Autopilot Suddenly Possessed?

[markn455]

I am running firmware 2018.18 4c72e69 which was installed a few weeks ago. Recently, when I activate EAP the car will seemingly without reason suddenly slam on the brakes within a minute of activation. Then, everything appears normal and operates as expected. There has been one other time where I am driving along on a clear day, no traffic, on straight and level road and suddenly it will abruptly apply the brakes.

On March 23, I came out of a Non AP MS and purchased a 2018 MS100D with EAP. You could say my EAP experience is almost zero. I do not yet trust AP. Consequently I rarely use it. This type of behavior is certainly not giving me any confidence to trust it enough to use it. My past AP experience was with an AP1 car which seemed to work far better than this AP 2.5 car.

Is this expected behavior?

 

[Tam]

...Is this expected behavior?

It's beta, an unfinished product, so you should expect it to behave like one.

 

[xyeahtony]

occasionally I hack into random teslas and play around with my xbox controller.

(disclaimer: totally kidding in case someone actually believes im serious)

I've had phantom braking before too, happens rarely and is super annoying but thankfully nobody is tailgating me.

 

[buttershrimp]

I am running firmware 2018.18 4c72e69 which was installed a few weeks ago. Recently, when I activate EAP the car will seemingly without reason suddenly slam on the brakes within a minute of activation. Then, everything appears normal and operates as expected. There has been one other time where I am driving along on a clear day, no traffic, on straight and level road and suddenly it will abruptly apply the brakes.

On March 23, I came out of a Non AP MS and purchased a 2018 MS100D with EAP. You could say my EAP experience is almost zero. I do not yet trust AP. Consequently I rarely use it. This type of behavior is certainly not giving me any confidence to trust it enough to use it. My past AP experience was with an AP1 car which seemed to work far better than this AP 2.5 car.

Is this expected behavior?

For what it's worth, if you don't use it often, it may not be fully calibrated. I am more confident that this is a real thing after driving a newish model X loaner for 4 days with 20.5 firmware and had it go from consistently too far left.... to slowly centering over the course of a few hundred miles. My car on the same firmware, has been true center the whole time. It makes sense that Tesla has built in some ability for the car to correct and zero in to deal with variances in camera placement. My advice (not that you are asking) is to use AP as much as you can just keeping your hands on the wheel as though you are driving.... I'd do that and not be afraid to step on the gas pedal while on autopilot when it brakes too much, such as when a car in front of you takes off quickly from a red light and autopilot is too slow to accelerate.... a lot of people don't know that they can accelerate while on autopilot in these scenarios without penalty.

hope this helps.

 

[Acho]

It's beta, an unfinished product, so you should expect it to behave like one.

May be yellow bumper sticker with "Autopilot /Beta/ in Use" will help avoiding unpleasant surprises for drivers behind

 

[Daniellane]

I am running firmware 2018.18 4c72e69 which was installed a few weeks ago. Recently, when I activate EAP the car will seemingly without reason suddenly slam on the brakes within a minute of activation. Then, everything appears normal and operates as expected. There has been one other time where I am driving along on a clear day, no traffic, on straight and level road and suddenly it will abruptly apply the brakes.

On March 23, I came out of a Non AP MS and purchased a 2018 MS100D with EAP. You could say my EAP experience is almost zero. I do not yet trust AP. Consequently I rarely use it. This type of behavior is certainly not giving me any confidence to trust it enough to use it. My past AP experience was with an AP1 car which seemed to work far better than this AP 2.5 car.

Is this expected behavior?

I have never experienced abrupt braking without some observable external cause. Definitely overreactions to situations like stopping for cars exiting the roadway left or right that I would anticipate being out of my way by the time I get to them. I just press down on the accelerator to override...
If I were you, I would make an appointment at your SC and get it diagnosed. Could be a bad camera, sensor or cable or possibly the ECU.

 

[Tam]

When I first got mine more than a year ago in March 2017, Autopilots applied brakes quite often when I thought there's no reasons.

I then associated its frequency with

1) coming toward an overpass ahead

2) there's a car turning left or right far ahead.

3) there's a parking car within the same lane as I was traveling.

I didn't mind because that beats being killed in Florida accident because the radar detected the tractor-trailer and the:

"Radar tunes out what looks like an overhead road sign to avoid false braking events"

The system is now doing a much better job but it is still being finetuned.

 

[Ande]

Sounds like it's time for an excorsism.

 

[Impala1ss]

It's beta, an unfinished product, so you should expect it to behave like one.

I guess everyone accepts your explanation but why is it still BETA after so many years? Seems like it is better for Tesla to cover its A** by saying BETA rather than complete the software out of BETA.

 

[animorph]

It's beta because they keep modifying it, QC is a little less than we would like, and Tesla can't be sure how it will behave in all the corner cases we may expose it to. Doesn't hurt that people might be a little more cautious if they think the SW might have some bugs.

 

[markn455]

2018.21.9 update was installed over the weekend. The best I can tell the problem I mentioned does not occur on this version. I will know more over the next few days. For now, this build is a great improvement in addition to seemingly resolving my issue.

 

[The Duke]

If/when you get phantom braking, please note the speed limit. I have one freeway interchange that we used to go 70 mph across now slowing to 40 because Nicki thinks that is the speed limit there. 237 east to 880 north in the car pool lane, just south of the Tesla factory.

 

[RandyJ]

Prior to the latest software version update, phantom braking was a regular occurrence. I was almost rear ended by a semi when my S braked from 75 to 30 mph. I stopped using the feature.

I now have 2018.21.9 and the problem is still there. Off the freeway, it is worse at night, with phantom cars showing on the display at the same time as the braking. On the freeway, as I pass cars in adjacent lanes they appear on the display to be merging into my lane, when in fact they are clearly in their defined lane. The adjacent car is highlighted red and the brakes are applied.

On left hand curves, as I pass cars in the right hand adjacent lane they appear on the display to merge into my lane just ahead of me and then disappear.

Software still needs some major improvements.

 

[MorrisonHiker]

If/when you get phantom braking, please note the speed limit. I have one freeway interchange that we used to go 70 mph across now slowing to 40 because Nicki thinks that is the speed limit there. 237 east to 880 north in the car pool lane, just south of the Tesla factory.

I had a similar experience where I had TACC set to 70 and the car lowered it to 55 even though I was still on the main highway. I guess it thought I was taking the exit ramp. A few seconds later, it went back up to 65 but I had to manually set it back to 70. I have driven that same stretch of road several times since then and it only happened the first time.

I had it happen on a different highway today. TACC was set to 70 and it lowered to 60 but I was still on the main highway.

 

[Destiny1701]

I really miss AP1. Phantom braking? Never.
Drunken curves? Nope.
This was back in the glory days of 2015/16 when steering nags were also almost non existent.

Current .21 AP2 on our MX just phantom braked for no reason on a residential street at 30mph...no obstruction or vehicle around.

Sigh...

 

[TaoJones]

I really miss AP1. Phantom braking? Never.
Drunken curves? Nope.
This was back in the glory days of 2015/16 when steering nags were also almost non existent.

Current .21 AP2 on our MX just phantom braked for no reason on a residential street at 30mph...no obstruction or vehicle around.

Sigh...

Agreed 100%. My goal is to be back in an AP1 car by the end of the year and then to just ride out the next 2-3 years of development. By then, we will hopefully have FSD (specifically traffic signal reaction) and other AP(3) features that have been promised.

Put another way, assuming that v9.1 will be underwhelming, my limit for the whole “fool me once, fool me twice game” is pretty much at hand with regard to the embarrassment that is AP2 relative to AP1.

Fortunately, TACC is still decent which is to say not screwed up by recent updates, and TACC is just fine with AP1. That’s enough for me until a few more years pass and the rest shakes itself out. The nag thing was pretty much the last straw, and the phantom braking thing was and is just flat out dangerous. So unless Tesla’s going to step up and pay my deductible when I get rear-ended as a result of said phantom braking, I’m done.

 

[buttershrimp]

Agreed 100%. My goal is to be back in an AP1 car by the end of the year and then to just ride out the next 2-3 years of development. By then, we will hopefully have FSD (specifically traffic signal reaction) and other AP(3) features that have been promised.

Put another way, assuming that v9.1 will be underwhelming, my limit for the whole “fool me once, fool me twice game” is pretty much at hand with regard to the embarrassment that is AP2 relative to AP1.

Fortunately, TACC is still decent which is to say not screwed up by recent updates, and TACC is just fine with AP1. That’s enough for me until a few more years pass and the rest shakes itself out. The nag thing was pretty much the last straw, and the phantom braking thing was and is just flat out dangerous. So unless Tesla’s going to step up and pay my deductible when I get rear-ended as a result of said phantom braking, I’m done.

The real question to me will be, what features will be available in 9.1 in an AP2 car that are not available in a AP1 car. One might hope that they are holding off on implementing obvious and glaring differences with AP1,e.g. speed limit sign recognition and more frequent autolane change avilability, because they are trying to do something AP1 can't possibly do. I'm not hopeful on this because people like George Hotz are showing that little is needed in terms of hardware to do a lot of the tasks.

Anyway, I'm also hoping that some of Bjorn's ideas will see the light of day (from his video on top needed features of tesla software), ability to monitor battery temp, turn off the center screen, etc.

I for one, would love to see the ability to put the map on the IC without engaging the navigation system.

However, I will say this.... If I can predict one reason why @TaoJones ends up not going back to AP1, it's what is showing in this picture. If they can fix the speed limit sign issue, and make lane change more available, AP1 will officially become second best assuming 10.4 gains will hold.

 

[TaoJones]

I for one, would love to see the ability to put the map on the IC without engaging the navigation system.

However, I will say this.... If I can predict one reason why @TaoJones ends up not going back to AP1, it's what is showing in this picture. If they can fix the speed limit sign issue, and make lane change more available, AP1 will officially become second best assuming 10.4 gains will hold.

View attachment 311550

Eh.

Alas, a) long term, the IC will go the way of the dodo, as the Model 3 interface is clearly the way of the future, and b) short term, it’s not just fixing the broken speed limit database - it’s that there is no fix even being given a real timeline for that that leverages a camera or cameras, and not as part of a hybrid solution for AP3. Certainly not Tesla - see their website. It’s wishful thinking based upon merely a strong supposition that they’ve got the extra cameras and might as well use them at some point for something other than selling the non-existent E in EAP or FSD.

Putting the speed limit database mess aside, it seems realistic that we also surely won’t see traffic signal recognition for some time. See 10cm GPS for starters. What does that leave - the auto lane change fantasy that the welcher who shall not be named lost $11,120 over? Big deal. It’ll come, but so what?

I see yer picture and will add a picture I took the other day. What does it matter if that information is not utilized?

Here’s what I think we *will* get with v9.1 - blind spot recognition with flashing sides of the IC. That’s it. We might get blind spot *reaction* which would be admittedly very cool, except that you may recall the video from 2 years ago in which an AP1 car evaded an encroaching work truck during an on ramp merge. So again, level pegging at best with AP1, *with the potential of additional use cases* due to the 4x-8x hardware that comes with AP2.

What do we really need for useful full-featured FSD? Aside from 10cm GPS interaction, more processing power and better data comm. Tesla *will* deliver both of the last two via their commitment to a new SoC/board and via Skylink in lieu of 5G, the latter of which ain’t gonna be continental let alone global for at least 5 years. Tesla even wins *that* battle in the next 5 years if they can execute their launch schedule.

Anyway, if I had to guess, we’re still 18-24 months out before getting something very cool and useful. Would love to be wrong if it arrived sooner. But it ain’t coming on August 31st.

Until that day, whenever it is, it will be an AP1 car for me.

Meanwhile, here’s to v9.1 this coming late Fall or early Winter. Whatever it will be, it’ll be the next step in the journey toward what many of us paid for 1.5+ years ago now. Chump change compared to how long the earliest adopters had to wait, right? (That right there being a freebie for the fanbois as I’m out of Kool Aid).

 

[buttershrimp]

Eh.

Alas, a) long term, the IC will go the way of the dodo, as the Model 3 interface is clearly the way of the future, and b) short term, it’s not just fixing the broken speed limit database - it’s that there is no fix even being given a real timeline for that that leverages a camera or cameras, and not as part of a hybrid solution for AP3. Certainly not Tesla - see their website. It’s wishful thinking based upon merely a strong supposition that they’ve got the extra cameras and might as well use them at some point for something other than selling the non-existent E in EAP or FSD.

Putting the speed limit database mess aside, it seems realistic that we also surely won’t see traffic signal recognition for some time. See 10cm GPS for starters. What does that leave - the auto lane change fantasy that the welcher who shall not be named lost $11,120 over? Big deal. It’ll come, but so what?

I see yer picture and will add a picture I took the other day. What does it matter if that information is not utilized?

View attachment 311555

Here’s what I think we *will* get with v9.1 - blind spot recognition with flashing sides of the IC. That’s it. We might get blind spot *reaction* which would be admittedly very cool, except that you may recall the video from 2 years ago in which an AP1 car evaded an encroaching work truck during an on ramp merge. So again, level pegging at best with AP1, *with the potential of additional use cases* due to the 4x-8x hardware that comes with AP2.

What do we really need for useful full-featured FSD? Aside from 10cm GPS interaction, more processing power and better data comm. Tesla *will* deliver both of the last two via their commitment to a new SoC/board and via Skylink in lieu of 5G, the latter of which ain’t gonna be continental let alone global for at least 5 years. Tesla even wins *that* battle in the next 5 years if they can execute their launch schedule.

Anyway, if I had to guess, we’re still 18-24 months out before getting something very cool and useful. Would love to be wrong if it arrived sooner. But it ain’t coming on August 31st.

Until that day, whenever it is, it will be an AP1 car for me.

Meanwhile, here’s to v9.1 this coming late Fall or early Winter. Whatever it will be, it’ll be the next step in the journey toward what many of us paid for 1.5+ years ago now. Chump change compared to how long the earliest adopters had to wait, right? (That right there being a freebie for the fanbois as I’m out of Kool Aid).

I have to say, from your picture, I'm thrilled about Tesla's new limo detection capability.

I like your timeline on the giant leap for Autopilot and FSD... 18-24 months to see something close to FSD, I'm betting that my Mad Max Event will occur when the first few roadsters start getting test driven. By then, the prospect of a 200 kW, V3 charging of the model S and X and the HUD teasers are going to overshadow the miracle that is already autopilot.

This isn't blind optimism, in my opinion, autopilot, even more than the drive train, is the 8th wonder of the world.

However, acclimation,is every devoted Tesla owner's destiny.

I will cheer Tesla on (even with it's shortcomings) as the engineers slave away at developing safer software. I don't envy the task, as the public and media takes a piss on autopilot, as they ignore every other iteration of assisted driving that hides under the umbrella of "thorough safety testing" when the truth is they are scrambling to catch up on the tech front, or to escape the lawyers and bureaucracy that limit their release. Who knows, maybe MobilEye will actually one day be allowed to deploy some of their technology in 2020. Maybe then some other automaker will actually result in consequential and meaningful use of autonomous driving technology.

Unless we are drinking the heavily spiked Kool Aid that @Bladerskb sells with his steering wheel green lights nanny complaining about your eye contact)

The rest of us, so disillusioned with lofty goals of Elon's timelines and without the patience and persistence to bother exploring the auspices of autopilot capability for long enough to let a firmware calibrate or to realize that the dissatisfaction among certain forum folks has more to do with what feels comfortable to them as they rest their hands too perfectly on each side of the steering wheel..... but I digress.

As we discuss various interesting aspects of autopilot, I am very much looking forward to reading these posts a year or two from now and seeing how this all played out. I don't know what will happen. But here's to version 9... and here's to Elon frickin' Musk.... a person who is staying in the fight far longer than I would have anticipated. If he can limit himself to less than 3 mental breakdowns per year in the next five years, I think these goals will eventually happen.

 

[TaoJones]

I have to say, from your picture, I'm thrilled about Tesla's new limo detection capability.

I like your timeline on the giant leap for Autopilot and FSD... 18-24 months to see something close to FSD, I'm betting that my Mad Max Event will occur when the first few roadsters start getting test driven. By then, the prospect of a 200 kW, V3 charging of the model S and X and the HUD teasers are going to overshadow the miracle that is already autopilot.

This isn't blind optimism, in my opinion, autopilot, even more than the drive train, is the 8th wonder of the world.

However, acclimation,is every devoted Tesla owner's destiny.

I will cheer Tesla on (even with it's shortcomings) as the engineers slave away at developing safer software. I don't envy the task, as the public and media takes a piss on autopilot, as they ignore every other iteration of assisted driving that hides under the umbrella of "thorough safety testing" when the truth is they are scrambling to catch up on the tech front, or to escape the lawyers and bureaucracy that limit their release. Who knows, maybe MobilEye will actually one day be allowed to deploy some of their technology in 2020. Maybe then some other automaker will actually result in consequential and meaningful use of autonomous driving technology.

Unless we are drinking the heavily spiked Kool Aid that @Bladerskb sells with his steering wheel green lights nanny complaining about your eye contact)

The rest of us, so disillusioned with lofty goals of Elon's timelines and without the patience and persistence to bother exploring the auspices of autopilot capability for long enough to let a firmware calibrate or to realize that the dissatisfaction among certain forum folks has more to do with what feels comfortable to them as they rest their hands too perfectly on each side of the steering wheel..... but I digress.

As we discuss various interesting aspects of autopilot, I am very much looking forward to reading these posts a year or two from now and seeing how this all played out. I don't know what will happen. But here's to version 9... and here's to Elon frickin' Musk.... a person who is staying in the fight far longer than I would have anticipated. If he can limit himself to less than 3 mental breakdowns per year in the next five years, I think these goals will eventually happen.

Yer Mad Max event should be epic; I envision some who will participate end to end, and some who will show up for one leg or another as their schedules allow. The media coverage, done right, will be insane. Two innocent words of preamble: Goat leggings .

Moving on, SC v3 should solidify Tesla’s strongest competitive advantage to date (infrastructure). It will take years for the competition to catch up in that regard. And until that happens, ALL competing cars are competing to be #2 - the best *local* car. Woo hoo. How... positively underwhelming.

I’m sure they’ll try to leverage the four, count ‘em, four CCS pedestals at Baker to tout the first competitive interstate trip. My lawn chair is ready and waiting to see that line of pissed off Dolt/Leaf customers during a busy Vegas weekend. As far as I’m concerned, every one of them is a future Tesla owner just as so many Android phone users are one and done, and over to iOS they go.

Meanwhile, I have no idea which competitor will produce the best AP challenge - each of the 8-10 major players offers something but nobody has the complete package yet. That arena is very much in play, imo.

I’d like to see Karparthy win that battle. We’ll see.