AP2 / 8.1 Test Drive Perspective from AP1 Owner

[chillaban]

Ever since the 100D came out, I have been planning to upgrade from a 70D. The only reservation I've had is AP1 vs AP2's feature gap dramatically affecting my road trips. I eagerly arranged an AP2 test drive when 8.1 came out. I've had 17,000 miles in the last 8 months with AP1, and just finished around 150 miles of AP2 driving in a S P90D with firmware 17.11.10 (the car had 17.11.3 but the service center applied the update while we were doing paperwork)

Overall TLDR Parity Rating (compared to AP1, 8.0 17.4.14):

TACC: 95% parity
Highway Autosteer: 90% parity, even glimmers of superiority in certain situations
Local/Surface Street Autosteer: Less than 25% parity


Extended Notes:


TACC:

• For the most part, this behaves pretty similarly to AP1, only two comments here:
• Had 1 of ~30 overpasses result in an abrupt slowdown/hesitation. The car fairly quickly (I'd say 50% brake application) slowed 10mph from 75mph and then sped back up. This phenomenon has been documented before. Contrary to the opinion of some, I maintain this is not dangerous though highly annoying and startling. While the acceleration feels sudden because it's unexpected, overall it did not change the speed of the car enough to present a rear-ending hazard. To compare, I found a clear section of freeway at 75mph and jabbed the brake pedal as quickly as I could. The car slowed to 40mph in the blink of an eye, with a lot of ABS sounds.
• Had one or two situations in the city where TACC inappropriately slowed down reacting to a car in the adjacent lane. As I got closer, TACC corrected itself.

Highway Autosteer:

• Overall, very very usable. I'd call it 90% parity. In fact if you've not experienced AP1, you might feel this is the best lane holding system on the market...
• Around sharp curves, AP2 does not seem to slow the way that AP1 proactively slows. It also loses holding and goes slightly outside the lane from time to time when navigating sharp curves. Definitely some work to be done here.
• Stop and go traffic seemed to be handled quite well in terms of brake/acceleration ramping.
• Cut-ins were not handled as well as AP1. AP1 is a lot smoother and reactive when a slower car suddenly cuts into your lane. AP2 takes a little while to register, and then has to brake a bit more proactively.
• I had 1 case where AP2 wanted to follow an exit fork. It was sort of a fluke -- I spent a lot of time in the right lane to torture-test AP2, it passed at least 15 exits without diving for them.
• I actually found that with winding + hilly roads, AP2 is much faster than AP1 at correcting trajectory right after cresting a hill... AP1 seems to have a half-second deer-in-the-headlights moment before making a steering adjustment when cresting a hill, which is absolutely awful for winding mountain roads. AP2 actually handles this situation better than AP1, reacting virtually instantly after cresting a hill.
• Tracking of worn-out lane markings is actually remarkably good!
• Auto Lane Change is comparable on AP1 vs AP2. The only bug I observed: The moment you cross the lane line, AP2 shows a pretty jerky lane change animation on the screen. If you don't keep the turn stalk engaged, AP2 is more prone to canceling the lane change because it believes for a split second that you are aborting.
• In general, though, AP2 does not inspire as much confidence as AP1. With AP1, after 17,000 miles, I actually feel mildly comfortable to take my eyes off the road for a few seconds to mess with my navigation system. I don't feel as confident right now with AP2 -- I've seen it do some questionable things where if there's any cars or curves around, my eyes are GLUED to the road.

City Autosteer:

• It's almost not worth talking about, other than to say it's virtually a joke right now. Don't expect to use Autosteer on surface streets. Now I know what you're yelling already: "But the manual has always said Autosteer is designed for highways, how dare you use it on city streets".... This is mostly for illustrative comparisons. AP1 actually Autosteers reasonably well on multi-lane city expressways, such that I feel comfortable using it as a daily part of my commute albeit with more frequent overrides compared to highways.
• With AP2, the car goes maybe 15 seconds between attempting to kill me.
• 90% of the problem is when going through an intersection and the lane lines disappear, AP2 will draw wildly dancing lines and then jerk the steering wheel to follow one of those nonsensical lane lines. If Tesla just solves that problem, local auto steer would be a ton more useful.
• AP2 does not seem to want to go into "blue car following" mode. Even when it's coming up with completely BS dancing lane lines, it wants to trust those over the car in front.
• 35mph is not a very useful speed limit in my experience.
• There's one glimmer of hope: AP2 seems to recognize curbs, grass, etc as lane lines on the screen. AP1 has never managed to do that in my experience.

Other / Miscellaneous:

• The "improved" ultrasonics are, in my opinion, mostly marketing fluff. I saw almost no difference between AP1 and AP2 blind spot detection. It was able to recognize walls maybe from 1 foot further compared to AP1, but I can't imagine that makes a big difference. IMO improved blind spot detection will come when the side cameras are activated.
• The display of other cars on the road is awful compared to AP1. AP1 can continuously display up to 6 cars: 3 lanes, and 2 cars per lane. AP2 seems to only show 1 lead car in your lane and sometimes the car in front (less reliably than AP1), and it doesn't really show any cars in other lanes. Once in a while it shows a white car in a nearby lane, which is confusing: In AP1, white car means that TACC(or AP speed) is being limited from reaching the set point due to the distance to that car. But in AP2, white cars in adjacent lanes don't seem to have such a meaning: The car doesn't slow down for such cars.
• Also, randomly white cars blink onto the display without much meaning.
• The 1080p backup camera is a noticeable improvement over the 720p one. The screen is a lot sharper, and especially in low light there's fewer interlacing artifacts.

Overall, though, as an engineer in Silicon Valley, I respect the rate at which AP2 is progressing. Tesla has accomplished a pretty impressive feat, replicating the majority of the MobileEye EyeQ3 functionality on a completely new platform in a fraction of the time it took MobileEye to deliver that solution. I feel confident enough in AP2's continued improvements that this test drive has made me decide to go forward with upgrading my Tesla. But to be perfectly frank: AP2 is not at AP1 parity yet in any category. But at least with 8.1, it's genuinely useful on the freeways.

 

[thimel]

Great write up. Thank you.

 

[spentan]

Excellent Write-Up,

As a user of AP1 for about 30,000 miles over my S P85D and Signature X, to get AP2 last week in my new X90D, is an interesting change.

I can definitely feel the benefits, as you said, local driving, the kerbs, grass, are more distinctive and it kinda knows not to go into it. However there is some work to be done for sure.

I've actually found that Summon on AP2 is much better than AP1. I use it to back into my garage, and it is great. I found my AP1 used to steer itself too sharply into the corner of the garage, therefore halting, because of the cabinets on the side. AP2 is happy with the Tight Setting, and reverses confidently over the lip of the garage, straight in first time.

 

[IanR(forRover)]

Good to hear. In Brexit England we still await the renewal!

 

[Mark Z]

The details provided by the OP are excellent; a verification of the improvements made with version 8.1. There are times when AP2 is extremely accurate and sharp as a freeway "driver". However, the 35 mph city Autopilot tipsy "driver" still needs Drivers Ed.

I was very pleased with the latest AP2 lane change speed compared with AP1. No more waiting, it just does it.

 

[Nosken]

Great info by OP. I logged ~20,000 AP1 miles on my previous P85D. I have over 1,000 AP2 miles.
Your summary is very similar to my observations.
AP1 handled freeway curves much better, especially tight ones, but as more cameras get involved, it should be "no contest" later.
Just yesterday I was focused on the second car in front of me. The "differential radar" seems much better while moving than it was with AP1. A little "dancing around" of the second car, but it seemed show much more often while moving at speed.
On AP1 I liked knowing that it was picking up the other cars around me, in the other lanes. Not seeing them now, is less distracting, but it does not instill confidence that the computer sees those cars.
I've been able to initiate parallel park mode twice. It worked fine, but it seems to park at a slower speed. This is not a bad thing. Either park mode on AP1 was great, but the speed was nerve wracking. It may simply be my imagination, because I've gotten more used to it, but it does seem that it is slightly slower now. Not so much that you would get in trouble with oncoming traffic.
After all the incessant complaining in these forums, I think this is a very strong step in the right direction. When you look at the actual roll out, the most recent version took much less time than the original version. AP1 took 1 year to go active, initially it was much more crude than AP2 is. It wanted to take freeway off ramps often.
AP2 hardware rolled out only 6.5 months ago. I am encouraged by the progress, and look forward to more cameras becoming active.

 

[disagree]

But to be perfectly frank: AP2 is not at AP1 parity yet in any category. But at least with 8.1, it's genuinely useful on the freeways.

I don't think anyone would argue with that claim.

Where people differ is on whether this is acceptable for April 2017, especially for people who have had cars since December 2016, and whether in the places where AP2 doesn't meet AP1 there is significant hazard. Reading your list, which matches very closely my own feelings on 8.1, I have found situations in which AP2/8.1 is downright scary.

And I've got over 1000 miles on 8.1.

 

[drshan]

Ever since the 100D came out, I have been planning to upgrade from a 70D. The only reservation I've had is AP1 vs AP2's feature gap dramatically affecting my road trips. I eagerly arranged an AP2 test drive when 8.1 came out. I've had 17,000 miles in the last 8 months with AP1, and just finished around 150 miles of AP2 driving in a S P90D with firmware 17.11.10 (the car had 17.11.3 but the service center applied the update while we were doing paperwork)

Overall TLDR Parity Rating (compared to AP1, 8.0 17.4.14):

TACC: 95% parity
Highway Autosteer: 90% parity, even glimmers of superiority in certain situations
Local/Surface Street Autosteer: Less than 25% parity


Extended Notes:


TACC:

• For the most part, this behaves pretty similarly to AP1, only two comments here:
• Had 1 of ~30 overpasses result in an abrupt slowdown/hesitation. The car fairly quickly (I'd say 50% brake application) slowed 10mph from 75mph and then sped back up. This phenomenon has been documented before. Contrary to the opinion of some, I maintain this is not dangerous though highly annoying and startling. While the acceleration feels sudden because it's unexpected, overall it did not change the speed of the car enough to present a rear-ending hazard. To compare, I found a clear section of freeway at 75mph and jabbed the brake pedal as quickly as I could. The car slowed to 40mph in the blink of an eye, with a lot of ABS sounds.
• Had one or two situations in the city where TACC inappropriately slowed down reacting to a car in the adjacent lane. As I got closer, TACC corrected itself.

Highway Autosteer:

• Overall, very very usable. I'd call it 90% parity. In fact if you've not experienced AP1, you might feel this is the best lane holding system on the market...
• Around sharp curves, AP2 does not seem to slow the way that AP1 proactively slows. It also loses holding and goes slightly outside the lane from time to time when navigating sharp curves. Definitely some work to be done here.
• Stop and go traffic seemed to be handled quite well in terms of brake/acceleration ramping.
• Cut-ins were not handled as well as AP1. AP1 is a lot smoother and reactive when a slower car suddenly cuts into your lane. AP2 takes a little while to register, and then has to brake a bit more proactively.
• I had 1 case where AP2 wanted to follow an exit fork. It was sort of a fluke -- I spent a lot of time in the right lane to torture-test AP2, it passed at least 15 exits without diving for them.
• I actually found that with winding + hilly roads, AP2 is much faster than AP1 at correcting trajectory right after cresting a hill... AP1 seems to have a half-second deer-in-the-headlights moment before making a steering adjustment when cresting a hill, which is absolutely awful for winding mountain roads. AP2 actually handles this situation better than AP1, reacting virtually instantly after cresting a hill.
• Tracking of worn-out lane markings is actually remarkably good!
• Auto Lane Change is comparable on AP1 vs AP2. The only bug I observed: The moment you cross the lane line, AP2 shows a pretty jerky lane change animation on the screen. If you don't keep the turn stalk engaged, AP2 is more prone to canceling the lane change because it believes for a split second that you are aborting.
• In general, though, AP2 does not inspire as much confidence as AP1. With AP1, after 17,000 miles, I actually feel mildly comfortable to take my eyes off the road for a few seconds to mess with my navigation system. I don't feel as confident right now with AP2 -- I've seen it do some questionable things where if there's any cars or curves around, my eyes are GLUED to the road.

City Autosteer:

• It's almost not worth talking about, other than to say it's virtually a joke right now. Don't expect to use Autosteer on surface streets. Now I know what you're yelling already: "But the manual has always said Autosteer is designed for highways, how dare you use it on city streets".... This is mostly for illustrative comparisons. AP1 actually Autosteers reasonably well on multi-lane city expressways, such that I feel comfortable using it as a daily part of my commute albeit with more frequent overrides compared to highways.
• With AP2, the car goes maybe 15 seconds between attempting to kill me.
• 90% of the problem is when going through an intersection and the lane lines disappear, AP2 will draw wildly dancing lines and then jerk the steering wheel to follow one of those nonsensical lane lines. If Tesla just solves that problem, local auto steer would be a ton more useful.
• AP2 does not seem to want to go into "blue car following" mode. Even when it's coming up with completely BS dancing lane lines, it wants to trust those over the car in front.
• 35mph is not a very useful speed limit in my experience.
• There's one glimmer of hope: AP2 seems to recognize curbs, grass, etc as lane lines on the screen. AP1 has never managed to do that in my experience.

Other / Miscellaneous:

• The "improved" ultrasonics are, in my opinion, mostly marketing fluff. I saw almost no difference between AP1 and AP2 blind spot detection. It was able to recognize walls maybe from 1 foot further compared to AP1, but I can't imagine that makes a big difference. IMO improved blind spot detection will come when the side cameras are activated.
• The display of other cars on the road is awful compared to AP1. AP1 can continuously display up to 6 cars: 3 lanes, and 2 cars per lane. AP2 seems to only show 1 lead car in your lane and sometimes the car in front (less reliably than AP1), and it doesn't really show any cars in other lanes. Once in a while it shows a white car in a nearby lane, which is confusing: In AP1, white car means that TACC(or AP speed) is being limited from reaching the set point due to the distance to that car. But in AP2, white cars in adjacent lanes don't seem to have such a meaning: The car doesn't slow down for such cars.
• Also, randomly white cars blink onto the display without much meaning.
• The 1080p backup camera is a noticeable improvement over the 720p one. The screen is a lot sharper, and especially in low light there's fewer interlacing artifacts.

Overall, though, as an engineer in Silicon Valley, I respect the rate at which AP2 is progressing. Tesla has accomplished a pretty impressive feat, replicating the majority of the MobileEye EyeQ3 functionality on a completely new platform in a fraction of the time it took MobileEye to deliver that solution. I feel confident enough in AP2's continued improvements that this test drive has made me decide to go forward with upgrading my Tesla. But to be perfectly frank: AP2 is not at AP1 parity yet in any category. But at least with 8.1, it's genuinely useful on the freeways.

Very helpful thread. Thank you!

 

[OBX John]

Thanks for this - now my wife has taken over my MS 60, I was trying to decide between a CPO with AP1 or just going all out and getting a 100D. I think a new car will be better overall, but still tempted by the functionality and value of an AP1 car.

 

[chillaban]

I took my final spin this morning in stop and go traffic. Overall, I find that in heavy freeway traffic, the ~90% parity mark is about right.

I think sometimes, AP2 likes to initiate braking a little too late when approaching standstill traffic. It still slows down in time, but the braking is rougher than I would like.

In city though, be careful. It really does not want to recognize certain types of stopped cars. I was at a stoplight behind a Toyota Highlander and I engaged TACC for fun when there's no car shown in front, and it happily tried to lunge towards towards the Highlander before slamming on the brakes once it was in parking sensor range. Had that happened at 25+mph, it would've been an accident.


The only major incident I had on the freeway was that AP2 trusted a set of diagonal tire skid marks over the lane line, and drove 2 lanes over before using the lane lines again. There was no traffic nearby so I played along, but definitely keep a hand on the wheel when using AP2. It really can go from perfectly fine to doing something utterly nonsensical in a split second.

 

[lunitiks]

The 1080p backup camera is a noticeable improvement over the 720p one. The screen is a lot sharper, and especially in low light there's fewer interlacing artifacts.

What's you're source on this resolution upgrade on the BUC?

 

[chillaban]

What's you're source on this resolution upgrade on the BUC?

There were a few threads about it — and even a guy that was retrofitting it.

 

[Nosken]

The refresh models had a better camera even before HW2. It was distinctly higher resolution than my older 2015 AP1 P85D.

 

[chillaban]

The refresh models had a better camera even before HW2. It was distinctly higher resolution than my older 2015 AP1 P85D.

Just to add to that: my "old" Tesla is a Refresh model. I agree that it has a better camera, but the AP2-era camera is even better. I can read out license plates of the 3 cars immediately behind me at red lights, which I could not do before. It makes it a lot more pleasing to use as a rear view mirror replacement.

 

[bhzmark]

AP2 seems to recognize curbs, grass, etc as lane lines on the screen. AP1 has never managed to do that in my experience.

with winding + hilly roads, AP2 is much faster than AP1 at correcting trajectory right after cresting a hill... AP1 seems to have a half-second deer-in-the-headlights moment before making a steering adjustment when cresting a hill, which is absolutely awful for winding mountain roads. AP2 actually handles this situation better than AP1, reacting virtually instantly after cresting a

Summon on AP2 is much better than AP1.

Thanks. For my amusement I like calling out statements where people report AP2 is better than AP1.

 

[Electroman]

.. that was the best writeup on AP2 . Very informative. Thanks chillaban.

 

[Kant.Ing]

The OP is very informative, thanks!

Looks like pretending an AP1 parity has let Tesla engineers switch gears to update internet browser, kernel, and media app. Hooray!

 

[bob_p]

There should be separate development teams working on the console apps and AP.

Right now, Tesla placed a higher priority on quick releases of the AP2 updates - and adding in major console software changes would require integrating the AP2 & console software changes, which would require more testing before release.

Unfortunately, that probably means we won't see the promised 8.1 console software changes (media player fixes, browser, kernel update) until Tesla believes they've made enough progress with AP2 to allow them more time to add in the console changes.

 

[chillaban]

There should be separate development teams working on the console apps and AP.

Right now, Tesla placed a higher priority on quick releases of the AP2 updates - and adding in major console software changes would require integrating the AP2 & console software changes, which would require more testing before release.

Unfortunately, that probably means we won't see the promised 8.1 console software changes (media player fixes, browser, kernel update) until Tesla believes they've made enough progress with AP2 to allow them more time to add in the console changes.

My best guess is that the team working on the touchscreen/infotainment OS is busy working on the Model 3 UI. That's the more logical talent reuse versus trying to get UI engineers to work on self-driving systems.

 

[oktane]

AP2 does not inspire as much confidence as AP1. I've seen it do some questionable things where if there's any cars or curves around, my eyes are GLUED to the road.

City Autosteer: It's virtually a joke right now. Don't expect to use Autosteer on surface streets. With AP2, the car goes maybe 15 seconds between attempting to kill me.

The "improved" ultrasonics are, in my opinion, mostly marketing fluff. The display of other cars on the road is awful compared to AP1. AP2 is not at AP1 parity yet in any category.

Thank you for the very informative, unbiased post. I concur with your comments. In its current iteration, AP2 is "better than nothing" but still basically just a gimmick. Not even close to what I test drove (and rightfully expected would be delivered) with an AP1 car. Let's hope they keep improving rapidly.