6 ms·

That's my experience, having been on FSD beta for almost two years now. It's still "un-human-driver-like" in a bunch of surprising ways, but the actual meat an

by ajross 3y ago

That's my experience, having been on FSD beta for almost two years now. It's still "un-human-driver-like" in a bunch of surprising ways, but the actual meat and potatoes "does it do more or less stupid stuff" is I think ahead of my own driving at this point. At this point I trust its microdriving (lanekeeping, distance management, speed control, collision/obstacle avoidance, etc...) more than my own for sure. The higher level stuff is still a mixed bag (will it get stuck at that merge, can it make this lane change in time, does it understand that this other car is going to yield or will it stop, etc...).

Dylan16807 3y ago
Can it detect stationary obstacles at freeway speeds? Do we have good evidence? Because autopilot couldn't.
- ajross 3y ago
  Meh, we've been here before. That's a vague and impossible point to argue. I mean, sure, yes it does. Does it for all possible obstacles? Probably not? I certainly haven't tested. This kind of "$SYSTEM must have feature $X which is possessed by $COMPETITOR or else it will fail" is just plain bad analysis. It's possible for a Tesla to (1) be a better driver than you on the whole while still (2) doing weird stuff that you're sure you never would. FWIW: Stationary junk in the freeway causes accidents with human drivers every day, all the time. People are absolutely terrible at dodging crap in the road. But you and everyone else will swear up and down that you'd *never* drive over a shed tire or whatever, even though evidence says you totally would. Basically: no, you'll never get the evidence you want to "prove" the logical construct you've created. So Teslas, to you, will never be "safe". But in the real world they obviously are.
  - Dylan16807 3y ago
    > to "prove" the logical construct you've created You're making guesses about my standards that aren't warranted at all. They absolutely can be met. First off I'm specifically worried about large objects, a meter or larger. And let's see, autopilot has done 3 billion miles? Okay, if FSD can manage 1 billion miles of freeway driving with zero or one crashes into large objects, then I'll be convinced they're solving the problem. That should be fair, right? > But in the real world they obviously are. Obvious according to what? Autopilot wasn't, because it managed to hit entire vehicles repeatedly. There is (or was) a specific warning in the manual that it might hit stationary vehicles when going over 50mph. What makes it obvious that FSD avoids this specific problem? > But you and everyone else will swear up and down that you'd never drive over a shed tire or whatever, even though evidence says you totally would. You made this up about me from nowhere. Please never do that to anyone.
    - ajross 3y ago
      > You're making guesses about my standards that aren't warranted at all. They absolutely can be met. First off I'm specifically worried about large objects, a meter or larger. I've personally watched my car go around trash cans and bikes in the road, and (obviously) stop for halted vehicles. So, you'll abandon this argument and concede the point? I suspect you won't, because my anecdote isn't enough for you, and those obstacles aren't sufficient proof, etc... I think I'm more right than you want to admit. There's nothing I can say here to change your mind, and so coming here and demanding "evidence" isn't really an argument in good faith. And an edit just to pick on this bit: > Okay, if FSD can manage 1 billion miles of freeway driving with [emphasis in original] zero or one crashes into large objects, then I'll be convinced they're solving the problem. That's simply a ridiculous argument. That level of safety is way, way, WAY beyond anything you get from existing transportation systems of any kind, period. A quick google shows that in that "one billion miles on the freeway", you'd expect not merely the "one accident" you're demanding, but in fact TWENTY FATALITIES (I couldn't find statistics for mere collisions, but needless to say it's going to be at least an order of magnitude or two higher). So basically you're sitting here and glibly demanding that this product be 200x safer than its competitors before you'll consider it acceptable... and getting huffy when I call you unserious?
      - Dylan16807 3y ago
        > So, you'll abandon this argument and concede the point? I suspect you won't, because my anecdote isn't enough for you, and those obstacles aren't sufficient proof, etc... Are you trolling with this? I'm asking if it can reliably do that. Of course I need more data than one person can collect. I am not being unreasonable to ask for more data than your personal experience. Most people with autopilot never saw this problem either. And specifically I need to know about freeway speeds, because speed is an important factor. There is plenty you could do to change my mind. If you link to something published by tesla or a government body showing autopilot and FSD accident rates by type then that would be more than enough. > So basically you're sitting here and glibly demanding that this product be 200x safer than its competitors before you'll consider it acceptable I'm not asking for a lack of fatalities. I'm asking for a lack of hitting stationary vehicle-scale objects on the freeway. Do you think that particular kind of accident is responsible for a majority of fatalities, or something? My expectation is that it's a very rare kind of accident and I also feel like it's a good canary. Also the freeway fatality rate is about 5.4 per billion miles, not 20.

by ajross 3y ago

Dylan16807 3y ago

Can it detect stationary obstacles at freeway speeds? Do we have good evidence? Because autopilot couldn't.

ajross 3y ago
Meh, we've been here before. That's a vague and impossible point to argue. I mean, sure, yes it does. Does it for all possible obstacles? Probably not? I certainly haven't tested. This kind of "$SYSTEM must have feature $X which is possessed by $COMPETITOR or else it will fail" is just plain bad analysis. It's possible for a Tesla to (1) be a better driver than you on the whole while still (2) doing weird stuff that you're sure you never would. FWIW: Stationary junk in the freeway causes accidents with human drivers every day, all the time. People are absolutely terrible at dodging crap in the road. But you and everyone else will swear up and down that you'd *never* drive over a shed tire or whatever, even though evidence says you totally would. Basically: no, you'll never get the evidence you want to "prove" the logical construct you've created. So Teslas, to you, will never be "safe". But in the real world they obviously are.
- Dylan16807 3y ago
  > to "prove" the logical construct you've created You're making guesses about my standards that aren't warranted at all. They absolutely can be met. First off I'm specifically worried about large objects, a meter or larger. And let's see, autopilot has done 3 billion miles? Okay, if FSD can manage 1 billion miles of freeway driving with zero or one crashes into large objects, then I'll be convinced they're solving the problem. That should be fair, right? > But in the real world they obviously are. Obvious according to what? Autopilot wasn't, because it managed to hit entire vehicles repeatedly. There is (or was) a specific warning in the manual that it might hit stationary vehicles when going over 50mph. What makes it obvious that FSD avoids this specific problem? > But you and everyone else will swear up and down that you'd never drive over a shed tire or whatever, even though evidence says you totally would. You made this up about me from nowhere. Please never do that to anyone.
  - ajross 3y ago
    > You're making guesses about my standards that aren't warranted at all. They absolutely can be met. First off I'm specifically worried about large objects, a meter or larger. I've personally watched my car go around trash cans and bikes in the road, and (obviously) stop for halted vehicles. So, you'll abandon this argument and concede the point? I suspect you won't, because my anecdote isn't enough for you, and those obstacles aren't sufficient proof, etc... I think I'm more right than you want to admit. There's nothing I can say here to change your mind, and so coming here and demanding "evidence" isn't really an argument in good faith. And an edit just to pick on this bit: > Okay, if FSD can manage 1 billion miles of freeway driving with [emphasis in original] zero or one crashes into large objects, then I'll be convinced they're solving the problem. That's simply a ridiculous argument. That level of safety is way, way, WAY beyond anything you get from existing transportation systems of any kind, period. A quick google shows that in that "one billion miles on the freeway", you'd expect not merely the "one accident" you're demanding, but in fact TWENTY FATALITIES (I couldn't find statistics for mere collisions, but needless to say it's going to be at least an order of magnitude or two higher). So basically you're sitting here and glibly demanding that this product be 200x safer than its competitors before you'll consider it acceptable... and getting huffy when I call you unserious?
    - Dylan16807 3y ago
      > So, you'll abandon this argument and concede the point? I suspect you won't, because my anecdote isn't enough for you, and those obstacles aren't sufficient proof, etc... Are you trolling with this? I'm asking if it can reliably do that. Of course I need more data than one person can collect. I am not being unreasonable to ask for more data than your personal experience. Most people with autopilot never saw this problem either. And specifically I need to know about freeway speeds, because speed is an important factor. There is plenty you could do to change my mind. If you link to something published by tesla or a government body showing autopilot and FSD accident rates by type then that would be more than enough. > So basically you're sitting here and glibly demanding that this product be 200x safer than its competitors before you'll consider it acceptable I'm not asking for a lack of fatalities. I'm asking for a lack of hitting stationary vehicle-scale objects on the freeway. Do you think that particular kind of accident is responsible for a majority of fatalities, or something? My expectation is that it's a very rare kind of accident and I also feel like it's a good canary. Also the freeway fatality rate is about 5.4 per billion miles, not 20.