Auto pilot moving left and right in the lane with huge over correction?

[Mardak]

This lane centering regression had 2 especially bad results newly introduced in 2019.40.2.1 for me:
- ping-pong behavior during a curve that resulted in a "take over immediately"
- an automatic lane change during a curve unnecessarily crossed over to a 3rd lane (but corrected back to the center of the 2nd lane)

 

[CharleyBC]

Could someone who is experiencing extreme pinging and ponging please capture a short dashcam video and post it on this thread? I would like to see what it looks like, and I imagine others do too.

 

[Takeo]

Has anyone had their Model 3 with the newest update see the car over correcting within the same lane moving back and forth from side to side? Mine was going crazy last night and got my wife very scared... anyone else with the same issue?

Has anyone had their Model 3 with the newest update see the car over correcting within the same lane moving back and forth from side to side? Mine was going crazy last night and got my wife very scared... anyone else with the same issue?

I've recently experienced the same.

Since late last year/early this year, I've been impressed with how well "auto-pilot" keeps the car pretty steady inside the lane, but within the past week something has definitely changed. I'm experiencing a minor bit of the "ping pong" affect I remember reading about 2 years or so ago. In my case, it's not BAD, but it is definitely noticeable.

 

[gjunky]

I can definitely feel the difference with 40.2.1 as well. It ping pongs back and forth in a lane. It was very steady in the previous releases. It is not super strong in 40.2.1 but i can feel it, with or without my hands on the wheel (for testing, no worries)

regular 18” tires and a noticeable difference from 36.x,, my last release

 

[dusterbuster]

I experienced the ping-ponging overcorrection with 40.2.1 yesterday on a long highway drive. It was especially noticeable with lane shifts due to highway construction, etc. I started getting concerned that other drivers would think I was a drunk driver.

On the plus side, the lange change with signal (I don't use auto lane change) works much better now (doesn't hesitate and wait forever when the lane is clear before initiating and completing the lane change).

 

[emmz0r]

Started ping ponging for me too!

I look at the visualization and the lane lines are much more uncertain and wobbly. Did something have to give in AP 2.5 because of the new features needing more frame time?

 

[Rojer Lockless]

The Ping pong is just crazy now.

 

[Pilot7478]

If you remember, the ping-pong behavior was completely eliminated in v8 just before v9 was released.
With v9 it suddenly returned.
With v10 it even worse.

My theory is that it's their way to figure out if a driver has hands on the wheel and it was back after those bad crashes about a year ago or so...

 

[mhan00]

LOL. There are NEVER any issues for you. Come drive a long weekend in LA, then report back.

I did a long drive from LA to SF and back last week and have driven LA freeways all week (been on an El Chato Taco and Diddy Riese kick since I’ve been back, so I’ve been driving from the South Bay to the UCLA area). No issues. The car does drift more in the lane for me than in previous releases (it seems to try to edge away from cars in adjacent lanes, which it hasn’t done before), but other than that no problem.

 

[tvad]

I did a long drive from LA to SF and back last week and have driven LA freeways all week (been on an El Chato Taco and Diddy Riese kick since I’ve been back, so I’ve been driving from the South Bay to the UCLA area). No issues. The car does drift more in the lane for me than in previous releases (it seems to try to edge away from cars in adjacent lanes, which it hasn’t done before), but other than that no problem.

Wish our experience was the same.

Cheers.

 

[Mardak]

Did something have to give in AP 2.5 because of the new features needing more frame time?

Are people noticing ping-pong behavior only with HW2.5- and not HW3? It seems possible that the software is primarily optimized and tested with HW3 now, and there could have been some limits with HW2.5 that are introducing these regressions.

 

[Sasquatch]

I notice pretty severe ping-ponging with HW2.5 after the most recent update. In fact, I have turned off AP twice with cops behind me because I didn't want to appear drunk.

 

[emmz0r]

Are people noticing ping-pong behavior only with HW2.5- and not HW3? It seems possible that the software is primarily optimized and tested with HW3 now, and there could have been some limits with HW2.5 that are introducing these regressions.

I have 2.5. It seems like they have changed or scaled down the lane detection. It's stable and doesn't weave when the lines are clear. But it's way less robust now when the lines are unclear. This is the source of the weaving.

Also ut seems to detect dotted lines better at the same time. Seems more responsive in that effect. As usual, one step forward, sometimes multiple steps backwards

 

[Routybouty]

40.2.1 is horrible for me on HW2.5. It is so bad since 40.2.1 and it was slightly detectable in the previous update also. I haven't had nice solid lane keeping since .28-.32ish. Do the developers even try these in the real world? Do the early access people even give feedback? It's unacceptable to the point I'm not even using AP on certain roads that I always have before.

Updates need a rating system. If 50% or more people don't like it, you get the option to revert!

 

[Pilot7478]

I thought people just overreacting to this ping-pong problem as I did not find it to be that bad on my usual route.
However, I did a few hours freeway trip this weekend and found it so bad that I had to even overtake the steering a few times as people in the car started to be thrown side to side and obviously protesting.
I could not find a pattern or specific conditions at which it starts.
The quick lane changes were working great, but this ping-ponging did not allow me to enjoy this trip as I could not rely on AP as I used to.

 

[CA_Elektra]

VERY noticeable regression in holding lane position for me. The ping-pong back and forth is very unsettling. Also, it seems to swerve way too much when it needs to change position due to change of lane layout like a split into 2 lanes or an onramp that disappears forcing the car to move left over to into the right-most lane. Under those circumstances the car swerves so badly I've repeatedly though to myself if a cop was behind me he'd pull me over for suspected intoxication. This is the first update that has been a major step backward for me.

The post above about this being a known issue on a subset of cars is interesting. I do have the winter tire package on (also 2.5HW). Whatever it is, it's almost not usable which is a first.

And unrelated to that seems to be a regression in the front camera where it will suddenly say it's blocked and AP won't work. The first time I was driving into the sun so thought that was the cause, however it has since happened when the glass was clear and the sun was not the cause. So there seems to be a regression on camera processing.

 

[CA_Elektra]

have had subtle weave/ping pong for almost half a year. Comes and goes. Sometimes it's gone completely, other times it's noticeable (but not leaving the lane noticeable, just I can feel it when holding the steering wheel). I did notice when driving with a lot of trucks that the "move away from them" feature does make the car seem like it's ping ponging a lot, but when you realize that it's just passing trucks, and goes away when there are no trucks around, it makes you feel a bit better (ie it's still annoying, but you understand why it is happening).

I wondered if Tesla implemented a subtle sway in the driving in order to be able to pick up on whether you've got your hand on the wheel during long straight drives (ie if it stays straight, there's no torque on the wheel, vs. if it moves ever so slightly, it can detect if your hand is still there without having to nag you to jiggle the wheel).

I only notice the swimming effect when the cement/asphalt of the lane is mismatched with the painted lines, as is common in highway construction zones in California.

 

[emmz0r]

Has this been fixed in 2019.40.50.1 ?

 

[Pilot7478]

Has this been fixed in 2019.40.50.1 ?

No, it has not.
Did a drive yesterday - exact same behavior.

 

[diplomat33]

I could be wrong but I suspect that the ping ponging is the result of Tesla tweaking lane keeping for city streets since they are focused on "city NOA" now and obviously they did not get it just right yet. I am sure it will get fixed in a future update.