhaps a heavy rain at night, and self driving will disengage. In my experience in rain and snow, those are points when humans should not be driving, either.

Like I said earlier, I can understand why the driverless taxi services stop at the “degraded” point, but in my experience that point is far from when FSD is non-functional.

The biggest issue FSD has in snow is when the road is covered so it can’t see the lines. In my experience, which is mostly on the older pre-full-AI systems it was erratic. It might follow the tire tracks, give up, pick some other lane that doesn’t exist. Definitely too dangerous to play with when there is any other traffic.

2736

But what I’m pushing back against is, I believe it’ll probably be generally useless in an actual self-driving mode far earlier than the current mode you’re using, which is still a “fancy autopilot” mode. Technically, it’s still not really self-driving even when it’s used in bright sunshine at noon outside it’s geofenced area that specifically excludes “difficult” intersections. It still requires a butt in the seat in far more than 99% of US cities.

2737

While I’m not @scabpicker, I’m absolutely not an EV hater. If the next MX-5 generation is not electric, I’m leaning heavily towards the Taycan as my “fun” vehicle, or hoping that the 718 comes in electric by then. I’m also pushing my wife (who drives trucks due to her horses to get either the new RamCharger or the F150 lightning when she replaces her current F150). I’m also not a self-driving hater. I probably want self-driving more than you do. I just want it to be unsupervised.

We’re right back to you trying to explain to me why right field is the position I want to play, when I want to either play shortstop or do something else completely.

2738

This morning I dropped DesertRoomie off at the local Costco for her shift. It was 11am so the store was in full madhouse mode for a Saturday with autos rampant and people pushing shopping carts everywhere. I went down the parking lane closest to the front door and stopped behind a Waymo that was about ten feet from the intersection with the lane that ran immediately in front of the door.

While DR was getting out and collecting her gear the Waymo moved up and wanted to make a right turn towards the store’s door. It swung a little wide to avoid some cart-pushers along the curb and intruded into the left half of the lane. A car coming the other way stopped and the two of them were about fifteen apart staring at each other – at least the human was; I don’t know about the Waymo – while more cart-pushers blithely moved between them.

Luckily I wanted to go left and there was plenty of room to get around the Waymo and escape. The two of them were still sitting still last I saw them. In this situation I would put the majority of the blame on the human. When I’m near a Waymo in a parking lot I give it plenty of room in a confusing situation.

2739

Neither of you seem to have understood the argument, which nothing to do with EVs. Let me try again:
“I deem technology X useless because it does not work in a situation that only applies to a subset of the population, which, coincidentally, includes me. I will also dismiss the possibility of narrowing the set of circumstances in which it does not work, despite this pattern holding for all of the technology’s existence. I am deeply ignorant of the current state of the technology and am also unable to extrapolate to the future.”

2740

I only spoke to EVs because you brought it up. You’re not understanding my argument, so I’ll make it clear…

“Full Self Driving*” provides no value to me at the moment because it isn’t actually full self driving if you have to effectively put an asterisk after it. If I have to pay attention to the road, I might as well just drive myself.

2741

I’m not @Dr.Strangelove.

My latest car has a pretty good ADAS, but it’s far short of Tesla’s current “FSD”, much less a truly AI-powered fully self-driving car where there is no human driver at all, just zero or more human passengers being whisked obliviously by their e-chauffer to their destinations.

Prior to that car I thought as you did.

Having gotten used to ADAS, I now find it a great convenience on both boulevards and highways. Yes, I’m still paying attention. But it’s far less attention at a far slower rate of repetition. It’s a very different and much easier experience letting the ADAS drive as it does. My expectation is Tesla’s “FSD” would be even more relaxing.

I’ll suggest that you might find the same after you tried it too. So not “useless” but rather “less useful than perfection would be, but still darn handy.”

2742

People who love to drive will find it annoying because it won’t do what they want it to do and they think their way is better, and it may well be. They would probably appreciate it in stop and go traffic though.

For people like me who hate driving, it’s a godsend. It has improved my life considerably. It’s what sold me on getting the car.

2743

The answer to that is the same as to the EV haters, smartphone haters, etc.:
It’s fine. Don’t use the technology. It isn’t for you, and may never be for you. But it will get incrementally better over time, as we’ve already seen. And at each step it becomes useful to a larger subset of the population, and eventually become ubiquitous or even indispensable.

2744

Long post 1 of 2

So, I’m with you on this. I like driving, yes, but it’s also something I’ve been training my brain to do for 30 years now. It’s one of the few things in my life that I have the requisite 10,000 hours to declare expert status. I instruct high performance driving, I teach a lot of my friends’ teenagers how to drive, etc.

I’ve had access to regular old cruise control since I got my license, and I’ve always hated it. I just can’t seem to rewire my brain into “supervisory” mode. I use it, even now, sometimes, but it’s rare. If I drive for 8 hours, I’m driving for 8 hours. I’ve driven cars with adaptive cruise control and lane keeping, and I find them to be even worse. I’ve trained my brain to anticipate the unexpected on the roadways; I’ve NOT trained my brain to anticipate the edge cases where I will suddenly need to grab the wheel. So I’m much, much more anxious when I’m in that mode.

I’ve not “driven” with FSD, but I’ve ridden shotgun while a friend drove, and I found it maddening. Yes, he was able to pay much less attention, and that freaked me out! All the edge cases, the situations where FSD just isn’t good enough yet, that’s what my brain is working out when I’m driving. What sudden movements will other drivers make, are there kids between those cars, will the lane markings suddenly get confusing, etc. The actual tasks of driving the car are 2nd nature, so if all I had to do for an 8 hour trip was keep the car in the center of the line and maintain the speed limit, I wouldn’t be exhausted at the end. But that’s… not at all what I’m doing. The reason I’m exhausted after 8 hours behind the wheel is because my brain is in constant alert mode, trying to figure out what could possibly go wrong.

I take driving seriously, and so when I see my friend basically “check out” even though we’re both relying on him to take over in the case of one of those 1-in-a-million edge cases that he’s not even looking out for… ack! And hearing FSD fans say “Yeah, I can basically check out on my commute.” Ack! If your driving style meant you were spending the bulk of your mental facilities just trying to drive the car… ack! You’ve been doing it wrong!

So I agree with you, FSD is not for me, not because I hate technology but because the things that might happen behind the wheel, that I’ve been training for my whole life for, won’t go away. And I’ll have to learn a new skill, which is “anticipate what this robot that nobody knows how it works will react in various situations so I know how to take over.” It sounds stressful, and awful, and I think this is where you’re coming from.

Now, once my car is effectively an uber, where I can get in and play on my phone the whole time… great! That’s a different experience. Supervised self driving is for the birds, though.

2745

Long post 2 of 2

I wanted to address Dr Strangelove’s notion that EV haters, or self-driving haters, or Tesla haters, are all anti-futurists who hate progress.

I think he’s coming out this from a place of altruism, and that’s commendable. Cars are dangerous and while I personally take driving seriously, most people don’t. There’s a bunch of yahoos out there, and driving is supremely dangerous, the most dangerous thing any of us do on a regular basis, but we’ve all adjusted our internal risk assessments to downplay the risk so we can just, you know, get on with our lives. So people will keep dying by the 10s of thousands, and the sooner we can let technology take over, the better. Lives will be saved. It’s good for humans. I get it.

I think this viewpoint sidesteps a bunch of issues that the world is currently facing, though. I’ve ranted about all of these on this board in the past, possibly in this very thread, but I’ll rehash them again.

  1. Cars are terrible. I say this as a car lover, as a car racer, as a lifelong enthusiast of the freedom cars provide. They are undeniably terrible for the environment, and all modern societies should be speed-running away from them. It’s been said that “EVs are not being built to save the planet, they’re being built to save the car,” and that’s very true. Cars were getting a bunch of bad press for their emissions, and so EVs let us pretend for a little bit longer that car culture is sustainable. It’s not. We need more population density, better public transportation, different ways of structuring society and our living spaces so that we’re not so car dependent.

Among the left, I don’t think this is controversial, although the right thinks that this is an anti-freedom sentiment and that it’s our God-given right to live as inefficiently as possible. Woohoo.

  1. [Most] EVs are terrible. They’re big, they’re too heavy, and they use way too many resources to build. They might be a net positive over heavy gas cars, but they’re the wrong direction for the automobile at this time. I think the Prius Prime might be the most sustainable new car, with it’s limited use of lithium and small size. If we can build 5 Prius Primes for the battery materials that go into a single Model Y, we should. And we probably shouldn’t be building 5 Priuses for every Model Y because cars are still bad (see point 1).

I hate that Tesla “solved” the range anxiety problem, and that that was the blueprint for every other automaker who wanted EVs that could sell on merit instead of “compliance cars” that would only sell at a huge loss.

EVs are not good for the planet, because they’re cars. All the hate that Hummer H3s got for being heavy and ostentatious should also apply to 5 or 6 thousand pound long range EVs. They’re disgusting examples of excess. Drive a Leaf and maybe we can talk. Maybe.

We’ve allowed big corporations to continue to make money while extracting a vast amount of natural resources for a select tiny percentage of global citizens. There is no utopian future where 10 billion people get around in 250 mile range Lithium power EVs. It’s a dead-end, and I don’t accept that it’s “better than the alternative.” I’m a zealot in that regard. I see people driving around in EVs and I just see people sticking their heads in the sand re: their own environment impact.

  1. None of the companies in the current self-driving arms race share Dr. Strangelove’s altruism.

All of these companies are burning through investor cash trying to be first to market so their owners and investors can get rich. That’s bad. It’s the worst of capitalism.

They’re not out there beta testing on public streets to hasten an accident-free utopia and save thousands of lives per year. Maybe that will happen, but if it does, they want to be billionaires because of it. And billionaires are evil. I will not root for any of them, just like I’m not rooting for an AI companies. They’re all problematic.

  1. In the speculative space, I think FSD and Waymo are amazing! Look at everything they’ve accomplished. I think Tesla is ultimately correct that vision only self-driving is the only way to get it done; LIDAR is a dead end, IMHO. It’s better in the short term, but I don’t think there will be a true L5 LIDAR powered self-driving car before there’s a true L5 vision-only self-driving car. I think once we get to the point where self-driving is a thing, it’ll be a thing with whatever equipment humans currently have – 2 eyes on a stalk in the driver’s seat and maybe a couple of ears.

BUT – I also think that 5 minutes after Tesla perfects unsupervised FSD, they’ll be able to put an Optimus robot in the driver’s seat and have it drive the car. That’s how far I think we are from this being an actual working technology – and I don’t think it’s that far. It’ll happen within my lifetime. But I don’t think that chasing specific AIs are going to shortcut us there for a task a general as driving. I think Tesla and Waymo are basically taking shortcuts for their investors benefit and no other reason.

  1. Tesla, specifically, is evil, and I will not root for them. And that’s because Elon Musk is a hateful piece of shit who’s made my life hell by going full MAGA and fucking up the US government, which is, in turn, fucking up the country that I love. Fuck Tesla, and fuck anything that puts dollars in the pockets of that human garbage.
2746

Well, if you can’t do it you can’t do it. It’s fine. I’m the exact opposite situation–I always crave more high-level supervising vs. low-level fiddling. And really, not just in driving but in most aspects of my life. I wouldn’t be a programmer if I didn’t write a script for every annoying task.

You learn over time what works and what doesn’t. No, you can’t check out and you won’t hear me say that until they achieve L4 on their non-Robotaxi vehicles. But you develop a sense for the things it handles well and the things it doesn’t, and take over if you aren’t comfortable enough. It actually works pretty well in construction zones these days, but I tend to take over anyway. You can spot them miles in advance so it’s not like you have to be on constant alert.

In many says, FSD acts as pure augmentation. There’s never been a human solution to “what if I’m changing lanes on the highway, look over my shoulder to check my blind spot, and right at that instant the guy in front slams his brakes hard?” It’s just a complete hole in the driving experience that I can only reduce, not eliminate. FSD eliminates that. It will brake for the guy in front, and check my blind spot, which I check anyway with high confidence that I won’t rear-end the guy in front.

As an aside, my “supervisor” instincts are at least as engaged when I’m a front-seat passenger as when I’m a driver. And I’ve been in one actual accident and a few near-misses where I could see well in advance the mistakes the driver made to get them in that situation. At least with FSD I can grab the wheel! Thankfully the one accident was just a fender-bender, but still.

2747

Not what I said. It’s about a specific type of argument, which really isn’t even limited to tech, though that’s a common application. Really it can be summed up as “It doesn’t work for me, therefore it’s not useful.” It’s basically a non-sequitur. So what if it doesn’t work for you?

Not that this is not symmetric with the enthusiast’s claim that “This does work for me, therefore it is useful.” Even the most enthusiastic people rarely claim that a certain thing is a good fit for everyone. But things can be useful even if they only apply to a small fraction of the population. And it’s undeniable that something is useful if 50%, 90%, etc. embrace it.

I won’t address the rest of the post, which belongs in a different thread.