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

[marcmerlin]

I'm well aware that the hardware on APv2 is substantially better, but I've failed to find any information of whether APv2 currently has an AP that behaves functionally better than APv1. I have v1 and I've been waiting for v2 to be sufficiently better for me to consider upgrading (i.e. changing cars).

1) APv1 is of course blind to red lights and street signs, and isn't supposed to be used on surface streets anyway. I believe the same is true of APv2 today (not talking about the promised future on the tesla web site)
2) The fact that APv1 is able to change lanes without creating an accident on most days seems like a mix of miracle and good luck. It seems pretty blind to cars that may have snuck up from behind and can switch into a lane while cutting someone over. Is APv2 better today?
3) APv1 is totally blind to cones or anything that may block my lane, or close the lane. I have found nothing saying that APv2 is better there. Is it?
4) APv1 does not always see stopped traffic in front of me. Usually, but not always. That's obviously scary.
5) APv1 is totally blind to a car in the next lane putting its blinker and changing into my lane, expecting that I'll slow down (AP won't since it doesn't see that) until the car is 50% in front of my car, and only then will my car slam on its brakes to avoid hitting the other car (at least I'm assuming so, since I don't let it drive that long when someone is merging in front of me). Again, is APv2 better here?
6) Not counting the above, is APv2 significantly better than APv1 today in some scenarios?
7) Is there any real info that it will be very soon?

In the meantime:
Tesla Collecting More Data As Engineers Quit Autopilot Division - Gas 2
Tesla's Autopilot software head leaves the company six months after joining from Apple
Some Tesla Engineers Think Autopilot Isn’t Safe

this does not fill me with fuzzies...

 

[calisnow]

Thank you for bringing up this topic - it's high time someone showed the courage to discuss the elephant in the room.

 

[No2DinosaurFuel]

As far as I can tell ap2 is still worst than ap1. Who knows when it will be better. I think tesla should just contract the work to Intel who now owns mobieye. Mobieye was the developer for ap1 and they did a he'll of a job getting it super reliable as much as it is now.

 

[verygreen]

I'm well aware that the hardware on APv2 is substantially better, but I've failed to find any information of whether APv2 currently has an AP that behaves functionally better than APv1. I have v1 and I've been waiting for v2 to be sufficiently better for me to consider upgrading (i.e. changing cars).

1) APv1 is of course blind to red lights and street signs, and isn't supposed to be used on surface streets anyway. I believe the same is true of APv2 today (not talking about the promised future on the tesla web site)

AP1 is not blind to traffic lights, the code is just not enabled in the customer cars.

2) The fact that APv1 is able to change lanes without creating an accident on most days seems like a mix of miracle and good luck. It seems pretty blind to cars that may have snuck up from behind and can switch into a lane while cutting someone over. Is APv2 better today?
3) APv1 is totally blind to cones or anything that may block my lane, or close the lane. I have found nothing saying that APv2 is better there. Is it?
4) APv1 does not always see stopped traffic in front of me. Usually, but not always. That's obviously scary.
5) APv1 is totally blind to a car in the next lane putting its blinker and changing into my lane, expecting that I'll slow down (AP won't since it doesn't see that) until the car is 50% in front of my car, and only then will my car slam on its brakes to avoid hitting the other car (at least I'm assuming so, since I don't let it drive that long when someone is merging in front of me). Again, is APv2 better here?

AP2 only uses two front cameras so it sees about as much with them.
there's same radar I believe so the rest is about the same.

6) Not counting the above, is APv2 significantly better than APv1 today in some scenarios?

Yes! When you get into an accident, AP1 only saved 8 pictures at 1fps prior to it.
With AP2 you (meaning Tesla) get gorgeous 30FPS from two front cameras for 10 seconds around the event and 1 fps from the rest of the cameras. This is much better.

Reportedly AP2 can see stopped cars that it never saw as moving before where as AP1 cannot. I never owned AP1 so cannot really tell you if it's true. (I mean AP2 most definitely sees stopped cars, just no idea about what AP1 did)

7) Is there any real info that it will be very soon?

We know about as much as you do.

 

[TaoJones]

So much misinformation. A sticky with the Cliff Notes version would be helpful.

AP1 is not "blind to street signs". In fact, and as an example, AP1 both recognizes and reacts to speed limit signs - AP2 does neither, relying instead upon a database riddled with errors.

AP2 is in no way better than AP1. I used to append the qualifier "yet", but now, after 60K+ miles with AP1 and 20K miles with AP2, as well as after the promise of parity and that video from/by December 2016, it's all I can do to remain a shareholder.

AP2 has been getting closer to AP1 on the highway, except that the new and problematic phantom braking behavior has been introduced - and that is a deal breaker for those not willing to be beta testers. Don't think so? Try having them as passengers during such episodes.

What really highlighted how far AP2 yet has to go was a recent road trip through a couple of western Canadian provinces. Those diagonal dotted lines were an adventure.

TACC continues to be worth the price of the car, but Autosteer has regressed. Every day it disengages constantly.

Speaking of TACC, I did recently witness it recognize a cyclist in front of me - unlike AP1 which displayed a bicycle icon, AP2 just displayed a car icon but it's something. v*.34*cfc.

Just tonight during a football game, there were ads for Mazda and Volkswagen advanced driving features. Will be interesting to see who wins this battle. I used to think Tesla's lead was insurmountable - or should I say MobilEye's. Tesla's infrastructure lead certainly is phenomenal.

Not so much the driving experience with Autosteer via AP2.

Nice safe ride though.

 

[marcmerlin]

Hi TaoJones: I wrote wrong as you noticed. I meant to say "blind to stop signs". Now, it's possible that maybe it's not blind to them but of course doesn't react to them at this time (which is as advertised). I didn't know it could tell red lights but that the code was not enabled (which again is understandable since it's not meant to work on surface streets).
To be clear, I am not trying to say "why isn't APv1 even better", when I genuinely believe APv1 is actually really damn good considering how limiting its hardware, is. Of course, that also means it's good enough that it gives some confidence that most of us probably shouldn't have in it, which is why I am still waiting for APv2 to provide that extra 5-10x safety level that it should be able to given the right software.
I didn't know that APv2 still wasn't at least on parity with APv1 (yet), I assumed it had at least caught up. Thank you for that information, as disappointing as it actually is.

Anyway, it sounds that for better or for worse, I should stick with my APv1 for some while longer, and that's very useful information, thank you.
I sure hope Tesla fixes its autopilot engineering. It's a hard enough problem that they can't afford to have employees just leave out of disgust, or fear of being responsible for cars that crash, with their name in the code that did it.

 

[oktane]

My guess is somewhere between now and never unfortunately.

 

[BigD0g]

Here we go again....

I think the answer is yes AP2 is better then AP1 in "some" limited circumstances. However, we have a bit of a way to go before we eclipse AP1, it's coming soon...

 

[malcolm]

Depends who you ask.

Varies from: Truly we are living in a golden age of autonomy...

To: It's No, Nay, Never. No Nay Never No More...

They'll all be along here shortly.

 

[malcolm]

Yes! When you get into an accident, AP1 only saved 8 pictures at 1fps prior to it.
With AP2 you (meaning Tesla) get gorgeous 30FPS from two front cameras for 10 seconds around the event and 1 fps from the rest of the cameras. This is much better.

Can't wait for them to add the driver facing camera.

"Hey, look everyone. This is the exact moment when the driver presses what he thinks is the brake."

 

[RDoc]

<snip>
Reportedly AP2 can see stopped cars that it never saw as moving before where as AP1 cannot. I never owned AP1 so cannot really tell you if it's true. (I mean AP2 most definitely sees stopped cars, just no idea about what AP1 did)

This is a forum legend. AP1 definitely sees and reacts to stopped cars that it's never seen moving. (and before we go off into a doppler radar riff, look up how car radars actually work)

 

[marcmerlin]

This is a forum legend. AP1 definitely sees and reacts to stopped cars that it's never seen moving. (and before we go off into a doppler radar riff, look up how car radars actually work)

Err, that's not a legend at all. It's happened to me 2-3 times.
I was reporting a previous bug I had found (saved in my phone so that I don't forget) and testing how it would react to a car stopped at a red light. Car never saw it, I waited as long as I felt I could before I slammed on the brakes:

.
As you can see from the date, that was Feb 2016. It's possible that this specific case has been fixed though. I can test it again when I get the chance.

As for car is blind from the sides, I hope you believe that one is real too if not:

Of course, that's not very surprising given that the camera focusses on what's ahead, and you can't expect a parking sensor to see something on the sides like this, until it's too late I guess.

 

[mbhforum]

Err, that's not a legend at all. It's happened to me 2-3 times.
I was reporting a previous bug I had found (saved in my phone so that I don't forget) and testing how it would react to a car stopped at a red light. Car never saw it, I waited as long as I felt I could before I slammed on the brakes:

. As you can see from the date, that was Feb 2016. It's possible that this specific case has been fixed though

As for car is blind from the sides, I hope you believe that one is real too if not:

Of course, that's not very surprising given that the camera focusses on what's ahead, and you can't expect a parking sensor to see something on the sides like this, until it's too late I guess.

If you check out my link below, I got into an accident with a construction cone with AP 1.0 and it was straight ahead of me as the lane was being closed off and I had my eyes off the road for a brief few seconds which was my fault.. I have AP 1.0 and cannot comment on how AP 2.0 would have handled this, but I was not warned at all as I hit it straight on.

Autopilot and Pay attention!!!

 

[verygreen]

Can't wait for them to add the driver facing camera.

"Hey, look everyone. This is the exact moment when the driver presses what he thinks is the brake."

For that driver-facing camera is not enough, you also need one in the driver side footwell.

 

[marcmerlin]

If you check out my link below, I got into an accident with a construction cone with AP 1.0 and it was straight ahead of me as the lane was being closed off and I had my eyes off the road for a brief few seconds which was my fault.. I have AP 1.0 and cannot comment on how AP 2.0 would have handled this, but I was not warned at all as I hit it straight on.

Autopilot and Pay attention!!!

Right, so you basically had the same issue than me with

,
except hopefully your cones were smaller than the big ones I would have hit.
Or did the local road people actually just put cones in the middle of the lane without warning and without slowly closing the lane from the side, first?

 

[xyeahtony]

AP2.0 works fine for me but YMMV

 

[Saghost]

Err, that's not a legend at all. It's happened to me 2-3 times.
I was reporting a previous bug I had found (saved in my phone so that I don't forget) and testing how it would react to a car stopped at a red light. Car never saw it, I waited as long as I felt I could before I slammed on the brakes:

.
As you can see from the date, that was Feb 2016. It's possible that this specific case has been fixed though. I can test it again when I get the chance.

As for car is blind from the sides, I hope you believe that one is real too if not:

Of course, that's not very surprising given that the camera focusses on what's ahead, and you can't expect a parking sensor to see something on the sides like this, until it's too late I guess.

AP1 (and AP2, for that matter) will not correctly identify and react to every stopped car, every time - so far.

AP1 cars definitely do identify and react to stopped cars that they never saw move before in some (most?) cases. I don't have any way to know if one reacts to more cars than the other, but I would guess the rates are likely similar.

The radar whitelist is one of the tools Tesla is using to improve this rate (at the cost of some additional false positive braking events) - and it applies to both sets of hardware.

 

[Freedom101]

...except hopefully your cones were smaller than the big ones I would have hit...

It looks like you did hit one in the video. If not that was close!

So far if I've seen any type of construction, debris, or road abnormality, I just take AP off until I'm clear of it. I've found since the last update it's not quite as much of a semi truck magnet as it used to be.

 

[marcmerlin]

It looks like you did hit one in the video. If not that was close!

So far if I've seen any type of construction, debris, or road abnormality, I just take AP off until I'm clear of it. I've found since the last update it's not quite as much of a semi truck magnet as it used to be.

I didn't hit it, but waited as long as I could before taking over, so yeah, it was close.

 

[mbhforum]

Right, so you basically had the same issue than me with

,
except hopefully your cones were smaller than the big ones I would have hit.
Or did the local road people actually just put cones in the middle of the lane without warning and without slowly closing the lane from the side, first?

Same size cone. There wasn't much warning if any at all.