Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

So, when is APv2 going to be noticeably better than APv1? Or is it already?

This site may earn commission on affiliate links.
This scenario is just so unique in the whole "continuous improvement" thing so far. Imagine how much better the system would be today if they were allowed to continue working with MobilEye vision (eg. if the Intel acquisition happened earlier, preventing the divorce -- given Intel's willingness to play by Elon's rules that is). Even with "just" AP1 hardware, there has been little to no focus on further improvements (or attention to the browser for that matter) as all development cycles seem to consumed recovering from the divorce.
 
This scenario is just so unique in the whole "continuous improvement" thing so far. Imagine how much better the system would be today if they were allowed to continue working with MobilEye vision (eg. if the Intel acquisition happened earlier, preventing the divorce -- given Intel's willingness to play by Elon's rules that is). Even with "just" AP1 hardware, there has been little to no focus on further improvements (or attention to the browser for that matter) as all development cycles seem to consumed recovering from the divorce.
I'd have expected APv2 to be so much better than v1 by now that I'd have gone to buy a new model S, but honestly I never expected APv1 to get much better. In case you don't know the details of the APv1 hardware, it's really quite limited. That's why I'm actually pretty happy with APv1 performance given what it has to work with.
The hardware is limited/insufficient enough that when I went to buy my model S as APv1 was just coming out, that I told the salesperson "you know that those sensors you have are totally insufficient for what you're trying to do, I'll be impressed if AP can perform well with them".
I'll gladly say that I was somewhat proven wrong, but again there is only so much APv1 can do with the onboard hardware.
There is however indeed little excuse for APv2 not doing more as of yet, except lack of sufficient software engineering (IMO).
The good news is that it can get better over time, but the longer it takes, the more tesla will get leapfrogged by someone else.
 
Things got better from December '16 until May when 17.17.4 came out. What I mean is that measurable progress happened during that time. Since then they introduced a new control algorithm that was supposed to result in "smoother" operation, but actual usability or reliability or safety or some other difficult to quanitify quality has regressed or at least not progressed. I used 17.17.4 every day; I never use AP any more except to try a new version.

There is no extrapolation from actual data to indicate that AP2 will ever be even close to what EAP was sold as or for that matter even as good as AP1. We can have faith in Tesla or whatever, but there is no data. And since it actually got worse after May (June?) and it still isn't fixed, there's a real possibility that the development system is "out of control" and no longer actually producing improvements. It's kind of weird.
 
In my experience, AP1 almost never identifies previously untracked stopped cars that you come upon. However, AP2 has become pretty good at it (probably because AP2 uses vision more, and AP1 relies almost entirely on radar). I would say that AP2 picks up stopped cars that were never tracked in your direct lane about 95% of the time.

In all other regards, AP2 performs worse than AP1 (and, at least for me, AP2 has gotten worse with every release since 17.4. I no longer use it on local or highway and have fallen back to TACC (except for stop-and-go, where it is OK and easy to monitor safely). There are just too many dangerous and unpredictable lurches and lane departures.
 
  • Like
Reactions: 1 person
In short, AP2 is supposed to be better than AP1 at some point in the as yet undefined future (don't believe me, just tweet Elon). At this point, I've just got a bunch of extra cameras around my car that either make me look really cool or really shady depending on where I'm parked.
 
  • Funny
Reactions: davidc18
In my experience, AP1 almost never identifies previously untracked stopped cars that you come upon. However, AP2 has become pretty good at it (probably because AP2 uses vision more, and AP1 relies almost entirely on radar). I would say that AP2 picks up stopped cars that were never tracked in your direct lane about 95% of the time.

In all other regards, AP2 performs worse than AP1 (and, at least for me, AP2 has gotten worse with every release since 17.4. I no longer use it on local or highway and have fallen back to TACC (except for stop-and-go, where it is OK and easy to monitor safely). There are just too many dangerous and unpredictable lurches and lane departures.
What's interesting is the rate of new firmware releases.
Don't ask me what they are fixing, or what new features are being added because most of them feel exactly the same.
But someone's doing something.
 
In my experience, AP1 almost never identifies previously untracked stopped cars that you come upon. However, AP2 has become pretty good at it (probably because AP2 uses vision more, and AP1 relies almost entirely on radar). I would say that AP2 picks up stopped cars that were never tracked in your direct lane about 95% of the time.

In all other regards, AP2 performs worse than AP1 (and, at least for me, AP2 has gotten worse with every release since 17.4. I no longer use it on local or highway and have fallen back to TACC (except for stop-and-go, where it is OK and easy to monitor safely). There are just too many dangerous and unpredictable lurches and lane departures.
My AP1 always detects and stops behind untracked stopped cars. I always monitor, and it always has worked.
 
Things got better from December '16 until May when 17.17.4 came out. What I mean is that measurable progress happened during that time. Since then they introduced a new control algorithm that was supposed to result in "smoother" operation, but actual usability or reliability or safety or some other difficult to quanitify quality has regressed or at least not progressed. I used 17.17.4 every day; I never use AP any more except to try a new version.

There is no extrapolation from actual data to indicate that AP2 will ever be even close to what EAP was sold as or for that matter even as good as AP1. We can have faith in Tesla or whatever, but there is no data. And since it actually got worse after May (June?) and it still isn't fixed, there's a real possibility that the development system is "out of control" and no longer actually producing improvements. It's kind of weird.

There was, of course, another event that happened right around the same time as that particular software release within the company... I'm hopeful that Karpathy is just turning things around and getting everything back on track.
 
But perhaps that might change now that Intel owns ME.

I agree! Intel change of hands of the AP1 technology gives Tesla a way to save face while getting something back ... It's sitting over in Santa Clara on a bench and maybe by now Intel has put enough of their own juice into the tech it could possibly pass as "Intel Inside" instead of EyeQ ..

Or, in a few weeks AMD will have cloned it, so Tesla could buy the vision tech from them for a few bucks less per car.
 
  • Like
Reactions: HankLloydRight