-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
AP1 vs. AP2
- Thread starter aleagle
- Start date
-
- Tags
- Driving Dynamics Model S
AFAIK, AP1 uses rule based system and traditional machine learning. So, it can learn fast but only up to a point. AP2 uses deep learning. It learns slower and is more complicated but will be better than AP1 in the long run.
Elon said AP2 will reach parity in December 2016. That never happened.
Elon said AP2 will reach parity in December 2016. That never happened.
No, Elon said "enhanced autopilot" would be released December 2016, meaning exceeding parity. Never happened probably not for 1-2 years if ever.
Which is why Elon owes us all an apology and refund.
Kant.Ing
Member
The improved result over time and over data from these loosely programmed machine learning algorithms looks like kids learning curve to improve their grade. Only problem is their score will never reach 100 yet we feel satisfactory to lower traffic accident and fatality. They are playing with statistics and ML is not AI.
ML is old. We need something better otherwise FSD stays as a dream.
ML is old. We need something better otherwise FSD stays as a dream.
Last edited:
Any source for this ? Or more info on the distinction between the type of software design used for both?
I was under the impression that both used deep neural networks except that AP1 used Mobileye for visual object recognition (which itself used its own DNN) and Ap2 brought that vision object recognition in house , but it is all DNN, or at least the key parts for both are DNN.
Last informative post mentioning a FW version and some comparsion is from mars, any news here ? - I plan to buy one soon, and I feel confused by the broken promises/delay situation. (depending on who-feels-what)
But of course, paying more top get EAP if there's a endless wait until it catches up.
But of course, paying more top get EAP if there's a endless wait until it catches up.
Hmm? Plenty of us have been posting our reviews of 2017.28, the latest version, and there's a mix of positive and mixed reviews amongst there.
Some of us believe AP2 has reached highway parity if not more compared to AP1. Basically all of us agree that local driving is still not close. But others are still seeing some fundamental glitches with AP2 like driving over lane lines, etc, that most don't see at all. I believe there to be a hardware/alignment/calibration component there.
Overall, it's steadily coming along. There's visible progress month over month. It's just nowhere near where we were told it would be as of last October.
Probably best to test drive at the dealer...and report back here. The situation is changing week-week, but buying while expecting more than the current capability has historically led to disappointment.
Regardless AP1 or AP2. For me, all I am using it for is freeway during stop/go traffic and long distance driving. I don't trust any AP at this moment in time or in the next 2-3 years to utilize full autonomy if AP2 does come to that for several reasons
1. It's first generation full autonomy, don't trust first generation anything 100%
2. There will be still significant of non AP drivers, human errors will always give you the unexpected
3. Using it in Surface street to me is still not solid, going back to #1. Something as the traffic lights. Across US there at least 1/2 dozen or more variation of traffic light position of hanging, locations of hanging, size of light, style of lights, etc. too many variations to programs against just in traffic lights alone, not to mention stop sign, walk sign, people holding stop sign, etc. to trust full autonomy on surface street.
4. We don't have proper regulations and oversight on it at the moment. There should be some common standards that all manufacturers should adhere to for full autonomy. Also shared technology so each car can talk to each other to help the full autonomy be better in prevention measures.
So at this point, AP1 or AP2 mean little usable meaningful differences to me. AP3 I might rethink the whole spectrum again.
#
1. It's first generation full autonomy, don't trust first generation anything 100%
2. There will be still significant of non AP drivers, human errors will always give you the unexpected
3. Using it in Surface street to me is still not solid, going back to #1. Something as the traffic lights. Across US there at least 1/2 dozen or more variation of traffic light position of hanging, locations of hanging, size of light, style of lights, etc. too many variations to programs against just in traffic lights alone, not to mention stop sign, walk sign, people holding stop sign, etc. to trust full autonomy on surface street.
4. We don't have proper regulations and oversight on it at the moment. There should be some common standards that all manufacturers should adhere to for full autonomy. Also shared technology so each car can talk to each other to help the full autonomy be better in prevention measures.
So at this point, AP1 or AP2 mean little usable meaningful differences to me. AP3 I might rethink the whole spectrum again.
#
Sorry to bring an old topic back to life, but I just tried AP1 on a long road trip and also AP2. AP1 perform beautifully. AP2 still struggle to change lane and to see other cars on the side. The movement on freeway for AP2 is also more jerky compared to AP1.
I am confident AP2 still need to catch up to AP1.
I am confident AP2 still need to catch up to AP1.
Agree. I was behind an AP1 on the 405 this morning. I stayed behind him for about 70 miles. Before exiting the freeway, I noticed that he was on AP. His car was so smooth vs. my AP2. I could especially feel that when the sun hit at it's early morning low altitude. AP2 is literally blinded by the sun.
Much work is needed. I really need to hear an acknowledgement and an action plan from Tesla
How'd you know they were on AP? It's not that I don't believe you, I'm just wondering what your criteria was for judging that. If I see a Model S/X consistently centered in the lane, as opposed to biased towards the left side of the lane (assuming a LHD car of course), I'd guess they probably have autosteer enabled.
Bruce.
Not to mention most of the times when AP2 is "not smooth" there is no way you can see that without being a person in the cabin. Small steering adjustments and even jerky steering adjustments / throttle adjustments are not easily visible by following another car (unless they outright slam on the brakes while you're tailgating them)
It was pretty evident; I used the lane markings and the tar fillings as references.
After the last update 2017.32 anyone notice on AP1, there some issue with lane recognition? When in a traffic condition and you have a vehicle close in front of you, AP1 seem to have trouble recognizing the lane and the blue markers disappeared.
I drive the same route to and front work daily and have been using AP1 on these stop and go traffic condition. Starting to notice this lane disappearing issue occurring more frequent with 2017.32 patch.
I drive the same route to and front work daily and have been using AP1 on these stop and go traffic condition. Starting to notice this lane disappearing issue occurring more frequent with 2017.32 patch.
[Cross posted from another topic. This is probably a better place to post this. Again these are my experiences from a limited 1 day test drive. Not discounting those that have issues]
All right - finally I had an opportunity to check out AP2 on an X75D through an overnight test drive. I drove around 200+ miles most of it using AP. Went through the same route that I take every day on my commute and some more. Most of my driving was on the highway and a few on the local roads where I am usually comfortable driving with AP.
Now, I have driven close to 30k miles on AP1 in my S85 and I am very familiar with AP1. Armed with that knowledge and with a dose of basic common sense - which some seem to lack here - I turned on AP2 with a lot of apprehension. My first 30 odd miles were a bit nervous watching the road and the dashboard like a hawk. Most of that nervousness was due to the excessive time I spend here reading the experiences of folks who have driven AP2 and how it tried to kill them. But to my disappointment, it never tried to kill me.
- It drove pretty much like AP1 on the highways. I simply can't tell if I am driving AP1 or AP2 just by driving on the highway.
- it centered the lane quite nicely for the most part even in sweeping turns. Agreed I didn't have any sharp 90 degree turns on the freeways I drove.
- Never experienced any ping-pongy stuff - as long as the lanes are clearly visible. Just rock solid like driving on tracks even on sweeping curves. Take a look at the video below.
- Did not experience any slow down due to over passes, but then I admit the overpasses I went under may not have been challenging enough. BUT I did notice a slow down once on a curve with traffic going much slower on my adjacent lane, and it got confused. Never had that happen with AP1.
- was able to drive for much longer distance and for more time without being harassed to 'hold the wheel' This was especially surprising in construction areas, whereas in AP1 it wouldn't let me drive without holding the wheel for more than 15 seconds. It did not even ask me once to hold wheel the entire 3 mile construction zone I drove (121/Airport), whereas in AP1 I would have had alteast 5 warnings on that same stretch.
- It seems to detect cars in front of you from quite a distance and slows down better than AP1, but the final stop seems abrupt. I can understand how it can leave an uneasy feeling. Starting from a stop is also painfully slow that it leaves a large gap in front before it catches up. I often end up pressing 'gas' to keep the gap as little as possible.
- Changing lanes - no different from AP1. Sometimes no hesitation at all. Sometimes a lot of hesitation and it gets annoying.
- The lack of AP cancel push button switch is a pain.
- Once driving at around 60, someone cut in front about just a feet away and it never detected although his tires were mostly in my lane. I had to press the brakes myself and take control. I have seen AP1 do this too.
- The lack of cars on the display in adjacent lanes is a clear lack of parity with AP1.
- few times the display car on the dash was riding right on the lane lines or very close to it, but the car itself was much closer to the center than the display leads you to believe. My video below has an example of that.
- on backroads though it was a mixed experience. It seems to detect faint lines much better than AP1, which is a good thing. This means I was able to use AP2 on areas I cannot use AP1. But I guess it was also the source of a few heart stopping moments, because it was detecting multiple spurious lanes and was getting confused. It detects crack as lanes. Detects shoulders as lanes. Detects tar lines as lanes. So sometimes it ends up taking a path which is actually not a lane. They need to calibrate the lane detecting algo to be not so sensitive. This is a clear negative compared to AP1
- Lanes on the display bounces often which is unsettling. But the good thing is, it still drives pretty steady and neat. So I am guessing it is just a display thing.
Bottom line: Love my AP1, perhaps the most important reason why I bought a Tesla and still love driving it. I was initially skeptical if I get an X then perhaps I am better off looking for an AP1 (inventory or CPO). Now I have zero hesitation switching to AP2. It does everything quite well, in what I like and use most in AP1.
Of course your usage patterns in your current AP1 might be different. Your expectations on what an Autopilot should do might be different. Your ability to use common sense when using a driver assistance like AP might be different. But as far as I can see AP2 is pretty close to AP1 parity. Pluses and minuses are insignificant.
All right - finally I had an opportunity to check out AP2 on an X75D through an overnight test drive. I drove around 200+ miles most of it using AP. Went through the same route that I take every day on my commute and some more. Most of my driving was on the highway and a few on the local roads where I am usually comfortable driving with AP.
Now, I have driven close to 30k miles on AP1 in my S85 and I am very familiar with AP1. Armed with that knowledge and with a dose of basic common sense - which some seem to lack here - I turned on AP2 with a lot of apprehension. My first 30 odd miles were a bit nervous watching the road and the dashboard like a hawk. Most of that nervousness was due to the excessive time I spend here reading the experiences of folks who have driven AP2 and how it tried to kill them. But to my disappointment, it never tried to kill me.
- It drove pretty much like AP1 on the highways. I simply can't tell if I am driving AP1 or AP2 just by driving on the highway.
- it centered the lane quite nicely for the most part even in sweeping turns. Agreed I didn't have any sharp 90 degree turns on the freeways I drove.
- Never experienced any ping-pongy stuff - as long as the lanes are clearly visible. Just rock solid like driving on tracks even on sweeping curves. Take a look at the video below.
- Did not experience any slow down due to over passes, but then I admit the overpasses I went under may not have been challenging enough. BUT I did notice a slow down once on a curve with traffic going much slower on my adjacent lane, and it got confused. Never had that happen with AP1.
- was able to drive for much longer distance and for more time without being harassed to 'hold the wheel' This was especially surprising in construction areas, whereas in AP1 it wouldn't let me drive without holding the wheel for more than 15 seconds. It did not even ask me once to hold wheel the entire 3 mile construction zone I drove (121/Airport), whereas in AP1 I would have had alteast 5 warnings on that same stretch.
- It seems to detect cars in front of you from quite a distance and slows down better than AP1, but the final stop seems abrupt. I can understand how it can leave an uneasy feeling. Starting from a stop is also painfully slow that it leaves a large gap in front before it catches up. I often end up pressing 'gas' to keep the gap as little as possible.
- Changing lanes - no different from AP1. Sometimes no hesitation at all. Sometimes a lot of hesitation and it gets annoying.
- The lack of AP cancel push button switch is a pain.
- Once driving at around 60, someone cut in front about just a feet away and it never detected although his tires were mostly in my lane. I had to press the brakes myself and take control. I have seen AP1 do this too.
- The lack of cars on the display in adjacent lanes is a clear lack of parity with AP1.
- few times the display car on the dash was riding right on the lane lines or very close to it, but the car itself was much closer to the center than the display leads you to believe. My video below has an example of that.
- on backroads though it was a mixed experience. It seems to detect faint lines much better than AP1, which is a good thing. This means I was able to use AP2 on areas I cannot use AP1. But I guess it was also the source of a few heart stopping moments, because it was detecting multiple spurious lanes and was getting confused. It detects crack as lanes. Detects shoulders as lanes. Detects tar lines as lanes. So sometimes it ends up taking a path which is actually not a lane. They need to calibrate the lane detecting algo to be not so sensitive. This is a clear negative compared to AP1
- Lanes on the display bounces often which is unsettling. But the good thing is, it still drives pretty steady and neat. So I am guessing it is just a display thing.
Bottom line: Love my AP1, perhaps the most important reason why I bought a Tesla and still love driving it. I was initially skeptical if I get an X then perhaps I am better off looking for an AP1 (inventory or CPO). Now I have zero hesitation switching to AP2. It does everything quite well, in what I like and use most in AP1.
Of course your usage patterns in your current AP1 might be different. Your expectations on what an Autopilot should do might be different. Your ability to use common sense when using a driver assistance like AP might be different. But as far as I can see AP2 is pretty close to AP1 parity. Pluses and minuses are insignificant.
Sawyer8888
Member
Very good assessment of AP2. I could not agree more with your point about the painfully slow starts. It annoys me and makes AP2 in stop and go traffic less desirable.
The two systems are closer, but I still prefer the AP1 in the loaners I have driven. There are still many more pluses in the AP1 column for me.
