.48 feels like AP2 finally passed AP1

[TaoJones]

Let's talk when we have speed sign recognition, vehicle identification display and auto-wipers in AP2... 'mmkay?

Indeed. People who haven't had AP1 are awfully quick to pronounce AP2 at parity or better.

I could give two shites about wipers, since even with auto-sensing they never cleaned the top 1/3 of my windshields adequately. This malaise continues with the SvC this past week saying the jets remain non-adjustable and are functioning within spec. Feh.

I continue to hope that the vehicle differentiated icons are low-hanging fruit that they'll get around to eventually.

However, speed limit sign identification and reaction thereto remains problematic every day here in SoCal and of course elsewhere. It's not a matter of delayed loading - the database has an alarming number of errors, which of course are reflected in resulting TACC speeds.

The ultra-positive fanboi spin for this, as I asserted in a recent well-received post (go figure) is that at some point, [speculation purely mine] Tesla will upgrade the functionality to include *both* AP1's read-by-camera approach *and* a (fixed, with periodic updates) AP2 backup database, useful for those times the car is in an inner lane and a speed limit sign is blocked by a big truck adjacent.

I don't know how much they care about this sort of thing, but there can be no "AP2 is at parity with or better than AP1" statements without considerable qualifiers unless and until pesky details such as speed limit sign recognition and reaction are addressed.

On the other side of the coin, if one cherrypicks what *does* work often in AP2 (as well as in comparison with AP1), I don't know that many would disagree with the following order across the spectrum:

TACC
AS
.
.
.
Speed limit sign validity
Vehicle ID
Auto-sensing wipers
.
.
.
EAP
FSD

2018-2019 will probably continue to be painful but with good periodic advances, topped off with Tesla AI (new boards) in 2020 and a schweet Q4 2020 transcontinental roadtrip (I believe Elon should include Canada *and* remove the voided warranty restriction for Mexico by then since after all there are SCs there now, and add at least Canada so he can say the trip was international in scope), replete with Barry White music.

 

[buttershrimp]

Indeed. People who haven't had AP1 are awfully quick to pronounce AP2 at parity or better.

I've driven a decent amount in an AP1 model S... not @TaoJones distances though.... and his criticism of AP2 is the one I most listen to because I KNOW he's driven a lot. For me, I've been disinterested in the AP1 vs. AP2 discussion partly because we aren't going back, but I also am enthusiastic and fascinated by AP2... it's my favorite thing about the car.... partly because I know that everyone at Tesla is busting their ass to impress in this area.... I think they are more focused now, not less. So I try not to get too far into the AP1 versus AP2 debate and instead try to take the approach of learning to be an AP2 sommelier of sorts.... the tannins are strong on this firmware sort of thing.

I will say, if someone wants to loan me their AP1 car for 30 minutes, I'll drive 3 roads in Austin and tell you if we are indeed at parity, I feel that confident in my perception of what the car is doing at this point that I'd be able to answer that question... but no one is likely interested in having me do this test with rap music, etc.... and I'm sure there are plenty of other people who trust their feel of the car... so I digress.

The most intriguing suggestion I've had so far to show me where AP2 is lacking has been driving 2222 in Austin.... it makes AP2 feel deficient. I plan to use it as my high water mark. I thank @brkaus for that.

If I were to go back to the AP1 car and drive it these are my working hypotheses:

1) I predict it would confirm that speed limit sign recognition and vehicle identification in surrounding lanes are the MOST important things to me for AP2 improvements.
2) I predict that AP1 will be inferior to AP2 in rain and low light conditions currently
3) I predict that AP1 autolane change will impress me only if it can recognizes if someone is in my blind spot and behaves accordingly.... if it can also avoid having to try multiple times when switching multiple lanes, that would be great.
4) I predict AP1 will be inferior on tight lane and tight curbs... something that I distinctly remember sucking on AP1... though I'm sure that's changed in some respects.
5) I predict the user experience of AP1 will exceed AP2 at this time.... and perhaps that should be the measure of parity. BUT for me it isn't... instead I try to judge AP2 on complex tasks rather than reliability compared to AP1... that's what my videos are trying to elucidate because that's what I'm always trying to push autopilot to handle. It is this area that there is the most room to criticize my view on AP2... it's a different metric and one that is more likely to disappoint if they come up short.
6) The thing I currently imagine myself to disagree with @TaoJones about on this subject is stop and go traffic. I love AP2 in this regard... it's practically superb in my experience. Part of this is that I hate using TACC because I still have to do the work and I find it's prone to the same errors... but perhaps AP1 would make me see the light in this respect.

I guess my point is this: I knew AP2 was going to be a work in progress when I bought it, but that was one of the selling points...that they have a platoon of AP soldiers working on this constantly. Even with the slower pace that some of us expected, how can you not be excited about that?

Also Consider this... Volvo called it quits today on their autonomous driving program because they want to take time to figure out how to have "the safest car on the road"....

 

[Electroman]

partly because I know that everyone at Tesla is busting their ass to impress in this area.... I think they are more focused now, not less

that they have a platoon of AP soldiers working on this constantly.

How do you know? I am not disputing you and common sense suggest that they would make this their no.1 priority, but your statements seem to imply you know someone inside Tesla working on AP. Right?

 

[buttershrimp]

How do you know? I am not disputing you and common sense suggest that they would make this their no.1 priority, but your statements seem to imply you know someone inside Tesla working on AP. Right?

It's common knowledge in terms of the number of people... the round table discussion on AI did wonders for my faith as well on autopilot... It's a big group of folks working on it, I know that. I wish I had an insider though.

 

[⚡️ELECTROMAN⚡️]

Indeed. People who haven't had AP1 are awfully quick to pronounce AP2 at parity or better.

I could give two shites about wipers, since even with auto-sensing they never cleaned the top 1/3 of my windshields adequately. This malaise continues with the SvC this past week saying the jets remain non-adjustable and are functioning within spec. Feh.

I continue to hope that the vehicle differentiated icons are low-hanging fruit that they'll get around to eventually.

However, speed limit sign identification and reaction thereto remains problematic every day here in SoCal and of course elsewhere. It's not a matter of delayed loading - the database has an alarming number of errors, which of course are reflected in resulting TACC speeds.

The ultra-positive fanboi spin for this, as I asserted in a recent well-received post (go figure) is that at some point, [speculation purely mine] Tesla will upgrade the functionality to include *both* AP1's read-by-camera approach *and* a (fixed, with periodic updates) AP2 backup database, useful for those times the car is in an inner lane and a speed limit sign is blocked by a big truck adjacent.

I don't know how much they care about this sort of thing, but there can be no "AP2 is at parity with or better than AP1" statements without considerable qualifiers unless and until pesky details such as speed limit sign recognition and reaction are addressed.

On the other side of the coin, if one cherrypicks what *does* work often in AP2 (as well as in comparison with AP1), I don't know that many would disagree with the following order across the spectrum:

TACC
AS
.
.
.
Speed limit sign validity
Vehicle ID
Auto-sensing wipers
.
.
.
EAP
FSD

2018-2019 will probably continue to be painful but with good periodic advances, topped off with Tesla AI (new boards) in 2020 and a schweet Q4 2020 transcontinental roadtrip (I believe Elon should include Canada *and* remove the voided warranty restriction for Mexico by then since after all there are SCs there now, and add at least Canada so he can say the trip was international in scope), replete with Barry White music.

I last drove AP1 about 8 months ago. I prefer AP2 over AP1 of 8 months ago.

 

[WannabeOwner]

I've been disinterested in the AP1 vs. AP2 discussion partly because we aren't going back

I'm on AP1, I have no current plans to chop in the car for a new one, but I am interested in the AP1 / AP2 status:

Folk joke about Elon-time; Tesla achievements are both impressive on the one hand and late on the other.

But there will come a time when Fan Boy Support decreases to the point of Getting the Lawyers In. I think that's a state that would be an unmitigated disaster. I need Musk to succeed, in order to shake up the industry. I need his arch enemies to not find a way to bump him off, and I need him to not get sued for over-promise, under-deliver.

So at some point on his journey I need him to change to making realistic promises (well, on just the technologies that have matured to mainstream).

From the Fan Boy sidelines I find the over-promise on AP2 embarrassing. That aspect of Tesla is now a part of my sales-pitch to interested parties. I am sure that has significantly delayed the take-up amongst my chums, who were previously bought-in by my initial enthusiasm and the "launch demo" gave them

 

[BLKTSLA]

Great job tracking the progress of AP2 guys/gals...I think this will be helpful in getting a true sense of when AP2 is actually better and we start seeing the "+" features such as Autosteer+ and Summon+. However, for those that are comparing AP1 and AP2 from loaner experiences or when they used to own AP1, keep in mind that AP1 is also evolving with AP2 with each software update. The updates to AP1 are not going to be as extreme as AP2 but its evolving. One of Tesla's key challenges in all of this according to an insider I met, is that they need to create a unified code-base that works with both AP1 and AP2 hardware so they can push mass update packages that addresses all cars instead of updates for AP2 and AP1, and AP2.5 like they have been.

So in order to truly compare the 2 you need to really need to have access both AP1 and AP2 (same firmware) and can drive them interchangeable on the regular or at least on the same stretch of roads for a good comparison. If you are not doing that its really hard to say how much better or worse one is over the other. The recent snow on the East Coast and having to manually apply wipers was very frustrating for an AP2 car so as @AnxietyRanger states Auto-wipers would be a great leap forward but Vehicle ID would be even better!

 

[croman]

Dear Santa,
Vehicle ID and then local auto lane change please.

 

[Az_Rael]

Let's talk when we have speed sign recognition, vehicle identification display and auto-wipers in AP2... 'mmkay?

Speed Limit sign reading +1000. About 90% of commute can be completed with AP1 engaged, and only 50% with AP2 engaged because 20 miles of a 65mph road I drive on isn't in the database and AP2 limits to 45mph.

 

[jimmy_d]

I'm getting the strong sense that opinions about the relative merits of these two AP systems are dominated by people's individual use cases and pet peeves. Personally, I don't care about wipers or whether AS performs well in situations that it's not recommended for, I just want it to make road trips and bumper to bumper traffic more pleasant. I don't have a commute and I can drive myself to the grocery store. Other folks have a daily commute with some weird problem and that is really important to them.

And of course others are fixated on something that pissed them off one time and they aren't ever going to get over it. Welcome to the internet.

But as for the bread and butter of AS and TACC working well in situations that are recommended it does seems like 48 might be the crossover version.

If I were to go back to the AP1 car and drive it these are my working hypotheses:

I put 25k on AP1 before I swapped over and I still get AP1 rides in a friend's AP1 car. I drive AP2 constantly. Some of these I can predict for you.

1) I predict it would confirm that speed limit sign recognition and vehicle identification in surrounding lanes are the MOST important things to me for AP2 improvements.

I don't doubt that there are errors in there and that for some people an error on their regular route is really annoying, but those complaints are more heat than light.

2) I predict that AP1 will be inferior to AP2 in rain and low light conditions currently

This is all about lane markers. If they are good enough AP1 is great, but generally AP2 is better.

3) I predict that AP1 autolane change will impress me only if it can recognizes if someone is in my blind spot and behaves accordingly.... if it can also avoid having to try multiple times when switching multiple lanes, that would be great.

Lane changing in both of them seems to be related to how well they can see the adjacent lane - which once again is pretty good for AP1 if the far lane marking is good enough, but otherwise AP2 wins. AP2 seems to notice marginal cars in nearby blind spots better than AP1, which is probably just because the ultrasonics are a bit better. Practically speaking there's almost no difference.

4) I predict AP1 will be inferior on tight lane and tight curbs... something that I distinctly remember sucking on AP1... though I'm sure that's changed in some respects.

Later versions of AP1 are really good on tight turns if the lane markers are good and then road allows for a clear view around the curve. AP2 as of .42 was not as good (I have lots of recent data on that) but it was improving pretty quickly. Haven't been able to try 48 in that case yet.

5) I predict the user experience of AP1 will exceed AP2 at this time.... and perhaps that should be the measure of parity. BUT for me it isn't... instead I try to judge AP2 on complex tasks rather than reliability compared to AP1... that's what my videos are trying to elucidate because that's what I'm always trying to push autopilot to handle. It is this area that there is the most room to criticize my view on AP2... it's a different metric and one that is more likely to disappoint if they come up short.

Yes. There are a lot of little details that are still better in AP1 so the experience is more predictable over all (as of 42). But AP2 just crushes AP1 when the lane marking situation gets weird or ambiguous. Your videos seem like they are happening in that zone where AP2 does better. But of course that's not really how it's meant to be used so I'm a bit conflicted about how much weight should be applied there. For you it seems to have some importance, but to me it seems like that use case is never going to be comfortable until we get FSD.

6) The thing I currently imagine myself to disagree with @TaoJones about on this subject is stop and go traffic. I love AP2 in this regard... it's practically superb in my experience. Part of this is that I hate using TACC because I still have to do the work and I find it's prone to the same errors... but perhaps AP1 would make me see the light in this respect.

I'm with you on loving AS in stop and go, but I use TACC by itself a lot and there are definitely places where it's a huge win. Without AS I cannot drive stop and go - it just drives me insane. OTOH I've seen some of the failures that @TaoJones cites and I imagine that if I had to deal with them every day they would get old fast.

I guess my point is this: I knew AP2 was going to be a work in progress when I bought it, but that was one of the selling points...that they have a platoon of AP soldiers working on this constantly. Even with the slower pace that some of us expected, how can you not be excited about that?

Also Consider this... Volvo called it quits today on their autonomous driving program because they want to take time to figure out how to have "the safest car on the road"....

I'm totally with you there. I think it's a really hard problem and that, while we're going to see huge improvements in the majority of real world use cases it's going to be a long, long time before anyone fields a system that can't be criticized by the pugnacious, or the litigious.

 

[SlicedBr3ad]

It's common knowledge in terms of the number of people... the round table discussion on AI did wonders for my faith as well on autopilot... It's a big group of folks working on it, I know that. I wish I had an insider though.

I recall tesla tripled fsd team. I expect great things will be passed down by fsd team to eap team. Don't hold your breath though as we're two years away apparently.

 

[buttershrimp]

Great job tracking the progress of AP2 guys/gals...I think this will be helpful in getting a true sense of when AP2 is actually better and we start seeing the "+" features such as Autosteer+ and Summon+. However, for those that are comparing AP1 and AP2 from loaner experiences or when they used to own AP1, keep in mind that AP1 is also evolving with AP2 with each software update. The updates to AP1 are not going to be as extreme as AP2 but its evolving. One of Tesla's key challenges in all of this according to an insider I met, is that they need to create a unified code-base that works with both AP1 and AP2 hardware so they can push mass update packages that addresses all cars instead of updates for AP2 and AP1, and AP2.5 like they have been.

So in order to truly compare the 2 you need to really need to have access both AP1 and AP2 (same firmware) and can drive them interchangeable on the regular or at least on the same stretch of roads for a good comparison. If you are not doing that its really hard to say how much better or worse one is over the other. The recent snow on the East Coast and having to manually apply wipers was very frustrating for an AP2 car so as @AnxietyRanger states Auto-wipers would be a great leap forward but Vehicle ID would be even better!

He's far too modest since he didn't paste a link, so I'll do it! @BLKTSLA does great videos that strike the balance of informative and entertaining....

 

[AnxietyRanger]

One more thing AP2 lacks that we keep forgetting:

The one thing that is missing from ap2(but exists on ap1) is the ability to initiate an auto lane change on roads that aren’t freeways/interstates. Much of my driving is on roads with 2-3 lanes in each direction and not having this feature drastically reduces the usability of autopilot. Not sure why more people don’t mention this fact more often. To me, it’s a much more important feature than auto wipers, speed limit reading, or identifying trucks/cars/motorcycles.

 

[AnxietyRanger]

So basically the notion that AP2 has reached or surpassed AP1 relies on:

a) Forgetting that:

- AP2 does not have speed sign recognition, AP1 does
- AP2 does not have non-freeway lane changing, AP1 does
- AP2 does not have multi-type vehicle ID display, AP1 does
- AP2 does not have multi-lane vehicle ID display when not changing lanes, AP1 does
- AP2 does not have multi-lane vehicle ID display on non-freeways at all, AP1 does
- AP2 does not have auto-wipers (feature relegated to AP2 in AP2 cars), AP1 cars do

b) Relying on the notion that AP2 neural network is different from MobilEye and thus can offer different (in reality better or worse) auto-steering results that would reach or surpass AP1 in totality.

I find this hypothesis hard to support, objectively. The feature lack alone is significant in AP2, but the reporting on AP1 vs. AP2 lane-keeping is still too much all over the place to even safely conclude AP2 is as good as AP1 in totality, let alone that it would surpass AP1 so much that that difference could account for all the features it misses...

I'm not saying it is impossible, were someone to conduct a study on it, but it seems very, very far-fetched to me. Also, some people make a good point: How much of the improved lane-keeping actually is available also on the latest AP1 updates, is not very well known.

I think the safe bet still is AP2 remains behind AP1.

 

[avoigt]

Still like it! WIP for sure but thats no news....

 

[Phrixotrichus]

Still like it! WIP for sure but thats no news....

Do you really have to shake the steering wheel for the system to register that you`re still alive? No touch/squeeze sensors? And what about the interior camera?

 

[avoigt]

Do you really have to shake the steering wheel for the system to register that you`re still alive? No touch/squeeze sensors? And what about the interior camera?

yes you do, safety first.

However I agree to your point that it gets a hassle if the system works just perfect like in the video. But even then to shut down the people calling all fraud, fake and AP is killing people you have to implement this kind of measures. In the future that may change.

 

[Yinn]

People who haven't had AP1 are awfully quick to pronounce AP2 at parity or better.

This.

Sincerely, ex-AP1 owner, current AP2 owner who misses his AP1.

Do you really have to shake the steering wheel for the system to register that you`re still alive? No touch/squeeze sensors? And what about the interior camera?

It's torque based so squeezing it won't do anything. The jiggle, or sideways tension is what registers as hands on the wheel. There weren't any interior cameras in earlier cars, I believe it was added in hardware AP 2.5, but not sure if it's programmed yet.

 

[Carl]

Indeed. People who haven't had AP1 are awfully quick to pronounce AP2 at parity or better.

You see the opposite as well: people who never had an AP1 car comparing AP2 with AP1 and complaining. I wonder how they do that. (note that I have no opinion on the subject as I never drove an AP2 car)

 

[daktari]

An autonomous driving system has multiple components I guess. Camera sensing, sensor fusion, driving algorithms and more. In my experience, AP2 is mostly lacking in the last 2 but maybe better in sensing (except speed signs).

The most puzzling is I would have thought that driving algorithm should be similar across the 2 generations, but AP2 still wobbles quite a bit and drift outwards in bends. It is hard of course to know wether this is because of poor lane recognition or the driving algorithm itself but I believe maybe the algorithm.

Sensor fusion, and I mean how the system switch from vision to radar lock on is also lacking in AP2. AP1 can drive a prolonged stretch with a blue car and no lane lines, AP2 will not. Why, is it because they rely more on vision based tracking?

Vision based sensing, ie recognizing the lane, is pretty similar I would say. Many times AP2 has impressed after 2017.38 which was a big step forward. But AP2 is severly lacking in some scenarios: now with snow obscuring the lanes. Since "silky smooth" back in may/june, my car does not recognize curbs either. Not sure how AP1 performs, but I do remember from my test drives last year that performance was pretty good?