Hacker News Reader

Comment by MichaelZuo | Hacker News Reader

It might be a plausible argument for stuff sold 6 years ago when the future was uncertain. But in 2022 I'm not seeing how it's plausible as an excuse. Considering it's still being sold and marketed as 'Full self driving'.

Clearly even level 4 autonomy is far beyond what any near future technology is capable of. If at the very least due to the fact that a very commonly encountered condition, construction work, effectively requires an AGI to navigate.

Which would imply the best achievable self driving is around level 3 autonomy.

View on HN

Pretend you don’t know anything about the different levels. You’re just a random person looking to buy a car. What does “full self driving” sound like to you?

It would sound like it could handle all commonly occurring road conditions.

Such as rain, fog, snow, construction work, tunnels, bridges, downtown intersections, bike lanes, etc...

I wouldn't expect it to handle rare phenomena such as hurricanes, hail storms, a collapsed bridge, etc...

Tesla FSD can barely handle a balmy 70 degree day, so I think you'd be disappointed.

I might biased because I work on actual self-driving efforts, but nothing at all about Tesla has ever indicated they were serious about solving self driving in any capacity.

The day that Elon proclaimed they already had the sensor suite decided was the day that they openly admitted this was fraud.

My opinion doesn’t matter, but your last point resonates with me because this struck me as totally bizarre too.

I don’t work in sensing let alone using AI with sensing data, but I’m around it a lot and my second-hand knowledge kind of set off some red flags.

In remote sensing for landscape architecture, hydrography, and other similar fields (making sense of static things where no one is at risk of dying immediately, and humans review the data), the care and knowledge that goes into choosing sensor suites and processing the data properly is huge. Yet Tesla comes out saying “we’ve got some mediocre cameras and we’re gonna run with it!”

Seriously? What? I get that it’s a lot of cameras but making sense of 3d space isn’t trivial at all. It’s so bizarre.

Again, I know nothing, they know more than I do. The whole thing just doesn’t sit right.

+1 I’m not in the field and my eyebrows raised when I heard that.

> Yet Tesla comes out saying “we’ve got some mediocre cameras and we’re gonna run with it!”

To be fair, humans managed to solve self-driving with some mediocre cameras, some crappy microphones, and some unreliable force-feedback and chemical sensors, without (importantly) adding any new sensors to the hardware between start-of-project and successful full driving. So it's not a priori implausible that "use whatever sensors we have now, and just learn to compensate for any deficiencies in software", could work, since our only preexisting examples of full-self-driving (humans) essentially do that.

It's just kind of stupid to hobble youself that way when you don't have to.

Humans don't have cameras.

Sure they do. They're those two spherical mostly-white-with-colored-circles-on-the-front things embedded in the upper half of their face. (They usually tend to call them "eyes" for some reason.) They're closed-circuit and wired pretty directly into the brain, so it's not like webcams or recording cameras where you can semi-easily get the data off somewhere else, and obviouly they're kind of wet and squishy and not-easily-electronics-compatible like most human components, but in terms of actual functionality, they're cameras.

> wired pretty directly into the brain

Eyes are actually the only visible part of our brains. They aren’t wired pretty directly, they themselves are brain material.

The difference here is that our “cameras” are processed in ways we really don’t understand and at speeds we can’t seem to compete with yet (at least not with the quality of processing we have, so far).

Our sight is also enhanced by our hearing or familiarity with certain driving situations and, in most cases, familiarity with certain places we drive frequently.

If we isolated our eyes from our other senses I have a feeling we might actually be much worse at detecting things around us or making sense of what we’re seeing.

Our senses like road feel can even help us use our eyes in anticipation of dangerous conditions like loose gravel or black ice.

I know machines can do this too, but ours works so well and so quickly; it’s hard to compare with what’s on the market at the moment.

Regardless, I think we agree on the final point. Such a strange way to limit a self driving car.

Unspoken knowledge is also not one of the 5 human senses.

FSD needs the human sense that seeing a stop sign on the freeway (sticking out of the back of a highway maintenance trailer) doesn't mean what a stop sign usually means - but that stopped highway trailer, with a person holding the same sign and looking at you means you should follow it.

Should you run into a pile of gravel to avoid a baby stroller? If a loose garbage can and a baby stroller appear on opposite sides of the road and both block your path, which one do you hit? A soccer ball coming out between parked cars probably has a kid following it...

The people who have died with FSD ran into problems like these. A human would see a transport truck cab, and the wheels on the other side of the road and infer that - even though it blends into the sky - that there's a trailer in between. One of the Tesla accidents was that the side of the trailer appeared the same color and location as the sky, so the Tesla didn't see it and went under the trailer at highway speeds. The driver wasn't paying attention and didn't make it.

Sensing packages will only get you so far, human intuition for new unexpected situations is impossible to get AI to learn - at least at this point.

Agree 100% but I wouldn’t call it intuition, but human understanding. We don’t just see stuff, we understand it. We understand so much context surrounding the things we see. We can predict the weight of an object by observing it’s motion in response to wind, essentially running a physics simulation, to know that that 4x8 sheet of styrofoam blowing in our lane poses no risk. Or that an empty cardboard box tumbling in the wind is actually empty based on it's motion.

We have an incredible level of understanding.

Jeez I even put myself in other drivers’ shoes/heads so as to predict their behavior. Their subtle cues in lane positioning, distraction, etc. they’re all incredibly telling if you can model human psychology.

Not to take away from your core point, but I wouldn't say the human sensory system is mediocre.

Your eyes are sensitive enough to see a candle flame at 2.6 km away. A rod can actually be stimulated by a single photon, but the brain squelches activation that small. The human ear is sensitive enough to hear a watch tick from about 20 feet away in a quiet room. That's pretty darn good.

Why not? Do you expect legal immunity?

Depends if you read the fine print.

The only time I touch the controls on the car is to input the address I want to be driven to.

And you can use it like a taxi service and send a non-driving human as a passenger. Or no passenger and just send it to an address to retrieve or deliver a package.

It's probably reasonable to expect that there are some asterisks like paved roads only and can't drive in literally all weather conditions. But, yeah, no human supervision needed.

> some asterisks like paved roads only

Some asterisks like non-paved roads will cause poor and potentially dangerous performance, technically.

If I tell it to drive into a volcano, and getting it to do so is significantly harder than five minutes of repeating various variations on "Yes, I know it's a volcano and will likely destroy the car and everything in it; do so anyway.", it's defective. (Preferably it also should not be significantly easier, of course, at least not to the point of being able to do so accidentally.)

My ultimate dream is to be able to more or less avoid flying. Hop in my car at 11pm, punch in address, pull out the lay-flat seat, go to sleep, wake-up 7-8 hours away. If I make it a day trip, I can do the entire thing in reverse never having had to pay for a hotel.

Living on the coast, I would still have to resort to flying, but there are many places I would enjoy that are reachable in an eight hour drive.

You can do this in practically every EU country, not to mention Asia (I don't know about other parts of the globe) via trains. I don't get why it has to be a car. It's just that the US has zero infrastructure for both inter city as well as intra city public transport outside of like 4 cities.

I am an American, so lacking the ability to reverse 80 years of public policy decisions, I want the sleeper-car-car.

For one, pull up a satellite view and look at the US anywhere 150 miles away from the west coast.

And we have buses that go between lots of cities(greyhounds). They just suck compared to driving yourself or flying, which is why you'll pretty much only find poor and desperate people on them.

The US has one of the most extensive rail networks in the world. What we do differently than Europe is use the rail network almost exclusively for freight. If we followed Europe's lead and did a lot more freight via long haul truck, we could free up a bunch of infrastructure for passengers, but would the end result be a net win? Or would we actually increase emissions by doing that?

There is absolutely no reason to make that tradeoff. If a rail line is hitting maximum capacity, then that means there's plenty of money going around to expand it. Adding more passenger trains does not require decreasing the amount of freight trains. That's not the reason we're bad at doing passenger rail.

Freight rail has significantly different requirements than passenger. People want to move quickly, with frequent daily trips (eg 24 trips per day, moving at 90mph). Freight does not care about train speed so long as it is reliable (eg 1 trip per day moving at 50mph). Logistics networks can plan around reliability.

Well, it's a bit more PITA to get your ass to station with all the baggage, then also possibly changeover to another line. But yeah, to the airplane it's an altenative.

This is already happening with regularity, at least in California.

Engineering is about solving part of the problem first, then iterating. If you look at what they've accomplished only in the last year with FSD, I find it highly unlikely that 0-disengagement drives don't become the norm by the end of 2023.

Currently, 90+% of drives have 0 safety-critical disengagements, and 50+% have no human-discomfort-related disengagements.

Are you referring to Tesla's FSD effort, or Waymo?

I hear excuses when I point this out, but lane keeping and adaptive cruise on the Model 3/Y are inferior to the competition now. AP was a selling point when the Model 3 first came out, it has become a bit of a joke now. If you buy a Model 3 today, it comes with an asterisk. If you want cruise control, beware, adaptive cruise has notable bugs and there is no option to turn it off and go old-school.

> If you look at what they've accomplished only in the last year with FSD, I find it highly unlikely that 0-disengagement drives don't become the norm by the end of 2023.

> Currently, 90+% of drives have 0 safety-critical disengagements, and 50+% have no human-discomfort-related disengagements.

No one is going to put their pre-teen child in a car for a 5 hour trip when there is a 10% chance that the car will crash.

And yet you think that such a thing will be the norm in 12 months or less?

> Engineering is about solving part of the problem first, then iterating.

And fraud is about selling solutions to parts of the problems you haven’t yet solved. While there are engineering criticisms of Tesla FSD, that’s not the issue here, so presenting definitions of engineering is a non-sequitur.

I don't necessarily disagree, but I offer a couple of counterarguments: A. Most people who bought this knew what it was, that it was unfinished, and they were funding the research (which is a public good to some extent) B. There should be some degree of burden on the customer to research what they are buying, especially with a claim as extraordinary as "Full Self-Driving"

Counterargument to myself: Elon sold it with a claim that it would skyrocket the value of the vehicles.

I never did understand how that passed the sniff test. Unleashing a fleet of self driving robotaxis wouldn't skryrocket the value of anything, it would be the fastest race to the bottom in memory.

That particular aspect passes the sniff test for me. The race would only go so fast with a million cars produced per year, only some of which would be participating.

Maybe we should bring back patent medicines again? New, improved Radithor! If you're unhappy when your jaw falls off, it just means you should have read the fine print.

Maybe the burden should be on the seller to call it “unfinished and unreliable attempt at self driving” instead?

The idea that we don’t let marketers make misleading product claims is very old. As a society our point of view is that no, actually the burden of proof should not be on the customer to figure out if a tube labeled “toothpaste” is actually wet gypsum or something.

The research burden might lie with the customer for "don't take your hands off the wheel just because if you ignore the instructions it sounds like something an 'autopilot' would let you do"

But I don't think it's true for believing the car company CEO when he repeatedly said that the car came equipped with all the necessary hardware for full self driving, and if you purchased the $15000 package the full self driving software would be with you by the end of the year, with most of the remaining obstacles being regulatory...

It's a bit like consumers have a responsibility to read the label on medical devices to use them properly, but they don't have a responsibility to advance their medical understanding to the point they can tell if Elizabeth Holmes is telling the truth about her product actually doing what she said it was doing.

Sounds like fraud.

It means a car that can drive in a reasonably broad range set if expected traffic conditions and has no failure mode that is unsafe at any point.

It’s that second part that is the important one. If the car can’t drive in heavy snow or roadless areas or whatever that’s fine, as long as it’s able to recognize and avoid those situations in a graceful way, like refusing to engage with them and parking in a safe area, suggesting the trip be cancelled, that kind of thing.

The current approach of it works, but when it doesn’t it will say “just kidding take over for me now” while going 70mph towards the side of a truck is definitely not that.

> What does “full self driving” sound like to you?

Well, that I would be fully self driving the car? ^^

No it means that you have to drive the car with your full self.

I feel our government should regulate this kind of speech more aggressively, to boost the safety of the public.

Announcing a capability, and importantly MUSK acting as a widely-recognized public figure like it's ready - shaming every other fledgling competitor, before it's ready, is amounting to GASLIGHTING.

A speculate this is what we should expect from the company-owner's notable, controversial personality type.

Immediately upon receiving "Full Self Driving Beta", in the 201x's, some Tesla drivers CRASHED because they turned ON the lane-keep-assist and then left the car to do the driving .. straight into the back of a stopped truck, or a split open lane-divider, and died.

To reiterate, this mis-leadership has resulted in death.

I suspect that CEO personality-type is a salient factor in whether a company should be allowed to operate. Should we stop this .. should we prevent actual Companies operating? Not just a group of hobbyists but tax-paying, licensed, employing companies. It is OUR government. We can stop this predictable travesty of gaslighting -> good-faith belief -> bad outcomes.

No. I stopped at “the government should…”

That never goes the way you think it will.

Advertising claims are in fact regulated. There just isn't a lot of enforcement but, at normal companies, legal teams are very conservative about making claims in advertising and marketing materials. If I claim our product is better than Brand X, you'd better believe legal is going to want to see the evidence and probably qualify the statement.

Indeed false advertising is a thing and it's not legal(depending on where you live ? it is in most of the world that has teslas at least). The government reaching further than going for false advertising is where the slope gp mentioned would come in I assume, and on that I fully agree.

It is fine (and correct) to be cautious about any situation where the government gets involved and starts setting limits.

But to discard an argument entirely is to bury your head in the sand.

Regulations are applied to many industries and limit the kinds of claims that can be made about products, and those regulations are important to public safety.

Such regulations most often apply to products that involve life or death consequences.

A company marketing vitamin supplements can’t claim that the supplements are proven to cure cancer, and that’s a good thing.

This is not to say that the government can’t or doesn’t get things wrong. But I’m curious how you feel about limitations placed on financial, food, and pharmaceutical companies.

At a minimum, there’s a fraud angle. You shouldn’t be able to claim to sell something that does A when that thing cannot in fact do A.

And when A happens to be critical to not dying, the implications of that claim just raise the stakes even higher.

The involvement of government is at most the courts to redress the kind of issues you're talking about, which is narrow to the particular situation involved, guided by precedent, but not fixed by it. Fraud, injury, death are all covered here.

Heavy-handed regulation I can't back though, in part because it is (increasingly these days) as broad as possible, as vague as possible, ensnaring far more then the initial case merited.

Why would you brag about not reading what someone wrote?

I did read it, actually, but there wasn't much of merit there. The usual spend more, make more rules, enforced by someone with a gun somewhere such that bad thing X can't happen again. Then of course, it does.

Not bragging, just bored.

[citation needed]

Fair.

Although honestly, it should also be on the regulation proposer to say cite why such a rule is necessary, and more importantly, under which conditions, if it doesn't work out, such a rule might be withdrawn.

I mean: We've had Medicaid, but still have problems in elderly healthcare. We have the Patriot Act, but still have terrorists. We have a Department of Education, but still have low (possibly lower) quality schools. We have a CDC, that failed in testing and containment of disease. We have an SEC, but SBF's fraud went unabated (although hopefully punished soon). We have an FAA, but still have unworthy aircraft (737 Max looking at you).

Perhaps then next round of rules proposed by folk looking to win their next popularity contest will work next time, I just doubt it very much. I just don't see much value in these rules - if bad things keep happening, what was the point? Might as well not have them, and make people directly responsible, through a court of actual law, when actual damages have occurred.

I’m not against regulation but I’m not sure it’s needed here either.

In the case of misleading statements and marketing causing death, that seems like something that a lawsuit would handle, not regulation.

In the case of false marketing misleading investors, we would see investor lawsuits for that if the stock price suffered as a result. Right now, what damages would you show for musks lies?

Instead of new regulation, a lot of nonsense would have been avoided had the SEC not given him a free pass after the last round of investor fraud. It was entirely predictable that an egomaniacal child would not change behaviour after a slap on the wrist and a warning.

Regulation often happens after companies have mislead people, it’s more a chicken egg scenario, do we let disasters happen to legislate against or do we have the foreknowledge to see the disaster happening and stop it?

Musk is a charlatan, but why do people act like it is common to buy $40k+ cars like picking up some gum in the checkout aisle? What the features name sounds like is not the only factor here. People don’t buys cars with that lack of thought. What the salespeople say, what the marketing material says, what the manual says, what the CEO says in public, what the feature feels like during a test drive should all matter more. This isn’t a defense of Tesla. I think it still leaves an argument for fraud, but we really need a better argument than the same “the name sounds misleading when you first hear it” argument that has been repeated for years now. If that was all it took, we all would have sued cellphone companies for their “unlimited plans”.

This isn't a "9 out of 10 dentists recommend using our product" situation, and cellphone companies have been sued for failing to deliver true unlimited plans.

It is being marketed as "Full Self Driving", and the CEO is publicly saying it is "in beta", "coming next year", and "just waiting for regulatory approval". In reality it needs constant driver supervision, is still primitive enough that it is directly linked to multiple fatalities, has been coming "next year" for over 6 years, and is clearly not even remotely finished. Customers have paid thousands of dollars for a feature they are not able to use, half a decade down the line. They may never be able to use it as advertised at all. The product as delivered is substantially different from anything the company says.

It's like saying your car can do 0 to 60 in less than ten seconds and then delivering a car without an engine - while putting a statement in the small writing that the car currently only reaches this in freefall from a mountain top, but pinky promise that they will install the engine in a few years.

Your comment helps prove my point because you didn’t stop at just the name. It takes more than “the name is a misleading” to be considered fraud. Tesla is guilty of more than just giving the feature a bad name. If you are arguing fraud, mention those other things. Because people don’t spend $40k+ based off a name of a feature.

I can still go to Verizon’s website today and find “Unlimited” plans so the name clearly wasn’t the problem in whatever lawsuits they might have been involved in.

And to reiterate, my point is not that Tesla is innocent. My point is that the most common argument used to point to their guilt is a stupid argument, especially considering there are better arguments available to argue the same underlying point.

I'm sorry, it's a scam to sell X "coming next year" and six years later, not to have it.

> I can still go to Verizon’s website today and find “Unlimited” plans

That other scams exist doesn't negate that.

> I'm sorry, it's a scam to sell X "coming next year" and six years later, not to have it.

Where did I state otherwise?

Once again, all these comments are reinforcing my point which is that there are plenty of more concrete complaints to make about this than “Autopilot is a misleading name”. I don’t know how many times I need to say I’m not defending Tesla. I am criticizing a bad argument. But I guess I should just accept my downvotes. I should know better than to try to have a nuanced conversation about Tesla and Musk.

You said: we really need a better argument than the same “the name sounds misleading when you first hear it” argument that has been repeated for years now

We do have that. We have a very thorough argument, and it has been repeated endlessly. So that's why you got downvoted.

Context matters. Reread the comment to which that was a reply. It focused specifically and exclusively on the name so I called out that focus.

You didn't focus your comment well enough on that aspect.

Sorry, I guess I shouldn’t have assumed that people would read my reply in the context of the comment I replied to.

People often elaborate moderately beyond the direct point that they started off in reply to. You are making an incorrect characterization of the problem. Your reply was unclear in context.

In your opinion, exactly how much time should people making purchases put in to investigate the extent to which they're being lied to? Because in my view, the burden should first fall on companies to be clear and honest.

The name itself is not a lie. You can call it misleading, but it isn’t a lie by itself. When the CEO is obviously lying to the public and you focus on the name instead, it weakens the case for fraud in the court of public opinion.

> If that was all it took, we all would have sued cellphone companies for their “unlimited plans”.

Personally, i think it’s bullshit that that’s not cracked down on too.

Will not crash into stationary objects, bicyclists, pedestrians, or the cars in front of me. Will not randomly halt the car either.

Pretending I knew nothing about the levels “full self driving” sounds like it means I could safely nap in the back seat of the car while en route .

To me, “full self driving” has always meant the car has a built-in cab driver so I can work while the car drives to my parents for dinner 6 hours away, then sleep while the car drives me home. I’m not sure how the phrase “full self-driving” could mean anything other than that.

Edit: When I pre-ordered a Model 3 Musk’s promise at that time was literally “press a button and your car will work as a taxi to earn money when you’re not using it”.

> I’m not sure how the phrase “full self-driving” could mean anything other than that.

There is some room to argue about the conditions under which the car can drive itself. If the car couldn't handle a high-grade, winding mountain road with an inch of snow on the ground and yet more coming down, I don't think too many people would be fussed is a "full self-driving" car said "uh, no" (especially as a good fraction of drivers couldn't handle that either).

To my end, if the car can't handle any condition that doesn't make me go "I think I should stay home," that should disqualify it.

Humans often drive in conditions were it's unsafe -- icy roads, dense fog, heavy rain, strong winds. We take the risk, and most times it's okay, except for the occasional pile-up on the highway.

But I don't think a "full self-driving" car refusing to drive in those conditions would be a deal breaker. I'd prefer it if autonomous vehicles were more conservative and just refused to drive in unsafe conditions.

Well then, call it, self driving under perfect conditions but not as a star modifier

Right. But you then end in a sport where driver barely does any driving 99% of the time but the 1% they do need to drive, they are out of any practice for normal driving, let alone in hard conditions.

We already had problems back when people started coming back from mandatory WFH and accident rate spiked, this would be essentially similar or worse, a bunch of drivers that might do 10k miles a year but drive 500 of them and have extremely rusty normal driving skill, let alone driving in bad conditions.

It would be essentially like letting your kid drive your car only in blizzard and pouring rain...

I'm not sure that experience with driving in normal conditions actually helps when driving in dense fog or on icy roads. People are overconfident and drive just like they normally would, which is why the pile-ups happen every time we have worse than usual road conditions.

A lot like a git merge, then :)

There is not room to argue. Unless we call it “Full self-driving*”.

Full describes the set of use cases the car can drive in. All of them, not some with conditions and exceptions.

Not many people can fully drive a car then. Which makes sense.

I think excluding off road / hazardous mountain roads etc. might be OK for the definition. And still cover many 9s of actual trips.

Driving into a parked ambulance is definitely a no no though.

Sure... regardless calling a Tesla self driving is blasphemous at best, though personally I find is criminally negligent.

I know what I want. I want to tell my car to go the the shop and get serviced, come back, and take me home after work. After a flight, have it pick up my inlaws at the baggage claim exit. etc...

It sounds a lot like something Musk promised in 2019: "Next year for sure, we will have over a million robotaxis on the road," "The fleet wakes up with an over-the-air update. That's all it takes."

The most generous interpretation is that Musk is delusional and out of touch with his company and his technology. But I think it is more accurately described as bullshit ("speech intended to persuade without regard for truth") or a lie (making knowingly false statements).

I think the interesting question is whether that legally ends up being fraud in the context of a civil suit. Whatever was going on in Musk's head (deluded, reckless, or lying) it seems like Tesla as an entity should have known that the cars were not actually self driving and that customers could reasonably think they were getting self driving.

> The most generous interpretation is that Musk is delusional and out of touch with his company and his technology.

I'm not sure I'd describe that as a generous interpretation, but there is other evidence that he's just wildly optimistic/over-confident. E.g., he still pours significant investment/effort into his project to colonize Mars, even after describing falcon wing doors as "engineering hubris"

It's not a particularly positive interpretation, but because it's the least bad, it's the most generous.

Basically, steering wheel is optional, it wasn’t going to happen, it’s been clear for years

Getting drunk every day.

Great, so we're back to "unlimited data" doesn't really mean "unlimited" but, just a large amount, but not really a large amount, because...

Words have meaning. Full self driving, to anyone who isn't a nerd who likes to pick shit apart, means the freaking car will drive all by itself as if a human were driving it, maybe even better.

I think 2016 it was already widely known by the industry it was going to be very difficult compared to what Tesla was saying, especially without the handicaps they imposed on themselves. No one with knowledge took Tesla's timeline seriously

> Clearly even level 4 autonomy is far beyond what any near future technology is capable of

I've always been pessimistic about FSD but I think it'd pretty hard to reconcile that quote with actual L4 deployments by Waymo and Cruise

> actual L4 deployments

In some of the cities with the easiest driving conditions in the world.

I wouldn’t call San Francisco easy…

Yes. And that's L4 and it exists

Apart from SF, they have also started testing in LA.

An easy city is still a city.

Only the low end of L4 is being reached right now, but that's massively better than "far beyond what any near future technology is capable of".

They also have remote operators standing by who jump in when the car gets confused tho…

That's the definition of L4, and the reason Waymo isn't L5. L5 is no human in the loop is ever necessary.

It's pretty amazing how much money Musk's lies helped companies like Uber to waste on this stuff.

I don't like how this whole thing has played out, but to be fair, I think fear of missing out over something that would, if successful, be an existential threat to Uber is probably a rational response.

I don't completely disagree. They definitely have to be weighing the likelihood of that outcome, and Musk's statements probably made it seem more likely. That said, there are a lot of existential threats to companies that they should probably not be spending investor money on, such as asteroid defense.

What do you mean? Waymo and Cruise have already deployed level 4 vehicles.

Can you link to the source claiming that?

Waymo offers a taxi service in SF where the cars have no driver.

From https://spectrum.ieee.org/amp/full-autonomy-waymo-driver-265...

> The Waymo Driver operates at Level 4 autonomy, meaning, Waymo says, that “no human driver is needed in our defined operational conditions.” This, Waymo continues, represents “fully autonomous driving technology,” with the Waymo Driver being “fully independent from a human driver.”

https://spectrum.ieee.org/full-autonomy-waymo-driver#:~:text...

Level 4 means full autonomy over all normal driving conditions. Waymo does not handle complex construction sites like a normal human driver would, a fact verifiable by many thousands of people. It does not have level 4 autonomy.

Looks like the article got backlash too because the editor inserted an editor's note in quite large font right in the middle of the article clarifying it.

> Level 4 means full autonomy over all normal driving conditions.

Aren't you describing Level 5?

From what I see online, Level 4 means

"The car can operate without human input or oversight but only under select conditions defined by factors such as road type or geographic area"

Waymo has been operating geo-fenced self-driving cars without drivers present. It may not handle complex situations, but that isn't expected out of L4.

Level 5 means full autonomy over all known driving conditions. Including dirt roads, etc.

Otherwise there would need to be a level 6.

What's the difference between "all normal driving conditions" (your definition of Level 4) and "all known driving conditions" (your definition of L5)?

These are not my definitions, they are the SAE's definitions. As anyone can verify themselves by visiting SAE.org

Even with SAE's definition, Waymo seems to be at L4: https://www.sae.org/blog/sae-j3016-update

It would only be the case if their claim is solely evaluated within certain limited areas. With the rest of the road network being ignored.

Level 4 means full autonomy within certain predefined conditions. In Cruise's case, this might be, "Only in certain parts of San Francisco at night." This qualifies for Level 4.

What's stopping someone from predefining it to mean "Only within a specific parking lot"? And then claiming they achieved Level 4?

Nothing. That is the definition of level 4. Level 5 is no restrictions.

Seems like a silly way of assessing levels of autonomy then. Maybe it doesn't make sense to allow it to be predefined to an arbitrarily small area?

If current Waymo counts as level 4 then we need sub-levels for them to have much meaning. They are a long way from supporting something like autonomous driving geofenced to all major US cities in all weather conditions. That big step up is still far behind true level 5, which can handle poorly maintained roads and all of the other weirdness one can encounter outside of major first world cities.

What about non California weather? I imagine radar might handle fog, but driving rain as in Carolina or serious snow as in Canada?

The rain can be no joke here. I've had to stop driving and pull over for the first time in my life after moving to NC. The rain was too heavy to see more than 10 feet or so.

> The rain was too heavy to see more than 10 feet or so.

Probably a hurricane. Did you see any cattle flying?

Not hurricane season at the time. There's a clearly defined season for that weather. This is just a normal thunderstorm.

Rain + city lights can make it hard to see even road markins in front. Technically self-driving car would theoretically have opportunity for advantage here (as it can just refer to maps) but practically it's super hard problem to solve.

Rain fade affects GPS does it not? Is the car relying on inertial navigation for a period? This rain is not something a camera and iffy GPS signal can keep you on the road. The intensity and amount of rain would definitely not let a camera see.

Quick add: this was at 12PM (noon) on a countryish road. City lights had nothing to do here.

I've wondered the same thing about FSD in Canada.

Often we Canadians drive on roads where we cant see the lines due to snow/ice. We sort of form our own "lanes" based on where we think the lines SHOULD be.

How would a FSD Tesla handle a situation where everything is white?

Even Humans who are use to these conditions still fail: https://ottawa.ctvnews.ca/opp-responds-to-numerous-crashes-o...

> We sort of form our own "lanes" based on where we think the lines SHOULD be.

And that is exactly what FSD does. You can search YouTube, there's plenty of examples if you want to see how it performs in different weather conditions. Snow is challenging but not impossible.

That is not my experience at all. It either says “Full self driving unavailable. Poor weather detected,” or it freaks the fuck out and tries to kill me.

FSD can’t adequately handle rain, let alone snow where the road lines are not visible.

Based on many hours of footage, rain doesn't seem to be a big problem unless it's very heavy, but the experience can vary I suppose. Snow for sure is still a problem, but there is a path to improving it. The current state of perception shows that it can be solved by deep learning (more training data for these conditions), in the sense that in many situations it can guess where the lines and lanes are based on other context, like tire tracks, positioning of other cars, width of the road and lane meta data, snow protrusions etc. There's also the problem of when to treat snow as an obstruction to go around and avoid vs when to just go over it. There will always be more uncertainty in these conditions though, just like there is for us humans, we are often unsure where the lanes are exactly, and one of the challenges if having the system deal with this uncertainty better. And then control has clearly not been worked on for snow and ice yet.

Forget rain and snow, they can't even handle non-California roads yet.

I had the misfortune of taking a rental on narrow mountain roads and it actively attempted to kill us twice. While moving over to pass an oncoming car, the lane keeping feature decided we were too close to the shoulder and started steering us back into the oncoming car.

After pulling over, regaining my composure for a few minutes, and figuring out how to disable it; somehow it activated again on a different day and it happened again.

> somehow it activated again on a different day and it happened again.

I dont have a tesla but my car has things like "lane assist" and most of these safety features can be turned off, but the default state is "ON" and this is reset once the car is turned off/on again.

Oddly enough, what it calls "advanced cruse control" defaults to OFF on my car and must be turned on any time you want to use it?

Yeah, I think it was something like that. Once I began checking for it, it seemed inconsistent - possibly related to how long the car was off?

The joys of modern technology - every time I get in my car, I have to spend 15 seconds checking the state of the "Car may decide to murder you setting".

Granted it did fine on wider, more developed roads once we got closer to larger towns, and it would probably do fine in my day-to-day life; but that whole experience has left me really shaken on any self driving tech. The roads it fell apart on weren't even that bad - well paved and maintained, just narrow and without any sort of center stripe.

"lane assist" can defiantly be annoying. While mine doesn't try to kill me by driving off a cliff... It does seem to prefer that i drive dangerously close to those cement "jersey barrier" walls they use in construction sites??

The lane narrows due to construction and it prefers to maintain the lane even thought it is too narrow... I don't get why it cant detect the barrier and realize that driving over the line is safer?

This is my big issue with the whole "FSD" - roads are not uniform and driving conditions are not always idea.

Let's say FSD is fine for 80% of the time, but what happens if drivers depend on it and it is now in the 20%?

I think giving drivers a false sense of security is very harmful.

Psychologists have shown time and again that as car safety improves, driver aggression only increases to remove the benefits.

ABS - You can now drive faster knowing you wont skid...

"lane assist" has shown that people will depend on it and allow their cars to drive right off the road and into the ditch (youtube has a lot of these). FSB will only be worse.

That is because NCAP and rest only counts features that are on after starting car into that rating.

Similarly MPG rating also requires that, which is why stuff like auto stop/start and the safety features don't remember settings after restart.

> Considering it's still being sold and marketed as 'Full self driving'.

The problem is that this isn’t what they do. They are extremely careful in their wording.

Go back through the archive.org backups of tesla.com/autopilot and watch how their wording changes.

Here’s what they say today:

> Tesla cars come standard with advanced hardware capable of providing Autopilot features, and full self-driving capabilities—through software updates designed to improve functionality over time.

Despite how you or any reasonable person reads it, Tesla can easily make the case that the cars (as sold today) do not have full-self driving available right now. The hardware in the cars does have the capability of allowing full self-driving at some nebulous point in the future (something that’s going to be very difficult to disprove).

The other thing the do is use Autopilot (capital A) as a branded feature that doesn’t mean anything like what autopilot (lowercase a) means to a reasonable person.

Reply on news.ycombinator.com