7 ms·

As someone who works in the autonomous driving space, it is absolutely mind boggling to me how much blind faith Tesla fans put into FSD. When you compare it to

As someone who works in the autonomous driving space, it is absolutely mind boggling to me how much blind faith Tesla fans put into FSD. When you compare it to any true L4 system it’s not even close and very unlikely to get close any time soon.

jvanderbot 3y ago
As a robotics person outside this space, it's not amazing that the shift to L4 as a goal has not been clearly grok'd by the public. After years of rhetoric, most people still believe that cars will be fully autonomous because they just don't understand the limitations. It's also quite amazing to me we, the so called smart ones, ever believed level 5 was happening but those were the good old days, huh.
- faeriechangling 3y ago
  It's hard to impossible for me to know how difficult level 5 driving even is since we don't have any machine that can do it. Trying to achieve it has been a good example of turtles all the way down. At the same time I feel like people are ignoring how well self-driving works RIGHT NOW, the rate of accidents is still less than with human drivers. So I'm still pretty optimistic.
  - paulryanrogers 3y ago
    How can you know if Tesla won't publicly release stats of interventions or accident per 100k miles?
    - bluGill 3y ago
      There is Cruise and Waymo data as well. Though I have not seen anyone independent who has access to that data making any claims about it being safer.
  - jvanderbot 3y ago
    > It's hard to impossible for me to know how difficult level 5 driving even is I have slightly a different opinion. I think you can easily guesstimate how difficult something will be by listing the "new" stuff. But I think you can't easily estimate how easy something will be because you can't list everything, so anything you miss could easily delay. Said another way: We should have known it would take at least a decade, but should never have said it will take at most two decades. (as technologists). And yeah - agree - lvl 3 is pretty good, feature-wise.
- lern_too_spel 3y ago
  L4 is all that's needed for a taxi service, and that's where all the money is. Tesla sold cars with "FSD" to customers living anywhere, so they have to solve L5, and they have to do it with fewer sensors and less compute. Of course it's not going to work. We're all just waiting for the other shoe to drop for the inevitable lawsuits.
ado__dev 3y ago
FSD is a Level 2 system and not autonomous as stated by Tesla in legal filings. So yes there is a huge difference between a real self-driving system and whatever the hell Tesla is selling.
ajross 3y ago
It's a supervised system as deployed. How much "blind faith" are you really seeing vs. what you think you're interpreting. None of us are taking naps in the cars. The system does continue to make mistakes, though virtually always of the "too conservative" variety. But it's also the most fun I've had in a vehicle in my whole life and an absolutely sale-maker for me personally. I simply won't buy a vehicle from another vendor until it can do this. So... I guess I repeat the question: where are you seeing "Tesla fans" using FSD unsupervised? I mean the linked article is almost unique in that it seems to have happened. And why? Well, it's buried but it turns out the driver was on autopilot because he was absolutely hammered: > Von Ohain and Rossiter had been drinking, and an autopsy found that von Ohain died with a blood alcohol level of 0.26 — more than three times the legal limit
- ra7 3y ago
  > So... I guess I repeat the question: where are you seeing "Tesla fans" using FSD unsupervised? One of the biggest Tesla influencers using FSD without his hands on the wheel: https://www.youtube.com/watch?v=fFW4e0Pkz_Y https://www.youtube.com/watch?v=fFW4e0Pkz_Y This is explicitly disallowed by FSD terms and conditions. He does this all the time.
  - ajross 3y ago
    That seems like a bad faith argument. Do you genuinely feel that driving with your hands in your lap constitutes dangerous operation? The car asks for your hands on the wheel because that's the primary sensor it uses to monitor attention. In particular, do you genuinely feel that driving an FSD car with your hands in your lap constitutes the kind of inattention we were actually talking about (driving with a .26 BAC!). I mean, you don't freak out if you see someone using cruise control without their foot hovering over the brake, do you? Objectively and largely inarguably, braking control latency is more important to safety than steering latency.
    - ra7 3y ago
      Yes, I genuinely feel driving with my hands on my lap is dangerous. I own a Tesla, use Autopilot (not FSD, but applies to that as well) and always have my hands on the wheel because I know, at highway speeds, I may not be able to react quickly that one time. This is the entire reason Tesla has hands on the wheel requirement. And the reason for having that requirement, which is crux of the problem, is that the system is incomplete and incapable of safe operation.
bkandel 3y ago
When I was in a Tesla Uber, I found watching the front screen that shows where cars are around you frightening. Cars would frequently move around and change position unexpectedly and in a way that had nothing to do with actual car movements, lane markings would be totally off, etc. If that screen is at all representative of what FSD "thinks" is going on, I'm not at all surprised that there are lots of crashes when relying on the system.
- fisherjeff 3y ago
  YES. Thank you. I rode in a Model Y last year and just could not believe the mistakes it was making. Disappearing and reappearing cars; the semi in front of us was apparently straddling the lane line for a few miles; somehow an early-2000s Dodge Ram was classified as a small sedan – the list goes on, and this was only a ~10-minute ride. I would be absolutely mortified if a product of mine ended up in front of a customer in that state.
  - eschneider 3y ago
    I spent a decade working on commercial computer vision applications that, among other things, had to recognize and track cars. Those are exactly the sort of transient errors you'd expect to see in shipping products and they usually have heuristics to "smooth" over those sorts of problems. That said, would I ever trust my life to a system like that? No.
    - oscb 3y ago
      I'm actually surprised they show it as raw as it looks. Doesn't inspire too much confidence, even though I bet the system must be reclassifying and changing way faster than it renders things on screen.
    - fisherjeff 3y ago
      Don’t get me wrong, my background is also pretty CV-heavy, and I don’t expect perfection by any means. But the display itself serves basically no purpose besides looking cool, and it just fails pretty badly at that. Also yeah, it made me maybe a little more nervous about being on the road with a Tesla than it should’ve…
myko 3y ago
Just sold my Model S and bought a Lyriq and am incredibly surprised at how much better Super Cruise is than FSD. It drives just like a human, incredibly smooth. Of course, it won't work on regular streets - just freeways, but maybe that's for the best, considering how inconsistent FSD is.
themaninthedark 3y ago
As an engineer, it boggles my mind how much people trust software in general. Toyota was put under a microscope after the Prius crashes....do you think anything has changed since? Do you think the software running the cars has gotten any better? VW cheated on emissions, they payed civil and criminal fines and a couple engineers got charged. No one from managment....Since then, it has come out that almost everyone was cheating. Cummins was most recently caught and it is now a civil fine(also being called record breaking in the news, at 1.6B where VW was 1.5B civil but 4.3B total...) We normalize the behavior and decrease the penalties.
- bluGill 3y ago
  I trust humans even less than software. At least software can be debugged over time and is consistent so if we get it working...
  - avgDev 3y ago
    I would imagine debugging something like FSD might be difficult. Not sure what kind of data they store, but no situation might be the same.
    - bluGill 3y ago
      Every try to debug a human? Worse, humans are not copies so making one better doesn't make it easy to upgrade the rest of us.
      - themaninthedark 3y ago
        If the software was broken up in a sane manner and followed the "do one thing and do it well" philosophy, it would be easier to debug. You don't debug humans, you train them. Yes, the training is imperfect and doesn't always stick. Which is why we have yet another imperfect system of checking and certifying the training has been done before release. And then we have yet another system of bug catchers that patrol around making sure that the training is being adhered to. And then a system to revoke certification if we detect to many issues with the driver. I would like to trust a self driving car, however it has also been programed by the very same humans that we don't want to trust. And I doubt that the review is nearly as robust as it should be. Also the code is not open so I can't review it even if I wanted to because "trade secrets" which might just translate to "we are way to embarrassed to let others see this code because of how shitty it is". Not being easy to upgrade might be a feature in this case; Right now I don't have to deal with a pushed update that makes all 2024 model Chevy SUV platform encounter a race condition that adversely affects lane keeping after being passed on the left side while in a turn on my morning commute.

jvanderbot 3y ago

As a robotics person outside this space, it's not amazing that the shift to L4 as a goal has not been clearly grok'd by the public. After years of rhetoric, most people still believe that cars will be fully autonomous because they just don't understand the limitations. It's also quite amazing to me we, the so called smart ones, ever believed level 5 was happening but those were the good old days, huh.

faeriechangling 3y ago
It's hard to impossible for me to know how difficult level 5 driving even is since we don't have any machine that can do it. Trying to achieve it has been a good example of turtles all the way down. At the same time I feel like people are ignoring how well self-driving works RIGHT NOW, the rate of accidents is still less than with human drivers. So I'm still pretty optimistic.
- paulryanrogers 3y ago
  How can you know if Tesla won't publicly release stats of interventions or accident per 100k miles?
  - bluGill 3y ago
    There is Cruise and Waymo data as well. Though I have not seen anyone independent who has access to that data making any claims about it being safer.
- jvanderbot 3y ago
  > It's hard to impossible for me to know how difficult level 5 driving even is I have slightly a different opinion. I think you can easily guesstimate how difficult something will be by listing the "new" stuff. But I think you can't easily estimate how easy something will be because you can't list everything, so anything you miss could easily delay. Said another way: We should have known it would take at least a decade, but should never have said it will take at most two decades. (as technologists). And yeah - agree - lvl 3 is pretty good, feature-wise.
lern_too_spel 3y ago
L4 is all that's needed for a taxi service, and that's where all the money is. Tesla sold cars with "FSD" to customers living anywhere, so they have to solve L5, and they have to do it with fewer sensors and less compute. Of course it's not going to work. We're all just waiting for the other shoe to drop for the inevitable lawsuits.

ado__dev 3y ago

FSD is a Level 2 system and not autonomous as stated by Tesla in legal filings. So yes there is a huge difference between a real self-driving system and whatever the hell Tesla is selling.

ajross 3y ago

It's a supervised system as deployed. How much "blind faith" are you really seeing vs. what you think you're interpreting. None of us are taking naps in the cars. The system does continue to make mistakes, though virtually always of the "too conservative" variety. But it's also the most fun I've had in a vehicle in my whole life and an absolutely sale-maker for me personally. I simply won't buy a vehicle from another vendor until it can do this. So... I guess I repeat the question: where are you seeing "Tesla fans" using FSD unsupervised? I mean the linked article is almost unique in that it seems to have happened. And why? Well, it's buried but it turns out the driver was on autopilot because he was absolutely hammered: > Von Ohain and Rossiter had been drinking, and an autopsy found that von Ohain died with a blood alcohol level of 0.26 — more than three times the legal limit

ra7 3y ago
> So... I guess I repeat the question: where are you seeing "Tesla fans" using FSD unsupervised? One of the biggest Tesla influencers using FSD without his hands on the wheel: https://www.youtube.com/watch?v=fFW4e0Pkz_Y https://www.youtube.com/watch?v=fFW4e0Pkz_Y This is explicitly disallowed by FSD terms and conditions. He does this all the time.
- ajross 3y ago
  That seems like a bad faith argument. Do you genuinely feel that driving with your hands in your lap constitutes dangerous operation? The car asks for your hands on the wheel because that's the primary sensor it uses to monitor attention. In particular, do you genuinely feel that driving an FSD car with your hands in your lap constitutes the kind of inattention we were actually talking about (driving with a .26 BAC!). I mean, you don't freak out if you see someone using cruise control without their foot hovering over the brake, do you? Objectively and largely inarguably, braking control latency is more important to safety than steering latency.
  - ra7 3y ago
    Yes, I genuinely feel driving with my hands on my lap is dangerous. I own a Tesla, use Autopilot (not FSD, but applies to that as well) and always have my hands on the wheel because I know, at highway speeds, I may not be able to react quickly that one time. This is the entire reason Tesla has hands on the wheel requirement. And the reason for having that requirement, which is crux of the problem, is that the system is incomplete and incapable of safe operation.

bkandel 3y ago

When I was in a Tesla Uber, I found watching the front screen that shows where cars are around you frightening. Cars would frequently move around and change position unexpectedly and in a way that had nothing to do with actual car movements, lane markings would be totally off, etc. If that screen is at all representative of what FSD "thinks" is going on, I'm not at all surprised that there are lots of crashes when relying on the system.

fisherjeff 3y ago
YES. Thank you. I rode in a Model Y last year and just could not believe the mistakes it was making. Disappearing and reappearing cars; the semi in front of us was apparently straddling the lane line for a few miles; somehow an early-2000s Dodge Ram was classified as a small sedan – the list goes on, and this was only a ~10-minute ride. I would be absolutely mortified if a product of mine ended up in front of a customer in that state.
- eschneider 3y ago
  I spent a decade working on commercial computer vision applications that, among other things, had to recognize and track cars. Those are exactly the sort of transient errors you'd expect to see in shipping products and they usually have heuristics to "smooth" over those sorts of problems. That said, would I ever trust my life to a system like that? No.
  - oscb 3y ago
    I'm actually surprised they show it as raw as it looks. Doesn't inspire too much confidence, even though I bet the system must be reclassifying and changing way faster than it renders things on screen.
  - fisherjeff 3y ago
    Don’t get me wrong, my background is also pretty CV-heavy, and I don’t expect perfection by any means. But the display itself serves basically no purpose besides looking cool, and it just fails pretty badly at that. Also yeah, it made me maybe a little more nervous about being on the road with a Tesla than it should’ve…

myko 3y ago

Just sold my Model S and bought a Lyriq and am incredibly surprised at how much better Super Cruise is than FSD. It drives just like a human, incredibly smooth. Of course, it won't work on regular streets - just freeways, but maybe that's for the best, considering how inconsistent FSD is.

themaninthedark 3y ago

As an engineer, it boggles my mind how much people trust software in general. Toyota was put under a microscope after the Prius crashes....do you think anything has changed since? Do you think the software running the cars has gotten any better? VW cheated on emissions, they payed civil and criminal fines and a couple engineers got charged. No one from managment....Since then, it has come out that almost everyone was cheating. Cummins was most recently caught and it is now a civil fine(also being called record breaking in the news, at 1.6B where VW was 1.5B civil but 4.3B total...) We normalize the behavior and decrease the penalties.

bluGill 3y ago
I trust humans even less than software. At least software can be debugged over time and is consistent so if we get it working...
- avgDev 3y ago
  I would imagine debugging something like FSD might be difficult. Not sure what kind of data they store, but no situation might be the same.
  - bluGill 3y ago
    Every try to debug a human? Worse, humans are not copies so making one better doesn't make it easy to upgrade the rest of us.
    - themaninthedark 3y ago
      If the software was broken up in a sane manner and followed the "do one thing and do it well" philosophy, it would be easier to debug. You don't debug humans, you train them. Yes, the training is imperfect and doesn't always stick. Which is why we have yet another imperfect system of checking and certifying the training has been done before release. And then we have yet another system of bug catchers that patrol around making sure that the training is being adhered to. And then a system to revoke certification if we detect to many issues with the driver. I would like to trust a self driving car, however it has also been programed by the very same humans that we don't want to trust. And I doubt that the review is nearly as robust as it should be. Also the code is not open so I can't review it even if I wanted to because "trade secrets" which might just translate to "we are way to embarrassed to let others see this code because of how shitty it is". Not being easy to upgrade might be a feature in this case; Right now I don't have to deal with a pushed update that makes all 2024 model Chevy SUV platform encounter a race condition that adversely affects lane keeping after being passed on the left side while in a turn on my morning commute.