The human accident rate is about one per 500K miles, so if they were able to get in that range, then yes, they would have succeeded; drivers would be able to stop paying attention to the road without putting themselves and others in danger.
But the current FSD beta's intervention rate is more like one per 10 miles, judging from some quick googling. I see no particular reason to assume that incremental improvement can take us from 10 to 500K.
This is from from November 2021, but I'm still highlighting it because it is just terrifying (Correct time, though the video later on also exhibits inabilities of the system): https://youtu.be/9wRRClg_aM8?t=113
But are you using confirmation bias to find a cognitive bias that fits here.
But, In all seriousness we don't have access to the data across all 60k FSD users to know what the intervention rate is and how it has been changing over time.
We do have previous statements that as they get better they are moving to harder situations. Start with empty roads, and once you can do them well start finding harder and harder situations. When you start you avoid construction zones, once you are doing well you start looking for them.
Which could be a sign some drivers are simply overly cautious. Suppose 1/10 of disconnects prevented a crash, reducing the risk of crashing to 0 only reduces the number of disconnects by 10%.
To actually reduce that number you would need to make drivers feel more confident in the vehicle which is a useful metric, but only indirectly relating to safety.
What is the appropriate point of comparison though? All human drivers? Sober human drivers? Sober cautious human drivers? Sober cautious human drivers with driver assistance technology (e.g. auto-braking and blind spot warning, or potentially even more sophisticated LiDAR tech)?
I don't think this question is even meaningfully-defined. There is no "the" point of comparison. The relevant point of comparison is whatever ride it's displacing.
The rideshare explosion has already had a measurable effect on drunk-driving deaths; to the extent that a theoretical lower-cost AV will make rideshare even more accessible, then its effect on drunk-driving reduction absolutely makes non-sober drivers a relevant comparison.
For an average young person who'd get in the car with one of their friends, or drives a bit recklessly themselves[1], an AV at sober-human-driver level would be valuable.
For a guy who needs his kids driven around, a "sober cautious human driver" level of safety may feel right.
For questions like "what should the regulatory bar for launch be", all human drivers seems like an easy answer.
[1] I'm probably guilty to a degree here, on the rare occasions I drive
Is it reasonable to assume AV will be lower-cost than rideshare? The key thing that makes Uber more affordable than a taxi is that vehicle purchase/maintenance/depreciation/liability are all externalised.
In a full-self-driving situation you no longer have to pay your driver, but you do have to pay for all of the above. With the inevitably higher standards of maintenance required for AV fleet vehicles I can't really imagine it being cheaper than it currently is.
Sure the sensor/cv/vision tech will get cheaper, but machines still wear down.
> Is it reasonable to assume AV will be lower-cost than rideshare?
That's what the industry is betting on. I think it's reasonable in the steady-state: labor costs are expensive as hell.
> vehicle purchase/maintenance/depreciation/liability are all externalised.
These aren't 100% externalized with Uber, as they show up in the labor cost. They're only externalized with Uber to the extent that drivers do the math wrong on the costs they're paying[1]. Most of the analyses I've seen of this choose every possible pessimistic assumption, and still end up with net wages that are very high. They're of course low relative to "a living wage", which is what the analyses are focusing on, but that's precisely the point of what we're talking about: even the floor of labor costs is very high, when you're looking at expenses.
[1] Completely tangentially, but also note that this ignores the extent to which people derive value from being able to convert assets around. It's hard to imagine for us SWEs making 1% salaries and sitting on mountains of wealth, but liquidity is a constant and pressing concern for a large portion of the country. See also: payday lenders, where there's a stark difference between the opinions of those who've actually studied the economics of the industry and the midwit affluent John-Oliver-watcher.
> The human accident rate is about one per 500K miles, so if they were able to get in that range, then yes, they would have succeeded; drivers would be able to stop paying attention to the road without putting themselves and others in danger.
Unfortunately, I expect that automation will be held to a higher standard than human drivers, rather than the same standard. When an accident happens, people want to know who to blame, and an unimpaired human driver gets somewhat more latitude for a genuine accident, while a piece of software is always going to be perceived to be at fault (which it may well be, even in a situation where a human wouldn't be considered to be). And conversely, people (somewhat validly) want to have more control: every driver thinks they're above average, and the software won't be as good as their accident rate, and if something happened at least they were in control when it happened.
I don't necessarily even think those are incorrect perspectives; we should hold software to a high standard, and not accept "just" being as good as human drivers when it could be much better. But at the same time, when software does become more reliable than human drivers, we should start switching over to make people safer, even while we continue making it better.
(Personally, I wish we had enough widespread coordination to just build underground automated-vehicles-only roads.)
The real averages of FSD intervention are unknown since some 2,000 Tesla employees also have NDA'd Beta access, and it would surely differ between rural, suburban, and urban roads.
The truth of it is that it’s just not possible (with currently existing technology/ML architectures) to create a truly autonomous taxi without HD maps. Everyone in the robotaxi industry knows this - even Tesla builds HD maps, they just don’t call them that.
My knowledge only comes from Karpathy's talks about this (which are great, worth watching if you haven't seen them).
I found his and Tesla's arguments convincing for the general case. That doesn't mean that the narrow cases aren't super cool or valuable (I signed up for this Cruise thing in SF).
I just think that if the software is unable to make decisions based on visual data alone without up to date high resolution maps it'll never achieve true FSD in the general case (not geo locked). You'll end up trapped in a local max otherwise because there are just too many conditions in the real world that vary (and the world is too large to economically map fast enough for that approach). You have to solve the vision problem.
I don't know enough to comment on the approach differences beyond that, but my understanding was that Tesla did not rely on the same stuff that Waymo and Cruise require (largely Lidar and these high resolution maps).
> I just think that if the software is unable to make decisions based on visual data alone without up to date high resolution maps it'll never achieve true FSD in the general case (not geo locked). You'll end up trapped in a local max otherwise because there are just too many conditions in the real world that vary.
My contention is that there’s no way to actually solve for the general case with currently existing technology. The amount of novelty in the real world is too great for any system to account for it without disambiguating via HD maps or remote support.
>You have to solve the vision problem.
This isn’t a vision problem specifically - even if you had LIDAR and high resolution imaging radar and 8 A100s on every Tesla, “true generalized self driving” wouldn’t be achievable without HD maps with our current understanding of Machine Learning.
>My understanding was that Tesla did not rely on the same stuff that Waymo and Cruise require.
Tesla maps individual traffic light elements, stop signs, and lane markings, but will attempt to drive even if the area isn’t mapped.
Disparities in FSD performance in different areas is largely attributable to some areas being better mapped than others - the mapping data has a huge effect on its performance. There are key elements of the driving task (including recognizing and reacting to every single type of sign other than a stop sign) that FSD can’t do and relies entirely on maps for.
Novelty isn’t nearly as big of a problem as you might think. One of Wamo’s famous videos was someone on an electric scooter chasing a duck in the middle of the street. That’s very odd behavior, but the car followed the rather simple option of just not hitting them and going forward when possible.
Cars really don’t need to identify what something is just it’s location and movement which is a vastly easier problem. A trash can rolling down the street can be treated just like an oil drum doing the same thing etc.
> Cars really don’t need to identify what something is just it’s location and movement which is a vastly easier problem. A trash can rolling down the street can be treated just like an oil drum doing the same thing etc.
You’d think that, until you encounter something like a turn restriction sign with a bizarre conditional restriction that it’s never seen before. At which point the car needs to OCR the text, parse the semantic meaning, and apply to the scene.
Right by my house I have a four lane (on one side) intersection with a traffic signal. Each of the lanes goes straight ahead. However, each lane has its own traffic light, and when the traffic light rotation is in that direction, it alternates the two left most straight lanes red while the right most are green, and then switches (because very shortly after the intersection there is a quick lane reduction to two lanes).
I can't imagine how AI would _correctly_ see four straight arrowed lights in front of it in the intersection, some of which are red, some are green. Humans of course recognize that they correlate to the lanes, but this is a more esoteric case for AI to assimilate.
And now we’re already making concessions about the car’s abilities.
There are 10 MPH speed limit signs on Market Street in SF that specify in incredibly small text “when behind trolleys”. Assuming we take your approach, the car will just always go down market at 10 MPH.
Imagine if it’s a negative turn restriction - IE, it’s permitting turns except for during certain hours and conditions. Now the car is treating it as always permitted and turning into traffic. An edge case, but something it’s going to encounter in the real world.
And now your moving the goalposts. We are talking extreme edge cases in some random small town not common signs in a major city. They can always get updates on what some random sign in some random location means as long as their safe and don’t block traffic that’s all that’s needed.
Also, negative restrictions can again default to full restrictions. Permitting a car to say park in a snow lane doesn’t require a car to park in the snow lane.
I don’t think I’m moving the goalposts - we were discussing whether autonomous driving (which I take to mean L4-L5 driving without the need for a human in the loop) is possible without geofences or HD maps. “Edge cases in some random small town” are exactly the sort of thing you need to worry about without a geofence.
Not to mention these sorts of edge cases are way more common in large cities than small towns - one of the examples I gave was down a central avenue in San Francisco.
>They can always get updates on what some random sign in some random location means as long as their safe and don’t block traffic that’s all that’s needed.
What if it truly fails to parse the sign accurately and does something illegal or dangerous? What does sending an update out look like? Does a human take a look at a crop of the sign and review it? Why not just map it in that case?
It’s not a question of parsing a known sign, even extremely complex rules can be encoded. Further that process can take place from a photo of the sign uploaded by the car to then be encoded by the rules. The general case is stopping and having a remote driver slowly tell the car what to do.
An unknown sign in a place without cellphone reception is about the only case where it really need to just figure it out on it’s own rather than simply avoid causing an accident.
> What if it truly fails to parse a sign accurately and does something illegal or dangerous?
Not much, people regularly disobey traffic signs especially ones with complex instructions. Don’t hit stuff or jump in front of another car is generally enough.
> Further that process can take place from a photo of the sign uploaded by the car to then be encoded by the rules. The general case is stopping and having a remote driver slowly tell the car what to do.
So you’re now agreeing that you need some level of remote support to handle edge cases like this?
>An unknown sign in a place without cellphone reception is about the only case where it really need to just figure it out on it’s own rather than simply avoid causing an accident.
Yes, and again, this is the sort of thing you actually need to worry about when trying to come up with generalized self driving solution.
> Not much, people regularly disobey traffic signs especially ones with complex instructions. Don’t hit stuff or jump in front of another car is generally enough.
What if it misinterprets a one way sign at night when there’s no other signal that it’s turning on to a one way lane and it suddenly finds itself traveling opposite the direction of traffic for a long period before encountering another car? You have to consider all of these edge cases when talking about a generalized solution.
Maybe you still disagree with me in sprit, but do you see how when we really look at edge cases how you have to fall back to some level of remote operation or mapping?
> So you’re now agreeing that you need some level of remote support to handle edge cases like this?
As a bootstrap step yes, after that no just regular updates for new traffic rules and such. You can’t make a purely offline self driving system that doesn’t get updated for 30 years because laws change. But presumably a non geofenced self driving car is going to be tested by driving on every road either directly or via someone’s mapping project.
> What if it misinterprets a one way sign at night when there’s no other signal that it’s turning on to a one way lane and it suddenly finds itself traveling opposite the direction of traffic for a long period before encountering another car? You have to consider all of these edge cases when talking about a generalized solution.
You mean in some location without maps? There are a finite number of roads in the world and they don’t change that quickly. If you’re worried that the AI is going to say end up on an ice road that melts, sure that’s the kind of thing that happens once. But the threshold isn’t perfection it’s ~30,000 dead people per year in the US. Beat that and you win.
Or you solve for a subset of highways in a subset of weather conditions. That would be more useful to a lot of people than city cabs which exist today (with human drivers).
Cruise is interesting insofar as they are not simply looking to sell their technology, but they also want to monetize it as a service. Not only will they not need a driver, they will also be able to buy the hardware (the car) at cost. If it's successful, their margins will be much higher than Uber and Lyft by a long shot.
As a taxpayer who pays for roads, and suffers from traffic congestion caused by one-occupant and zero-occupant vehicles, I'm eagerly looking forward to reducing the taxes I pay, by taxing those margins, instead.
Ideally, the taxes could be high enough that driverless taxis will operate at barely above break-even. The financial comfort of me and my neighbours are more important to me than the profit margins of a firm that barely employs anyone in my town.
Unlike a factory or a corporate office (that can threaten to move offshore, eliminating jobs and impoverishing a town), the firm in question is a hostage of local politics - not the other way around.
My cynical take: the government is not going to forgo collecting a tax from you that you are already paying. Instead it will tax you and start collecting per-ride fees from Cruise, etc.
Yes, because in the common case, a taxi (driverless or otherwise) drives empty at least some of the time, to pick someone up, thus creating congestion, compared to a private vehicle, which doesn't drive empty.
The cheaper and more convenient you make zero-person and single-person automobile transportation, the more people will use it, and the more congestion they will create.
The more expensive and less convenient you make it, the more trips will use non-automotive, or public transportation, both of which produce far less congestion.
I actually agree with everything here, but on the other hand the decision of whether and how to actually build the massive amounts of non-car infrastructure we need to have transport be efficient and accessible without private cars of any kind, is in a whole different place. At least in the US, it's pretty clear that in most areas there is very limited political will, even in the grass roots, for things like "build good high-speed trains" and "dig new billion dollar subways" etc. So I think pragmatically speaking things like robotaxis are going to be the "solutions" that we'll actually get.
(And yes, I agree that that's dumb since the same politicians and voters have no problem indefinitely subsidizing and expanding the massively money-losing infrastructure called Roads at taxpayer expense!)
On the other hand, once a sufficient percentage of cars on the road are autonomous, couldn’t they use cooperative navigation algorithms to improve throughput a whole lot?
There are so many inefficiencies with human drivers—chaotic merging, unnecessary lane changes, blocking of passing lanes, and so on. I could imagine that optimizing all those away would make a huge difference overall.
You could also probably increase speed limits. And fewer accidents should cause a significant reduction in traffic jams.
More importantly, "driverless" means no one at the driver seat. What Tesla has is barely even Level 3. Waymo right now is doing rides without anyone in the driverseat, aka Level 4.
I just drove through the Alps, at night, during a snow storm. This is hardly everyday driving, but it's the sort of experience Canadians are no strangers to.
Success is when I trust the autopilot to handle the weather conditions where I live, not just sunny days in California.
Does 50 miles of geofenced and daily mapped streets mean Cruise won self driving?
What if Waymo gets to 20k miles of geofenced roads and monthly mapped?
What if Tesla gets to the point of one intervention/crash every 100k miles? 10M Miles?