3 ms·

It does always devolve into the same discussion, I agree. All of the evidence says it doesn't work well. One guy says "but it does for me!" And the discussio

It does always devolve into the same discussion, I agree. All of the evidence says it doesn't work well. One guy says "but it does for me!" And the discussion ends. In some situations with well marked, simple road design, it works great. Give it a multiple-lane turn intersection to navigate and things get interesting. Give it a three lane turn where the outermost lane is fairly sharp, and it will just cut across the middle lane because it seems to have a built-in limit on how hard it can turn the wheel. What's noteworthy is that it doesn't significantly improve with each revision. It'll get a little better, then regress, over and over, but still the fundamentals are unreliable. At some point maybe it becomes clear that the current strategy is topped out. I do think it tends to be better behaved than AP on the highway, though, so I look forward to a stack merge there.

ajross 3y ago
> All of the evidence says it doesn't work well. Except actual accident statistics, you mean. > Give it a multiple-lane turn intersection to navigate and things get interesting. Give it a three lane turn where the outermost lane is fairly sharp, and it will just cut across the middle lane because it seems to have a built-in limit on how hard it can turn the wheel. Now you've digressed into navigation choices and not safety. And... I agree! This is exactly what I was talking about upthread. It's an issue for real improvement, and notably not something that can be helped by LIDAR at all. Tesla is actually way ahead here, as I see it. Waymo and Cruise operate in heavily constrained environments that disallow challenging maneuvers entirely. (FWIW: that three-lane turn behavior you're talking about doesn't seem familiar. I don't see that with my car, and we have a few intersecions like that around here. In general multi-lane turns are marked, and the car is quite good about figuring it out.)
- rootusrootus 3y ago
  > Except actual accident statistics, you mean. I thought we had gotten past that statistic? AP, in particular, gets used in what is already the safest driving situation there is. Those statistics can only be compared to human drivers in exactly that same narrow circumstance. If we were talking about FSD, then we'd need to see statistics where the human never touches the wheel; saving the car from it's own mistakes is cheating ;). > Now you've digressed into navigation choices and not safety. I think I understand what you are saying but I'm not sure I agree -- incorrectly navigating across the lane lines, especially when the car indicates it sees them, is definitely a safety problem in my view. I expect that it's kinda like the silly wipers. If you live somewhere with the right kind of rain, you think "what is everyone complaining about?" but if you live in the PNW where we get a lot of misty rain, you end up cursing at the car regularly during the winter months. Some places have just the right roads for FSD, and those people think it's pretty great. But then other people get constant failures and wonder why the heck they spent 6, 12, or 15 grand on something that doesn't deliver. I'm glad someone has a good experience with it, even when it doesn't match my own. But I also didn't buy it outright, so I just cancel the subscription after trying it again to see how things stand.
CRConrad 3y ago
rootusrootus: > It does always devolve into the same discussion, I agree. All of the evidence says it doesn't work well. One guy says "but it does for me!" ajross: It works on my machine!

ajross 3y ago

> All of the evidence says it doesn't work well. Except actual accident statistics, you mean. > Give it a multiple-lane turn intersection to navigate and things get interesting. Give it a three lane turn where the outermost lane is fairly sharp, and it will just cut across the middle lane because it seems to have a built-in limit on how hard it can turn the wheel. Now you've digressed into navigation choices and not safety. And... I agree! This is exactly what I was talking about upthread. It's an issue for real improvement, and notably not something that can be helped by LIDAR at all. Tesla is actually way ahead here, as I see it. Waymo and Cruise operate in heavily constrained environments that disallow challenging maneuvers entirely. (FWIW: that three-lane turn behavior you're talking about doesn't seem familiar. I don't see that with my car, and we have a few intersecions like that around here. In general multi-lane turns are marked, and the car is quite good about figuring it out.)

rootusrootus 3y ago
> Except actual accident statistics, you mean. I thought we had gotten past that statistic? AP, in particular, gets used in what is already the safest driving situation there is. Those statistics can only be compared to human drivers in exactly that same narrow circumstance. If we were talking about FSD, then we'd need to see statistics where the human never touches the wheel; saving the car from it's own mistakes is cheating ;). > Now you've digressed into navigation choices and not safety. I think I understand what you are saying but I'm not sure I agree -- incorrectly navigating across the lane lines, especially when the car indicates it sees them, is definitely a safety problem in my view. I expect that it's kinda like the silly wipers. If you live somewhere with the right kind of rain, you think "what is everyone complaining about?" but if you live in the PNW where we get a lot of misty rain, you end up cursing at the car regularly during the winter months. Some places have just the right roads for FSD, and those people think it's pretty great. But then other people get constant failures and wonder why the heck they spent 6, 12, or 15 grand on something that doesn't deliver. I'm glad someone has a good experience with it, even when it doesn't match my own. But I also didn't buy it outright, so I just cancel the subscription after trying it again to see how things stand.

CRConrad 3y ago

rootusrootus: > It does always devolve into the same discussion, I agree. All of the evidence says it doesn't work well. One guy says "but it does for me!" ajross: It works on my machine!