fasteddie7
Active Member
So many ways I want to take that.... probably only one way to actually take it....
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Poll: Has AP2 18.10.4 surpassed AP1 in terms of performance?
- Thread starter ab26
- Start date
⚡️ELECTROMAN⚡️
Village Idiot
Isn’t AP1 still improving with updates? Is Mobileye making these improvements? At what point did AP2 reach parity with late 2016 AP1?
fasteddie7
Active Member
It sure is. The latest 10.4 for ap1 really made a difference. The updates are on Tesla and after what I've experienced in the last 2 weeks, I expect more cool stuff for both Autopilots to come
I currently own both. In the areas I drive, AP2 is at least equal to AP1 in terms of the driving algorithm. Recognition wise, I feel like AP2 has surpassed the Mobileye capabilities in AP1, especially in low confidence areas. My anecdotal experience, of course.
arubinst
Member
Why do you think AP2 does not show the cars on the adjacent lanes?
Identifying cars and trucks and bikes is a whole different problem, since the NN needs to be trained to differentiate them. But since auto lane changing displays the cars on the target lane, it is peculiar that this information is not shown all the time.
Identifying cars and trucks and bikes is a whole different problem, since the NN needs to be trained to differentiate them. But since auto lane changing displays the cars on the target lane, it is peculiar that this information is not shown all the time.
pilotSteve
Active Member
It’s my hope that Tesla is not displaying adjacent lane vehicle icons until the side cameras are doing “blind spot” imaging so ALL nearby vehicles, perhaps especially the blind spots, will be accurately shown.
Showing *some* vehicles is a cute trick, showing *all* is a safety enhancement that will be a game changer for Tesla.
I recently had an AP1 loaner but I had .6.1 on my AP2 car at that time. My immediate observations:
- At the time, AP1 still showed other cars but didn't act upon that information. It seemed like a fun animation more than something useful. It seems like .10.4 owners see more cases of AP1 offsetting itself for overlapping cars, but I did not see that behavior in the previous update.
- AP1 drove very smoothly. "confident" is one word I'd use to describe it, but that didn't mean it drove more correctly either. It was also prone to departing lanes or completely not respecting lane markers when engaged in a challenging scenario.
- AP1 really loves to follow the car in front, while AP2 more or less doesn't care. So, if you're following another car, AP1's "monkey see monkey do" approach can feel like it's doing a great job of driving when it's really just parroting what the driver in front of you is doing. And I don't like the idea that I'm only as safe as some unknown person in front of me.
- Wow, AP1 really surprised me in terms of unavailability on many common roads that I take with suboptimal lane markings. It simply wouldn't offer itself at all, and I've taken for granted that AP2 allows engaging on those roads every single day. This is especially the case on commercial roads with just a central divider and a curb. AP1 really doesn't like offering AutoSteer unless conditions are pretty decent and the kind of situation where intuitively you'd expect AP to work. Meanwhile with AP2, I often engage it on roads I KNOW are kinda borderline, just to see how it'd do. This could get an inexperienced driver in trouble too.
Overall I still think both are really good systems. I still strongly prefer AP2. I'm okay with its style of driving and see a lot more potential in its sensor suite. Before it was something that needed a leap of faith, but especially as of .10.4, AP2 fills me with optimism.
- At the time, AP1 still showed other cars but didn't act upon that information. It seemed like a fun animation more than something useful. It seems like .10.4 owners see more cases of AP1 offsetting itself for overlapping cars, but I did not see that behavior in the previous update.
- AP1 drove very smoothly. "confident" is one word I'd use to describe it, but that didn't mean it drove more correctly either. It was also prone to departing lanes or completely not respecting lane markers when engaged in a challenging scenario.
- AP1 really loves to follow the car in front, while AP2 more or less doesn't care. So, if you're following another car, AP1's "monkey see monkey do" approach can feel like it's doing a great job of driving when it's really just parroting what the driver in front of you is doing. And I don't like the idea that I'm only as safe as some unknown person in front of me.
- Wow, AP1 really surprised me in terms of unavailability on many common roads that I take with suboptimal lane markings. It simply wouldn't offer itself at all, and I've taken for granted that AP2 allows engaging on those roads every single day. This is especially the case on commercial roads with just a central divider and a curb. AP1 really doesn't like offering AutoSteer unless conditions are pretty decent and the kind of situation where intuitively you'd expect AP to work. Meanwhile with AP2, I often engage it on roads I KNOW are kinda borderline, just to see how it'd do. This could get an inexperienced driver in trouble too.
Overall I still think both are really good systems. I still strongly prefer AP2. I'm okay with its style of driving and see a lot more potential in its sensor suite. Before it was something that needed a leap of faith, but especially as of .10.4, AP2 fills me with optimism.
WannabeOwner
Well-Known Member
Around me (rural roads in UK) we have a few roads with both central and verge lines, plenty with only central line, some of which have a curb and some a verge. UK roads probably narrower, on average?, and for me AP1 (i.e. on roads where it will engage) has been fairly hopeless with Centre-Line+Curb, and mostly-useless with Centre-Line+Verge. But with both Centre-Line and Curb/Verge-Line works well. (Highway is fine of course)
My thought is:
AP2 needs to have knowledge of vehicles in lane / other lanes in order to drive. Its obviously doing that part OK
Being able to convert that information into dashboard pictures is secondary, with software development priority to be given to "better driving". Also, releasing dashboard pictures which are unreliable, for any reason, would be a huge turnoff for Public - i.e. where AP2 is driving properly, but portraying the information wrongly. So I think Tesla is doing "better to display nothing until it is perfect".
I think the way that AP2 shows the car in front hopping from centre to side-of-lane, and back again, when clearly on dashcam [YouTube's that I have seen] the car in front is rock-solid centre-of-lane, is alarming. Clearly the presentation on dashboard is not the same as actual! and I think AP2 MUST have better Actual in order to be able to drive / avoid etc, which also leads me to believe they are not display any additional images until it works a lot better.
I hate to rehash an old thread, but I'm really curious to see where people stand on this issue now that 5 more months have passed.
I've been on AP1 with my 2015 P90D since it was first released and have grown extremely comfortable with it. Using it has become muscle memory and I've grown to know exactly where/when I can trust it and what it's limitations are. I rely on it everyday for my commute. Obviously, it's improved leaps and bounds since it was first released, though I agree with others who've complained about the frequency of nags.
My lease is now up and I'm about to switch out to a AP2 car next week. To be completely honest I really wish I could keep my P90D for at least another year or two...it's in excellent condition (55k miles later) and still feels like the day I got it. I made the mistake of leasing as I was paranoid about where things would go tech wise and FUD over premature obsolescence, but it was the choice I made so....my time with the car is up.
I'm quite nervous over the adjustment from AP1 to AP2 and having to relearn the system all over again. There are so many conflicting reports out there with some people saying AP1 is "clearly" superior and others saying the complete opposite...it leads me to believe that each one has it's own unique strengths which different drivers are exposed to differently. Can anyone here offer any more clarity on where the two stand now or any specific situations one should watch out for with AP2 coming from AP1?
I've been on AP1 with my 2015 P90D since it was first released and have grown extremely comfortable with it. Using it has become muscle memory and I've grown to know exactly where/when I can trust it and what it's limitations are. I rely on it everyday for my commute. Obviously, it's improved leaps and bounds since it was first released, though I agree with others who've complained about the frequency of nags.
My lease is now up and I'm about to switch out to a AP2 car next week. To be completely honest I really wish I could keep my P90D for at least another year or two...it's in excellent condition (55k miles later) and still feels like the day I got it. I made the mistake of leasing as I was paranoid about where things would go tech wise and FUD over premature obsolescence, but it was the choice I made so....my time with the car is up.
I'm quite nervous over the adjustment from AP1 to AP2 and having to relearn the system all over again. There are so many conflicting reports out there with some people saying AP1 is "clearly" superior and others saying the complete opposite...it leads me to believe that each one has it's own unique strengths which different drivers are exposed to differently. Can anyone here offer any more clarity on where the two stand now or any specific situations one should watch out for with AP2 coming from AP1?
fasteddie7
Active Member
I
I'm making a video on it now ap1 and 2 side by side. It's looking more and more like ap2 is finally the winner though.
TaoJones
Beyond Driven
You sure you wouldn't rather discuss religion or politics instead?
So after just driving another few thousand miles last week, I can say with continued certainty that AP2 remains not as good as AP1 in practical use.
Example 1: The continued broken speed limit data which directly affects AS in AP2 and one's, go figure, speed over distance.
Scenario a) Having to drive hundreds of miles through Wyoming and Utah at the speed limit because the speed limit data is lower than the posted limit. See Rawlins to Lusk, or Wheatland to Rawlins. For those somehow yet unaware, AP1 (Mobileye) uses a camera to read the speed limit signs. This works 99%+ of the time.
Scenario b) Having AP2 set speed through a school zone at 35mph, when in fact it is 25mph. Or at 35 when the limit is 45. This happened in SD, UT, and still happens in CA all over the place.
That this is still so bad for so long is beyond disappointing. It affects safety margin. That said, it is also fair to say that neither AP1 nor AP2 are bulletproof in this regard, and that an AP3 hybrid solution relying upon the former with the latter as a backup (presuming the thousands of errors are fixed) could be really good. In the meantime, A for effort. D for execution. F for maintenance.
Example 2: Phantom braking. That's braking - not breaking. So this still happens as it has since Day 1 with AP2. And I'm surprised I haven't been rear-ended yet as a result. Doesn't happen with AP1. Unnecessary braking? Sure. Phantom? No. This is a flat out F.
Example 3: An unnatural lust for center medians. As of *.28*, the car with AP2 will still head right for the (left of the, actually) center median past an intersection. Didn't have that with AP1.
Example 4: Crossing lane markers. More frequent with AP2 than AP1, especially at speed.
Now, if you ignore all of the above and watch video of the car managing curves over ascending and descending elevation, when it's not crossing lane markers or phantom braking or at the wrong speed limit for hundreds of miles at a stretch, AP2 does not suck quite as badly anymore. As one would expect after 2+ years post-launch.
The best way I can put it is that AP2/AP2.5 is another step along a long journey to get to AP3, which I'll define as based upon the new SoC/board and one could argue AP4 (or 3.5) once the S/X refresh comes along. It will be quite something to have traffic signal reaction that works from any lane.
Those who wait until next year's SoC, thus moving from AP1 to AP3, will not be as disappointed as those who paid for (silent E)AP or FSD in, say, late 2016.
Optimistically, we'll get stop *sign* reaction with the hardware we have and a code update. After all, this functionality was on full display in the staged December 2016 video with Mobileye gear.
Of course, it would be nice if the above 4 deficiencies were addressed as well *polite cough* before the new shiny object arrives.
Onward and upward. My 4th Model S will be one of those refreshed cars in, hopefully, late 2021 - just having traffic signal management alone will be worth the price of the car. My 3rd Model S will be an AP1 S85 to get from now until then. I'm done with AP2 as soon as possible.
You are going to see the usual suspects come and tell you both ways.
I am in the camp that believes AP2 is better in the core capabilities - basic autosteer and TACC - in more situations than AP1
- to start with it centers amazingly well in straights and curves. I only have ever go 5 miles over on curves.
- it detects even faint lines far better than AP1. Bots no problem.
- gone are the days every small rolling ups and downs are a thrill ride with swerving all over the place that AP1 tends to do. AP2 gets you over the hill without any issues (for the most part) even on curvy roads.
I have never had any issues with speed limits in all the places I have been. I am sure there are areas where this is iffy.
One feature that I miss is the lane changing in local roads, that is not available in AP2
I am in the camp that believes AP2 is better in the core capabilities - basic autosteer and TACC - in more situations than AP1
- to start with it centers amazingly well in straights and curves. I only have ever go 5 miles over on curves.
- it detects even faint lines far better than AP1. Bots no problem.
- gone are the days every small rolling ups and downs are a thrill ride with swerving all over the place that AP1 tends to do. AP2 gets you over the hill without any issues (for the most part) even on curvy roads.
I have never had any issues with speed limits in all the places I have been. I am sure there are areas where this is iffy.
One feature that I miss is the lane changing in local roads, that is not available in AP2
Are Tesla rolling out updates for AP1 that comes from MobileEye?
The ME tech is available in other cars and has some new features, like the ability to understand variable speed limit signs and improved lane following that anticipates corners more like a human driver. I don't know if the hardware needs to be newer for that or if Tesla could get those software updates from MobileEye.
The ME tech is available in other cars and has some new features, like the ability to understand variable speed limit signs and improved lane following that anticipates corners more like a human driver. I don't know if the hardware needs to be newer for that or if Tesla could get those software updates from MobileEye.
WannabeOwner
Well-Known Member
I had an AP2 loaner for 10 days recently whilst my car was in for service (yeah, service doesn't take 10 days but apparently getting something as simple as a boot/trunk-catch does)
I only use AP1 on highway / dual-carriageway with central reservation] - I do very little non-highway driving and what little I do is rarely on roads where AP would make much difference to me, either way. I also use AP on stop-go traffic (whatever the road type).
Out of curiosity I did do some testing in the Loaner of AP2 on my local roads; the roads that had central divide line and nearside verge line, AP2 drove pretty much for the whole length of my tests with no intervention (can say about "nags", I always have one hand on the wheel and consequently never get any nags).
I also drove on country lanes with central line but no verge-line (and grass verge rather than curb-stones). Only real negatives were that AP2 tended to move outwards where there were "driveways" or "gate entrances", and then the oncoming verge-bank was a surprise to it!! and AP2 was too slow to steer back into lane [so mostly I had to intervene], but overall "pretty good", but AP2s idea of where the various village signs were was a complete joke. mostly it sailed past the initial limit sign and then the dashboard sign came up quite a bit later ... I'd have a ticket by then if someone was checking that day ...)
Whilst AP1 reads the speed signs that isn't really the top-and-bottom of it, although WAY better than AP2. AP1 doesn't start slowing until it has seen, and passed, the speed sign, which is technically incorrect (and slowing from 60-70 MPH to 30 MPH happens gently, so uses up quite a few feet ...) and coming out of speed limit we normally have a "deregulated" sign, which means "national speed limit" and that depends on what type of road it is (only two choices though .. so "Not hard"). AP1 detects the delimit sign immediately, but doesn't change speed for quite some time afterwards (figuring out if it is an 60-type or a 70-type road somehow??). For following traffic that can be annoying, so I intervene ... AP2 had a poor idea where the limit was, but did start accelerating immediately as it knew what the delimited speed should be.
AP2 definitely much better on verge detection etc. and over-rise / speed-adjust for bend, but for me, using it on Highway, that would be irrelevant (well, not the "over rise" bit, but 90% of the highways I use aren't undulating like that, but AP1 has got better on that too), but the lack of speed sign reading would be a complete PITA for me, compounded (in UK) because we have variable speed limits (camera enforced) on busy highways
My AP1 car has been able to read those since day one, so not sure what an "improved" version from Mobile Eye would add.
As I mentioned other cars with it are able to read variable speed limits and adjust accordingly, so for example there is a wet and a dry limit on the same sign and depending on the rain sensor it selects the right one.
They can also read other types of sign, such as "no overtaking".
WannabeOwner
Well-Known Member
Stupidly [with benefit of hindsight !!] I didn't, no, I didn't have any highway journeys to make in that time interval. But my highway driving is so straightforward I reckon that AP1 is all I need, and AP2 would be at least as good, probably better. Except for the speed signs. Annoyed I didn't try a highway-test now ...
Ah, OK. We don't have that, just speed signs on light-boards for traffic flow. Mostly they look like speed signs (red circle with a number in it, but made up of individual bulbs, of course), but sometimes they are just a number on a black board.
I have seen those wet/dry ones on the continent (but I don't remember drivers obeying them in wet weather !!).
Lex
Member
I like the questions here about MobilEye updates.. piqued my curiosity. The system Tesla used was EyeQ3, check out this nifty Mobileye infographic:
https://www.mobileye.com/wp-content/uploads/2016/11/Mobileye_EyeQ_Infographic6-002.pdf
(did you know our AP1 MobilEye systems can identify stoplights and stop signs ?)
Based on my quick searches, no version history or firmware updates can be found on the web, so if there were updates, they were probably minor, and in closed distribution.
https://www.mobileye.com/wp-content/uploads/2016/11/Mobileye_EyeQ_Infographic6-002.pdf
(did you know our AP1 MobilEye systems can identify stoplights and stop signs ?)
Based on my quick searches, no version history or firmware updates can be found on the web, so if there were updates, they were probably minor, and in closed distribution.
Similar threads
- Poll
