this sounds scarily like AP1...hmmm now i'm getting more excited about AP2
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
2017.38 f87c64d5
- Thread starter Peteybabes
- Start date
-
- Tags
- 2017.38 f87c64d5 Model X
RE post #34 and post #39, 2017.38 f87c64d5,
Third day and now have 1000 miles on the clock. Sunday morning traffic leaving Coalinga on I-5 is light and the AP2 works well. I-5 is pretty straight for miles. Due to our own error in understanding Tesla's Navigation, we miss the cutoff for Bakersfield and end up backtracking through Arvin to get to CA 58. The car was in AP2 and we were following a car that pulled into a left turn lane at a light exposing a car that was already stopped. I expected to have to break for the stopped car especially as Tesla had notified us that EB was currently not enabled but the car made a clean stop. I did not apply the brakes.
Later in the day along CA 138, a car coming toward us made a left hand turn across our lane. Our car immediately slowed hard and then recovered as the crossing car cleared the road. Very good positive actions by the MX.
. More than once during the day the car failed to recognize that it was entering a turn. No idea why.
. Later in the day in heavier traffic on 10, the car again hesitated in the same fashion as described in post 39.
. In other posts above, it has been noted that with AP2 on a road with a dip in it, the car would have "trouble". On CA 138 there is a stretch of highway through an area prone to flash floods that has a whole bunch of dipsy-doodles. After a number of miles of going up and down and having the car repeatably try to cross into the oncoming lane at the bottom of nearly every trough, I gave up trying to use AP2 until out of the area. Something in the programing triggers the car to pull left when it enters a short trough. A very consistent action.
. Also, in other threads I believe it has been reported that sometimes the car will tend to drift from an inside turn lane and into the car or truck that is being overtaken, What I observed over the last three days is that the above described drift does happen especially near the end the beginning or near the end of a turn. I believe this "drift", for lack of a better word, is because AP1 and 2 are not able to anticipate turns. We are able to set the car up for the corner before a entering the corner. AP is not be yet able to be proactive in this situation.
. Ping-ponging also tends to arise from the above "drift" problem. Once the car has entered a turn and it has projected the lane boundaries, the car will want to keep following the projection until it has gathered enough information to recognize that the lane is straightening. The car has most probably drifted wide as it comes out of the corner so the first thing it does is try to center itself but it goes a bit to far and so it pongs back. It centers quickly but there is this back and forth cycle times 1 or 1.5.
That is all that comes to mind for now.
Third day and now have 1000 miles on the clock. Sunday morning traffic leaving Coalinga on I-5 is light and the AP2 works well. I-5 is pretty straight for miles. Due to our own error in understanding Tesla's Navigation, we miss the cutoff for Bakersfield and end up backtracking through Arvin to get to CA 58. The car was in AP2 and we were following a car that pulled into a left turn lane at a light exposing a car that was already stopped. I expected to have to break for the stopped car especially as Tesla had notified us that EB was currently not enabled but the car made a clean stop. I did not apply the brakes.
Later in the day along CA 138, a car coming toward us made a left hand turn across our lane. Our car immediately slowed hard and then recovered as the crossing car cleared the road. Very good positive actions by the MX.
. More than once during the day the car failed to recognize that it was entering a turn. No idea why.
. Later in the day in heavier traffic on 10, the car again hesitated in the same fashion as described in post 39.
. In other posts above, it has been noted that with AP2 on a road with a dip in it, the car would have "trouble". On CA 138 there is a stretch of highway through an area prone to flash floods that has a whole bunch of dipsy-doodles. After a number of miles of going up and down and having the car repeatably try to cross into the oncoming lane at the bottom of nearly every trough, I gave up trying to use AP2 until out of the area. Something in the programing triggers the car to pull left when it enters a short trough. A very consistent action.
. Also, in other threads I believe it has been reported that sometimes the car will tend to drift from an inside turn lane and into the car or truck that is being overtaken, What I observed over the last three days is that the above described drift does happen especially near the end the beginning or near the end of a turn. I believe this "drift", for lack of a better word, is because AP1 and 2 are not able to anticipate turns. We are able to set the car up for the corner before a entering the corner. AP is not be yet able to be proactive in this situation.
. Ping-ponging also tends to arise from the above "drift" problem. Once the car has entered a turn and it has projected the lane boundaries, the car will want to keep following the projection until it has gathered enough information to recognize that the lane is straightening. The car has most probably drifted wide as it comes out of the corner so the first thing it does is try to center itself but it goes a bit to far and so it pongs back. It centers quickly but there is this back and forth cycle times 1 or 1.5.
That is all that comes to mind for now.
Siggy101
Member
I have seen something awesome and perhaps promising.
My AP2 S90D now shows me cars in other lanes, but only when I engage the indicator to initiate a lane change.
It only shows them during a lane change but it is now consistent. I see cars in other lanes every time I auto-lanechange.
Has anyone else got this? I've had it since the weekend.
My AP2 S90D now shows me cars in other lanes, but only when I engage the indicator to initiate a lane change.
It only shows them during a lane change but it is now consistent. I see cars in other lanes every time I auto-lanechange.
Has anyone else got this? I've had it since the weekend.
fasteddie7
Active Member
You must have been behind on updates, the last several updates have had this. Glad you have it now!
arcus
Active Member
17.38 is picking up a *little* bit... these things are so slow it make you think there is some blockbuster release just around the corner... <sigh>...
Perhaps someone more aware could explain better, but I believe there's a deep neural network in the car that processes data from the cameras and radars - learning about things in the process. So, if the car improves, it could be either due to a patch or a the car learning itself. I could be very wrong though.
⚡️ELECTROMAN⚡️
Village Idiot
RE: posts 34, 39 and 42: Beyond the basic calibration procedures, such as what is needed for the car to engage in auto parking and engage AP, I see no evidence of the car learning anything new. At this point AI does not seem to have been incorporated even though there is an NVDA chip running the show.
We now have 1,600+ miles on 2017.38 f87c64d5 and I have seen no change or improvement in the car's ability to stay centered, it continues to prefer the left side of the lane especially coming out of a sweeping curve, It continues to want to drift into the left lane when going into a trough, for unknown reasons it sometimes does not recognize the lane line and will drift across it. It sometimes recovers and finds it's lane. There has been no improvement in it's tendency or hesitate to overtake a truck in the adjacent lane, not every time but often enough.
My point is that we have covered 1,600 miles in 5 days, all with the subject update, over freeways, highways, flats, mountains and small towns. The lighter the traffic, the better the car behaves. It's behavior seems to become more dicy as traffic, especially with trucks, increases. One must always be awake to the fact that AP is still in Beta. It's current form is a great feat but it is not yet perfect. Driver beware.
Again, over 5 days and 1,600 miles, I see no evidence of learning. I expect the next update will improve AP, but it will not do it on it's own today. Now in Winslow, NV.
StefanSarzio
Member
The learning part isn't supposed to happen on your car, but on the Tesla servers. And you get the result only via software update.
StefanSarzio
Member
Drove some time with 2017.38 f87c64d5 now and also don't see any improvement since the .32 version. The car still veers left and right when someone merges close in front of me onto my lane, almost crashes into the left rear side of a car when I accelerate during a lane change, stays long times on either side of the lane without any apparent reason, feels sloppy and insecure most of the time.
If you drive an AP1 car from time to time you immediately feel how much more confident AP1 cars drive. There is still no comparison. AP2 is about half as confident/mature as AP1 at the moment.
If you drive an AP1 car from time to time you immediately feel how much more confident AP1 cars drive. There is still no comparison. AP2 is about half as confident/mature as AP1 at the moment.
Update after a 750 km trip yesterday. Mixed roads, mostly single lane only/undivided (not highway/divided). Little trafic.
This is a really good update and very predictable when it will fail.
On divided highways really solid, tracking in centre.
Motorways class B ( wide lanes) also really solid.
In the dark, on the narrower roads, car seemed to stay more to the dividing line, a bit to close for comfort when semis came in the other direction. Could be because of fenders on the outside triggering the ultrasonic.
Narrower roads with lots of bends, speeding by 10 km/h, is a big challenge. Sometimes ok, sometimes not.
Did not try to take exits or on ramps, not even bus stops, but does move the wheel a bit.
Difficulties over crests, beware. Especially in bends, loses track and start searching for lanes. Very predictable.
Does ping and ping when lane markers are not straight. Not the old oscillation, car is just not normalizing the path but following all the small wrinkling of the lane markers.
Tacc ( no autosteer) also reduces speeds by itself in some bends when speeding.
This is a really good update and very predictable when it will fail.
On divided highways really solid, tracking in centre.
Motorways class B ( wide lanes) also really solid.
In the dark, on the narrower roads, car seemed to stay more to the dividing line, a bit to close for comfort when semis came in the other direction. Could be because of fenders on the outside triggering the ultrasonic.
Narrower roads with lots of bends, speeding by 10 km/h, is a big challenge. Sometimes ok, sometimes not.
Did not try to take exits or on ramps, not even bus stops, but does move the wheel a bit.
Difficulties over crests, beware. Especially in bends, loses track and start searching for lanes. Very predictable.
Does ping and ping when lane markers are not straight. Not the old oscillation, car is just not normalizing the path but following all the small wrinkling of the lane markers.
Tacc ( no autosteer) also reduces speeds by itself in some bends when speeding.
Forgot one issue, auto high beams.
The are quite specific to dim but are way to slow to fire up again after oncoming trafic passed. Between 2-3 seconds, must be fixed!
I usually practice engaging when at the side of oncoming car, meaning no driving in the dark. 3 seconds at 90 km/h in pitch black is just not good enough.
The are quite specific to dim but are way to slow to fire up again after oncoming trafic passed. Between 2-3 seconds, must be fixed!
I usually practice engaging when at the side of oncoming car, meaning no driving in the dark. 3 seconds at 90 km/h in pitch black is just not good enough.
