AP1 - MS Suddenly Veering Off Road On Road Crests - Any Similar Experiences?

[OK-S]

Hi Brainstrust,

My 2015 MS with AP1 plus MCU3 and all other SW upgrades was just on a 4,000 km cross-country trip. Flawless and enjoyable performance as usual EXCEPT that on completely straight sections of two-lane roadways (90-100 kph/55-60 mph) with rolling crests the autopilot began to exhibit abrupt veering behaviour. Even a couple of times on a rolling 4-lane section of motorway. These were short crests, but just high enough that one could not see the other side of them. In the past, the car would tend to keep its direction rather than veer off course.

Mostly the veering was towards the gutter (off the roadway) but a few times in the other direction towards the oncoming lane. I kept checking that it was not an anomaly (even did a dual button reset) but it was repeatable dozens of times and required quick intervention to correct it - essentially overriding the AP. The AP did not disengage by itself or offer any audible or visible warning - except when I took back control.

I have driven this car for over 40,000 km and have never experienced this behaviour before. The windscreen was not dirty and does not look visibly pitted in the lens area. The behaviour of the autopilot in other circumstances appears normal.

So, can this be a failure due to a more recent SW update or is the camera itself likely to be the cause? Or something else? There is clearly a safety issue here in my opinion. I will, of course, book for a service but I would like to be prepared should others have encountered and resolved this issue. (on SW 2022.8.2)

 

[SilverGS]

There's no way you have MCU3 on your car... I'm sure you meant MCU2

But the odd behavior of your car in AP is very strange.

 

[dannycamps]

I'm fairly certain there are no longer updates being done to the AP1 (MobileEye) stack at this point in time but you never know if Tesla tweaked something else that would cause the behavior you were seeing.

 

[PCMc]

I have a mid-2016 MS90D which I've owned for just under 6 years.

What is described as cresting incline is something which I've experienced ever since I first purchased my car in June 2016. It is difficult to judge the severity of the veering from the description to say if what is described is indication of capability regressing or now. We all have our subjective personal definitions of abrupt and I do not know if OPs definition and mine match.

I have found if you watch the lane markings closely when you crest I notice they may tend to bend towards the side of the roadway. I've associated that with how the vision system is reacting and perceiving the curvature of the road cresting down, essentially how it's mapping an out of plan curvature visual back to more of an in-plane representation. I've learned to recognize this as just one of the attributes of the monocular AP1 vision system, anticipate it, which is why I may not subjectively react to it as 'abrupt'.

I actually have a local stretch of road where this type of crest exists. I'll actually intentionally drive it at times when new releases come out just to 'recalibrate' myself on my expectations for how my car will respond.

 

[PCMc]

I'm fairly certain there are no longer updates being done to the AP1 (MobileEye) stack at this point in time but you never know if Tesla tweaked something else that would cause the behavior you were seeing.

I agree it's difficult to be certain, but I wouldn't jump so quickly to say there are no tweaks or fine tuning being done on AP1. While the rate of software updates with MCU1 have reduced to once every 2-3 months over late 2020 through 2021, I would continue to see at times subtle changes in how AP1 would handle certain types of conditions. I agree, nothing as dramatic as the evolution that's been going on with AP2/2.5/3 related to FSD, but I'm not so quick to say no subtle changes to calibration.

Just my opinion from what I perceive with my AP1 vehicle.

 

[OK-S]

There's no way you have MCU3 on your car... I'm sure you meant MCU2

But the odd behavior of your car in AP is very strange.

Yes, sorry, MCU 2 upgrade.

 

[OK-S]

I have a mid-2016 MS90D which I've owned for just under 6 years.

What is described as cresting incline is something which I've experienced ever since I first purchased my car in June 2016. It is difficult to judge the severity of the veering from the description to say if what is described is indication of capability regressing or now. We all have our subjective personal definitions of abrupt and I do not know if OPs definition and mine match.

I have found if you watch the lane markings closely when you crest I notice they may tend to bend towards the side of the roadway. I've associated that with how the vision system is reacting and perceiving the curvature of the road cresting down, essentially how it's mapping an out of plan curvature visual back to more of an in-plane representation. I've learned to recognize this as just one of the attributes of the monocular AP1 vision system, anticipate it, which is why I may not subjectively react to it as 'abrupt'.

I actually have a local stretch of road where this type of crest exists. I'll actually intentionally drive it at times when new releases come out just to 'recalibrate' myself on my expectations for how my car will respond.

There were no curvatures involved in the road itself but I understand the visual distortion that the rise may inflict but then also on the mid strip. Did you find this issue became more pronounced? Abrupt to me is that it would have surely put me off the road on the other side of the crest (or into the oncoming lane) - it was a significant and sudden adjustment that required that I quickly tug the wheel to overcome it and cancel AP in the process. Of course, at these speeds, it does not take a lot of input to change direction fast.

 

[OK-S]

I agree it's difficult to be certain, but I wouldn't jump so quickly to say there are no tweaks or fine tuning being done on AP1. While the rate of software updates with MCU1 have reduced to once every 2-3 months over late 2020 through 2021, I would continue to see at times subtle changes in how AP1 would handle certain types of conditions. I agree, nothing as dramatic as the evolution that's been going on with AP2/2.5/3 related to FSD, but I'm not so quick to say no subtle changes to calibration.

Just my opinion from what I perceive with my AP1 vehicle.

I also agree that there appear to be minor but ongoing changes to AP1 over the last two years - mostly for the better but also adding more sensitivity to checking on steering inputs. Speed sign sensitivity also appears to have progressed.

 

[AustinP]

I confirm that our AP1 car from 2015 has had that problem since always.
At some point, in the early days of Model 3, this same behaviour was noticed by, if memory serves, Consumer Report (or another media), and Tesla was quite quick to fix it... for the AP2.x & AP3. Not for AP1 unfortunately.
Note: I sent a mail to Tesla back then to document & ask about fixing this.

I'm used to it, and automatically anticipate.

But I indeed always hope for this to be fixed one day, and as @PCMc also does, I test this after each update... who knows

 

[OK-S]

I confirm that our AP1 car from 2015 has had that problem since always.
At some point, in the early days of Model 3, this same behaviour was noticed by, if memory serves, Consumer Report (or another media), and Tesla was quite quick to fix it... for the AP2.x & AP3. Not for AP1 unfortunately.
Note: I sent a mail to Tesla back then to document & ask about fixing this.

I'm used to it, and automatically anticipate.

But I indeed always hope for this to be fixed one day, and as @PCMc also does, I test this after each update... who knows

Thanks, Austin, it was a new experience for me after a couple of years driving this car. We may have had the odd drop-out but never this quite repeatable tug - my wife kept looking at me wondering if I was being intentionally annoying! She has never taken to autopilot like I have.

Anyway, I have asked for a service appointment and usually, that leads to a call and discussion of my issue - so perhaps there will be some info on this and whether it is something they are hearing more about. BTW, I find it interesting that our Autopilot is still listed as "Beta" on the screen!

 

[PCMc]

@OK-S - The examples I cited also were cases where there is no actual left/right curvature of the road, but the monocular vision of the single AP1 camera perceives the downward curvature coming over the crest as curvature to the side. I generally observe if I watch the lane markings on the display, I can actually see them start to bend as if there is a curve in the road. Autopilot then reacts to try and follow where it sees the lanes, just in this case, that's unfortunately out of the lane. The road crest I use for my reference is quite near my home, on the edge of a residential area, so the speed limit is only about 60-65 km/hr. If your experiences are more motor way at higher speeds 100 km/hr or higher, then that also can explain why you reacted to it as more abrupt.

I'm actually on an international assignment, so I spend 2-3 months away from my car at a time. That's where I do not have any driving experience now for last two months to provide any feedback on whether a recent software update has made it worse. I also upgraded to MCU2 and the last two software releases I have any real driving experience with were 2021.44.30.8 and 2022.4. I didn't sense any major changes, pro or con with those.

I have since received 2022.4.5, 2022.4.5.3, and now 2022.8.2. However those all been installed remotely since I left US with my car parked in my garage.

 

[dark cloud]

Yeah, perpetual beta.
Based on 150,000 km AP1 for me hasn't changed in a couple of years. The last real change involved lane changes, where less of the car needed to be in the changed lane for one to release the turn stalk and not have the car snap back into the original lane.

Yet AP1 seems to be semi-organic, having good days and bad days. The same conditions on the same roads, the same time of the day, the same amount of cleanliness of the windshield, and painted lines, and somehow AP1 has behaved differently, slightly. It makes no sense but that is my observations. Just like a laptop when sometimes a program runs speedily and other times it is not as efficient, like something running in the background: mothership communication?
So one cannot download an update and do one drive and determine a difference; the same drive needs to occur several times, and if their car is like mine its tracking ability and precision will vary slightly. On really bad days my wife complains about getting sea sick and non-driver assistance "normal" driving is necessary. But I still have MCU1; maybe it is more consistent with MCU2 now.

The one factor not discussed here is the location of the sun. The same crest of road into the sun at certain angles will definitely adversely affect AP1 performance. So that might be the reason AP1 might be presumed to not work as well as before.

 

[krsgio]

my car's AP has always been unreliable on hilly roads where the AP loses visibility of lines then tries to drive me off the road. I just don't use it in those scenarios anymore. I just never understood why they didn't use similar logic to crossing unmarked intersections where it defaults to going straight rather than freaking out.

 

[TWoodstream]

I'd like to chime in that I've had my S drift out of the carpool lane under the following conditions:
Top of the hill, sun was shining directly at the front of the car, slight curve in the road

Do the newer versions of AP fix this issue?

 

[krsgio]

I'd like to chime in that I've had my S drift out of the carpool lane under the following conditions:
Top of the hill, sun was shining directly at the front of the car, slight curve in the road

Do the newer versions of AP fix this issue?

After driving my moms FSD beta M3LR the other day I'll gladly take my AP1 losing sight over crest than the random braking and other weird things it was constantly doing.

 

[David29]

My 2015 S has AP1 and has exhibited similar but less severe behavior from the beginning. There is a nearby stretch of I-95 where i have observed this over the years. My sense had been that it got better after the first version, but then more recently has been drifting more than when it was at its best. As others have said, though, it is difficult to know if this is an accurate, consistent observation or just anecdotal -- the system may not have changed but the circumstances may vary (lighting, speed, lane marking condition, etc.).
But I have never experienced any deviation as severe as the OP describes, only a bit of wandering within my lane and maybe a bit of slip into an adjacent lane until the car crests the rise and can see the lanes clearly again.

One question to the OP -- you mention a "two lane" roadway. Do you mean one lane in each direction? In that case, I think you are talking about non-divided roads, and I think AP has different capabilities on divided vs, undivided roads.

 

[OK-S]

My 2015 S has AP1 and has exhibited similar but less severe behavior from the beginning. There is a nearby stretch of I-95 where i have observed this over the years. My sense had been that it got better after the first version, but then more recently has been drifting more than when it was at its best. As others have said, though, it is difficult to know if this is an accurate, consistent observation or just anecdotal -- the system may not have changed but the circumstances may vary (lighting, speed, lane marking condition, etc.).
But I have never experienced any deviation as severe as the OP describes, only a bit of wandering within my lane and maybe a bit of slip into an adjacent lane until the car crests the rise and can see the lanes clearly again.

One question to the OP -- you mention a "two lane" roadway. Do you mean one lane in each direction? In that case, I think you are talking about non-divided roads, and I think AP has different capabilities on divided vs, undivided roads.

The experience has been mostly on undivided roads - one lane each way. Well marked on both sides, straight. Multi-lane, divided freeways tend to have fewer similar crests but it did do the same behaviour in on a stretch where there were uncharacteristic crests as well.

 

[OK-S]

After driving my moms FSD beta M3LR the other day I'll gladly take my AP1 losing sight over crest than the random braking and other weird things it was constantly doing.

I hear that a lot. AP1 does a lot well enough that I use it often. False braking alarms are rare (perhaps two on our 4,000km journey) especially after fine-tuning the radar positioning that used to pause when passing. The only ongoing annoyance is the slowness to respond to a car that has already moved out of the lane after braking. There are a couple of seconds in lag that I often have to overcome manually to re-accelerate.

 

[OK-S]

I'd like to chime in that I've had my S drift out of the carpool lane under the following conditions:
Top of the hill, sun was shining directly at the front of the car, slight curve in the road

Do the newer versions of AP fix this issue?

If the sun is really low and direct (as in the early morning, worse with a slight mist) I find the AP may just cut out rather than continue - a preferable option! It is rare but does happen.

 

[OK-S]

my car's AP has always been unreliable on hilly roads where the AP loses visibility of lines then tries to drive me off the road. I just don't use it in those scenarios anymore. I just never understood why they didn't use similar logic to crossing unmarked intersections where it defaults to going straight rather than freaking out.

Agreed - you would assume this would be the temporary default and easily programmed in. The misreading looks to instead be initiating a more drastic response.