Hacker News Reader

Comment by zaroth | Hacker News Reader

The reason Tesla does not fall under this legislation seems pretty simple to me, but IANAL…

The law applies to vehicles with the "capability to drive a vehicle without the active physical control or monitoring by a human operator".

Tesla FSD does not have this capability. If the driver does not consistently demonstrate that they are actively monitoring the vehicle and in physical control, then the “AutoPilot(TM)” feature deactivates.

”An autonomous test vehicle does not include vehicles equipped with one or more systems that provide driver assistance and/or enhance safety benefits but are not capable of, singularly or in combination, performing the dynamic driving task on a sustained basis without the constant control or active monitoring of a natural person.”

View on HN

So, in summary Tesla "fully self driving" does not actually have the capability to "fully self drive." It seems like the hazy marketing around this feature hasn't been a great idea.

Other than the latest, bleeding edge beta ("FSD Beta") FSD is vaporware.

Autopilot on the other hand is a pretty good driver assist (lane keeping, TACC, etc.).

From one of the previous threads on FSD

> Tesla has so many versions of "FSD" or Full Self Driving because they keep shifting the post for what exactly is Full Self Driving.

> What even is FSD compared to SD, where did the need to tack on "Full" come from? And I bet it's because of them calling their lane assist cruise "Autopilot" which implies self driving.

I'd say the marketing has been perfect, the haze confuses both the consumers and regulators believe whatever they want to believe.

This is just a semantic argument about what "capability" and "autonomous" means. The system is obviously capable of autonomously navigating in an engineering sense, because that's what it does.

The system as deployed ("FSD beta") also requires a human driver, and actively monitors for driver attention. So it is not actually operating autonomously.

It seems that the spirit of the law is to disallow unregulated testing of cars that are actually operating autonomously. Cars with a human in the loop don't really seem to be what the regulations were thinking of.

The letter-of-the-law argument would seem to disallow any sort of uncommanded control by the vehicle, including lane assist and 1970's cruise control. And that's not right either.

The article addresses that explicitly.

Basically, the DMV is saying "oi, if you're testing a system that's capable of driving by itself, you have to follow certain rules"!

Then you reply "No, we're still testing it, and it requires human supervision, so it's not capable of driving by itself, so the rules don't apply!"

> "FSD" is aspirationally an automated driving system. The name unequivocally communicates Tesla's goal for development, and the company's "beta" qualifier communicates the stage of that development. Tesla intends for its "full self-driving" to become, well, full self-driving, and its limited beta release is a key step in that process.

> Tesla's "FSD" has the "capability to drive a vehicle without the active physical control or monitoring by a human operator," but it does not yet have the capability to do so safely.

I think the "loophole" here is that - despite the marketing claims - Tesla is NOT testing a system which is designed to run without active monitoring by a human. This is unlike Cruise, Waymo etc. who do fall under this law.

Marketing is admissible in court as evidence of intent.

That would be the case if they didn't also have controls in place to prevent the driver from disengaging while autopilot was active.

are you suggesting any developer of autonomous driving systems can straight up ignore this law as long as they make sure the system requires the presence of a human?

Yes? Tesla isn't the only company making driver assist systems. I have the equivalent functionality in my Genesis right now and no one is making noise about it.

The marketing is defense against any claims that you bypassed a safety, since marketing is saying publicly the safety systems are not needed.

My company has lost in court about similar cases where the court decided that the marketing picture showing someone closer than the warning label says is safe means we don't believe the warning label and so we are liable when a customer gets too close to cutting blades.

> Tesla's "FSD" has the "capability to drive a vehicle without the active physical control or monitoring by a human operator," but it does not yet have the capability to do so safely.

It’s that bit that seems like it would make Tesla’s actions fall even further under the DMV’s purview. Is their argument really going to be that they can’t be held to safety regulations because it’s too unsafe to be regulatable? That’s like saying you’ve stolen too much money to be guilty of robbing the bank.

I suppose it depends on what "capable" means.

If you mean that a Tesla with FSD can be used to drive without human control, then that is definitely true since the tests for driver alertness can be bypassed.

If you mean that a Tesla with FSD has the level of technological sophistication needed to drive autonomously, that is also true: if you bypass the checks for driver alertness, the Tesla can continue driving (albeit not safely). This is well beyond the range of most "driver assistance" technologies.

If you mean that a Tesla can drive autonomously without bypassing any checks for driver alertness, then no, it cannot, but only assuming these checks never fail and are always sufficient to prevent it from being used without active monitoring.

Tesla's case seems weak here, in that it relies on an extremely narrow interpretation of the law, according to which a system that can drive autonomously but tries to prevent you from doing so is not considered truly autonomous.

But IANAL, so perhaps Tesla has a better case than I'd assume.

> Tesla's case seems weak here, in that it relies on an extremely narrow interpretation of the law,

To be clear: there is no case, nor has Tesla tried to make one. The linked article is an opinion essay by an academic in the field. The CA DMV hasn't said anything.

For myself: I'd argue that we have ~23k FSD beta vehicles on US roads now that have been driving for 3-4 months and that if we're really concerned about the safety of the system (vs. getting into semantic arguments about the statute language of some random law) we should just go get the data and see. By my math that's going to be something in the tens of millions of miles now. That should be well above the noise floor.

How are "we" going to go get the data? Tesla isn't exactly open about these things. (no, marketing numbers aren't enough)

The subject being discussed is literally the regulatory authority of the California Department of Motor Vehicles. You don't think they track accidents statistic?

I mean, if someone was coming along from the DMV and saying "here's a report showing a 95% confidence of a higher than average accident rate among FSD beta vehicles" then I'd be first in line to admit that, yes, we should probably halt deployment while we figure out what's going wrong.

But that's not what's happening here at all. The people with the data are silent, which tells me something that confirms my priors as a FSD beta tester: this system as deployed is safe. It is.

Tesla isn't complying with the law, they're keeping the data to themselves. How would the CA DMV be getting accurate statistics?

What form are random cops going to fill out for "Tesla fucked it again and this was FSD Beta version whatever" ? Or will cops exactly and reliably fill out accident reports in exactly the same way with impossible precision, and then who is going to notice? How many accidents will they miss? How many _near_ accidents will they miss? How many takeovers by the driver will they have no idea about?

How does the CDC get covid infection statistics? How does the NTSB do aviation safety studies? How does OSHA report on occupation risks? Of course the government can do this. In fact in this very area there was big news over the summer that the NHTSA had opened an investigation into Tesla autopilot after finding a collection of just 13 accidents over six years where a vehicle on AP hit an emergency vehicle! If they can pick out needles from the haystack like that, they can surely run an analysis over FSD beta.

Look, if you're arguing that we need a unified system in place for reporting accidents and correctly tagging them to autonomous vehicle software stacks, then I'd tend to agree. But that's not an argument about Tesla safety, nor even with Tesla's compliance with existing regulation.

All I'm saying is that (1) my personal experience with this system as deployed is that it's quite safe, frankly safer than driving personally, (2) there is a ton of data out there being generated by all these cars, and (3) let's go find it instead of pursuing brand flamewars like this.

> "[...] NHTSA had opened an investigation into Tesla autopilot after finding a collection of just 13 accidents over six years"

If it takes them 6 years, they're not getting enough actionable data quickly enough.

> How does the CDC get covid infection statistics? How does the NTSB do aviation safety studies? How does OSHA report on occupation risks? Of course the government can do this.

By requiring that key industries and people report the data, then sometimes with inspectors. It's not magic, and it's not going to go well if you're expecting the data to just appear for them.

> If it takes them 6 years, they're not getting enough actionable data quickly enough.

The phrase "over 6 years" means that the sample set was six years long (since AP was first deployed), not that the data is six years old! This doesn't seem like a serious argument.

My point, to repeat, was that regulators absolutely look at this data, to the extent that a cluster just 13 accidents was enough to trigger an investigation. The data is absolutely available! And to the extent that it's not, let's talk about specific reforms you'd like to see to improve the quality.

But none of that is an argument about whether or not Tesla FSD beta is currently operating safely. Again, it's quite clear that it is.

> But none of that is an argument about whether or not Tesla FSD beta is currently operating safely. Again, it's quite clear that it is.

eyeroll I'm glad you've conceded the argument on my behalf. Any other decisions you'd like to make for me?

> And to the extent that it's not, let's talk about specific reforms you'd like to see to improve the quality.

I'd start with Tesla complying with the law and providing data about how many accidents, how many disengagements, how many miles, etc. and not having to depend on what gets noticed over 6 years.

CHP has alleged cases of drunk or sleepy drivers who disabled the dead man’s switch and were somehow in a full self-driving mode. Because people who presumably aren’t mechanics or automative engineers can enable an unsupervised self-driving mode, it seems like it might fall under this law.

CHP has even dealt with one guy who has been arrested multiple times for letting his Tesla drive itself down the interstate while he (the only occupant) was sitting in the back seat. https://www.mercurynews.com/2021/05/11/backseat-driver-chp-a...

Yes perhaps to further analogy at the extreme end - it could be seen analogous to if a caffeinated-weed-vodka energy drink company claimed its caffeinated-weed-vodka enhanced your driving ability and claimed to have internal metrics suggesting it worked, marketed the heck out of their "super alert" drink and wink-wink encouraged people to test it on the road but first to go to e-sign some click through waivers on their website.

just reading this now... this person wasn't even using FSD, but was using autopilot...

Put an ankle weight on the steering wheel.

This is how it fails:

Turn the system on, keep a finger touching the steering wheel, then close your eyes. Boom.

In fact no. It actively monitors driver attention with a camera. If you close your eyes or look down at your phone for more than a few seconds, the car will shriek at you, lock you out of autopilot for the drive[1], and if you don't take manual control will bring the vehicle to a stop. Recent versions have in fact been tuned to be too sensitive and people have been complaining about false positives.

I do wish people would spend some time actually investigating how this system works before pontificating about it. It's not done. But it's far, far closer to done than you seem to want to believe.

[1] And after three such events kick you out of the beta program entirely.

I can follow the argument on how FSD does not fall under the law because of its requirement for driver attention. I do wonder how much water that holds, but that is for another thread.

What I'm curious about is what you mean by "done". You mean an autonomous Tesla that can drive without human supervision? Is Tesla close to that in your view? On what type of roads?

Is this a recent change? There are many videos online of Tesla's "driving" with no-one in the driver seat. One tester in the bay area even had his car seized[1]

[1] https://www.nbcbayarea.com/news/local/backseat-tesla-driver-...

It's specifically enabled for those enrolled in the full self driving beta (and only while it's enabled). As far as I know it's not yet made its way to those who just use autopilot which is what was used with that remarkably dangerous backseat stupidity.

Edit: Zaroth's comment makes it sound like it's now enabled for normal autopilot use as well https://news.ycombinator.com/item?id=29598520

I am in the FSD Beta queue but I have not gotten access to it yet!

I wonder if the more aggressive monitoring and alerts turn on when you enter the queue for FSD Beta as a way to monitor your alertness before granting entry into the Beta?!

I guess Tesla feels you can "actively monitor" from the back seat? Or did those people defeat some safety?

Yes they weighted the seats, slipped out from a seatbelt, and weighted the steering wheel.

Newer software released in the last couple months also adds some kind of gaze tracking using the interior camera - because the car now alerts constantly if I am looking at my phone and will disengage auto-pilot and put me in the penalty box if I keep doing it. I don’t know how that tech would react if it didn’t see a human in the drivers seat at all, but I imagine ultimately that will be another safety measure. Tesla has a lot of incentive to cut down on these types of pranks.

For those that don’t know, disengage keeps autopilot running but with an extremely annoying alarm and massive red warning on the display, and as soon as you take over the penalty box means you can’t re-engage “AutoPilot(TM)” until the next drive.

There is another way to get it to stop alarming. Don't look at your phone while you're driving.

Please don't look at your phone while actively driving.

Thank you, I know that. I’m just explaining how I know the technology works, which is interesting and relevant in this context, not looking for driving tips.

But you bring up a great point! Very interestingly, the current version of the software will alert just as much and in exactly the same way when it detects the driver looking at their phone, whether the car is at a complete stop or driving freely down the highway.

> whether the car is at a complete stop or driving freely down the highway.

I’ve seen the police on off ramps looking for people texting or whatever while they are stopped to give them tickets.

Not that it matters in the context of TFA but in certain (most?) jurisdictions it isn’t legal at all.

My comment was more condescending then I intended, I appreciate the additional context.

One guy had this foot on the wheel from the back seat. Not sure how/if they defeat the face/eye detection.

that was old tech, and it's not part of the BETA topic is about

Reply on news.ycombinator.com