Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Fatal Flaw in AP2 Design

This site may earn commission on affiliate links.
Almost nobody is driving version 7 AP1. Version 8 AP1 fixed this and many other scenarios. Not perfect, but much better than it was in 7.
While it may be true in this case, with Tesla's fast update cycle some people like to silence critics by always saying "there is a new version since your experience - so this is probably fixed". The problem is, you don't know for sure whether something like this got fixed, or worse yet, a new even more dangerous behavior was added in the latest update. This by the way is another reason why I limit my AP1 usage - you think you learn the idiosyncrasies of the system, but then an update comes and you cannot rely on that knowledge to drive safely.
 
  • Like
Reactions: MS16
Don't know about the bump, but agreed Tesla starts hunting when it loses lane marking. Sunlight is dangerous as is cresting a hill. It will go beserk and kill you inside of a $150,000 coffin.

There needs to be dead reckoning tied in with GPS or some other fail safe like lead car tracking if lanes are not visible.

Maybe something like a HD Lidar based map for AP to rely on when it loses visual makings. This would allow a smoother transition back to the driver in uncertain conditions.
 
Almost nobody is driving version 7 AP1. Version 8 AP1 fixed this and many other scenarios. Not perfect, but much better than it was in 7.

Is the poster the author of the video I wonder? In any case, the video is nearing two years old. I only determined this by trying to watch it again from Youtube vs embedded.

These type of 'bugs', 'dangers', 'issues', have expiration times. It doesn't help Tesla and vehicle autonomy progress to rehash old stuff that is no longer relevant today.
 
  • Like
Reactions: NerdUno and _jal_
@bhzmark: I didn't say anything about "cresting hills." I said I drove over a damn bridge. Do I have to turn off AP2 at hills, bridges, overpasses, overhanging trees, new lanes, sunny days, sunsets, and whenever it rains? Really? Speaking of "little intelligence," your edge cases speak volumes about where the little intelligence lies. If I had a drill that always veered to the left to bore a hole in my finger, sooner or later I'd realize that my drill was a POS, and at least I would try to warn others instead of making up excuses for the f*cking drill.

You said the car was hunting for lane lines. Meaning the car was in a position where it could not see the lanes (aka cresting a hill). Keep your hand on the wheel and pay attention to where you are going. After driving with the system you should be able to understand its limits and adjust accordingly.


it requires more mental effort than driving by yourself where at least you're not worried about sudden swerve into oncoming traffic unless of course you suffer from seizures. At lest the driving instructors pay get paid for it, while Tesla drivers get to pay for it. Let's just hope insurance companies don't figure this out and require people using AutoPilot to get training and pay the same insurance as driving instructors on their teaching vehicles.

It definitely does not require more mental effort than driving yourself. If you know the limits of the system and generally pay attention to the road, AP can dramatically reduce fatigue on long distance trips. I just did 700 miles in AP1. About 100 or so miles were in 1-lane construction areas with heavy traffic. Did I use autosteer in the construction zones? No, because the lane markings were poor, so I just used TACC and it worked perfectly.
 
You said the car was hunting for lane lines. Meaning the car was in a position where it could not see the lanes (aka cresting a hill). Keep your hand on the wheel and pay attention to where you are going. After driving with the system you should be able to understand its limits and adjust accordingly.




It definitely does not require more mental effort than driving yourself. If you know the limits of the system and generally pay attention to the road, AP can dramatically reduce fatigue on long distance trips. I just did 700 miles in AP1. About 100 or so miles were in 1-lane construction areas with heavy traffic. Did I use autosteer in the construction zones? No, because the lane markings were poor, so I just used TACC and it worked perfectly.

Well, yes, it does require more mental effort because it requires you to be able to take over and correct in a fraction of a second, which is insufficient to gain complete situational awareness, therefore it follows you must always be as aware of your surroundings as if you were driving yourself or more. Think driving instructor, does it take more or less mental effort to drive with a student driver? After all, most students can keep the car in the lane probably better than Tesla AP (1 or 2).

Now, just because the AP fine print requires it, doesn't mean you can't get away with less. Lots if people do it and some eventually end up in accidents. Statistically speaking there will be people who will not get into accidents while completely trusting AP too. Just because someone survived Russian Roulette doesn't make it a safe game to play.
 
It definitely does not require more mental effort than driving yourself.

But there are times when it does. When AP suddenly decides to swerve for whatever reason (truck lust, faded markings, whatever) it does so with no warning and the driver has to "fight" the car to get it back in the center of the lane.

On a vacant stretch of highway this may be no big deal but in California lane splitting by motorcycles is legal. That means you share a lane with a passing motorcycle. When AP suddenly dives for the wall or starts lusting for the truck next to you and a motorcycle is screaming up to shoot the gap, it does require more mental effort as you fight the car while trying to avoid the motorcycle.
 
Well, yes, it does require more mental effort because it requires you to be able to take over and correct in a fraction of a second, which is insufficient to gain complete situational awareness, therefore it follows you must always be as aware of your surroundings as if you were driving yourself or more. Think driving instructor, does it take more or less mental effort to drive with a student driver? After all, most students can keep the car in the lane probably better than Tesla AP (1 or 2).

Now, just because the AP fine print requires it, doesn't mean you can't get away with less. Lots if people do it and some eventually end up in accidents. Statistically speaking there will be people who will not get into accidents while completely trusting AP too. Just because someone survived Russian Roulette doesn't make it a safe game to play.

From my own experience, while it's true that you need to be ready to take over, the situations where a failure will occur are pretty consistent. So I pay close attention and am at the ready on tight turns, cresting a big hill, etc. but can relax pretty well during the long stretches of straight-ish freeway or in slow moving traffic. That might seem somewhat limited, but it provides good coverage for most of the times where I'd want such an ability to relax. I generally don't drive wide open, curvy roads for hours at a time.

Also, "Lots if people do it and some eventually end up in accidents" can be said for every possible activity any human does. How important that observation is depends a lot on the relative values of "lots" and "some". In this case, so far as I know, the number of fatalities due to autopilot, world wide, is 1. Comparing that with Russian Roulette, where the chances of fatality are 1:6 is pretty misleading.
 
  • Like
Reactions: VikH
Since December, 2016, we have logged over 6,000 miles on our Model S P90D, and well over half of those miles have been driven using AP2, warts and all. The good news is it's gotten better. The bad news is it still will kill you without a moment's notice if you're not extremely careful. We decided to start a new thread about something which I, as a software developer, have regularly observed to be a fatal flaw (quite literally) in the current AP2 design.

Here is an example. We have a stretch of interstate highway that has a bridge with a slight lip on both ends. This lip provides a sufficient dip in the highway surface that, at 60 MPH, there is a noticeable bounce by the car when you leave the bridge surface. Despite excellent lane markings, AP2 switches off instantly every time, but it does it in a way that is extremely dangerous. Whenever the cameras detect a change in direction or disappearance of the lane markings, the car immediately swerves presumably to find another lane marking. When the nose of the car is elevated even slightly, it loses sight of the lane markings. When you couple the AP2 disengagement with the swerving that ensues from losing track of the lane markings, it invariably sends the car careening toward another lane of traffic regardless of whether there are vehicles beside you or not. Dangerous doesn't begin to describe it.

As a developer, it would seem to me the smarter and safer design would be to continue on in the previous direction when lane markings disappear at least until the driver can take over. This is especially true if there is a leading car for the Tesla to follow. As it happens, on this particular interstate, there is almost always a car immediately in front of you. And there is radar on the front of the car that can easily react to any unexpected condition.

Just curious. Have others experienced this?
exactly!! if lane markings disappear, swerving suddenly is not the right thing to do! Scares the hell out of the passengers! My sister refuses to use it now after several such sudden jerky motions on a crowded tight stretch of freeway.

AP2 is currently best when the traffic is light and the road is a nicely paved asphault. Not like our Socal roads where traffic is heavy, and ppl drive 80 and the lane markings are haphazard or wiped and repainted several dozen times.

Hopefully elons tweet of 5/21 promises real improvement from 17.17.17 (current version)

My biggest disappointment in this car is the TACC Cruise control - how the hell does the car slow down over a freeway underpass (elevated freeway, road under) - (not an overbridge above the freeway or an overhead sign even? but yeah I have had lots of those). Other's have figured this tech out - 2007 infiniti, 2014 BMW 2015 Lexus - I've driven all these TACC cars with never an unexpected slowdown! Even a one on a trip is one too many. It's not what you'd expect on a 90K+ car.

This is still "experimental software" and I'm suprised they made us pay $3k for it. There ought to be a disclaimer on every time the car is started.
 
Last edited:
  • Helpful
  • Like
Reactions: MS16 and NerdUno
At 6:45 PM tonight I was driving on the CA 91 freeway westbound through Corona on the new express lanes. While AP 2 has been flawless at night with the new lane markers, it's a disaster at sunset. All the lane marker locations used during construction are visible, and AP 2 gets confused. What the driver sees looks like endless tire skidding where the sandblasters removed the old paint lines. Since they didn't cover the new concrete with pavement in Corona, the result is a mess. The original express lanes were paved earlier in the year, and AP 2 works fine on that section at sunset.

With zero traffic at the time, I could allow for errors. Twice the 2017 MS 90D steered left across the lane markers. I would correct the steering and restart AP 2 when centered in the right lane. Time after time the software was attempting to steer using the wavy construction leftovers instead of concentrating on the double paint line on the right side. This is where Tesla should consider using the rear camera to provide extra information for lane centering verification. If the driver visibility is challenged, don't expect AP to steer accurately. Hopefully Tesla will find a solution soon to make autonomous driving a reality.

what version were u on? I've had it do fairly well thru that segment on AP2 17.17.17 but there are still times I have to intervene.
 
exactly!! if lane markings disappear, swerving suddenly is not the right thing to do! Scares the hell out of the passengers! My sister refuses to use it now after several such sudden jerky motions on a crowded tight stretch of freeway.

AP2 is currently best when the traffic is light and the road is a nicely paved asphault. Not like our Socal roads where traffic is heavy, and ppl drive 80 and the lane markings are haphazard or wiped and repainted several dozen times.

Hopefully elons tweet of 5/21 promises real improvement from 17.17.17 (current version)

My biggest disappointment in this car is the TACC Cruise control - how the hell does the car slow down over an overpass (not underpass or an overhead sign even? but yeah I have had lots of those). Other's have figured this tech out - 2007 infiniti, 2014 BMW 2015 Lexus - I've driven all these TACC cars with never an unexpected slowdown! Even a one on a trip is one too many. It's not what you'd expect on a 90K+ car.

This is still "experimental software" and I'm suprised they made us pay $3k for it. There ought to be a disclaimer on every time the car is started.
Context on the overpass logic. Tesla added this after the Florida accident where someone was killed by hitting a truck that was too high.
A Tragic Loss
Now the radar is used to detect such higher objects, which leads to false positives during overpasses. These are pruned out using a geocoded whitelist.
Upgrading Autopilot: Seeing the World in Radar
 
  • Informative
Reactions: bhzmark
Context on the overpass logic. Tesla added this after the Florida accident where someone was killed by hitting a truck that was too high.
A Tragic Loss
Now the radar is used to detect such higher objects, which leads to false positives during overpasses. These are pruned out using a geocoded whitelist.
Upgrading Autopilot: Seeing the World in Radar

does using report a problem work? Do I just say report a problem or provide some details?
 
Almost nobody is driving version 7 AP1. Version 8 AP1 fixed this and many other scenarios. Not perfect, but much better than it was in 7.

I have AP1 with 8.1 and I can attest that the car still swerves when it loses a the line due to a hill crest. I don't know if it'll actually swerve into oncoming traffic or just threatens to.
I remember when I was leaning to drive many years ago in the hills above Santa Barbara that I would get very nervous going over a crest where I couldn't see what was on the other side. Same with sharp turns. I takes a fair amount of faith to assume that the road will continue safely when you can't see it. I don't remember ever feeling a need to swerve into oncoming traffic in those situations though!

The reason it steer to the side, is because the camera sees the straight lines as curves.
Please elaborate.
 
  • Funny
Reactions: NerdUno
From my own experience, while it's true that you need to be ready to take over, the situations where a failure will occur are pretty consistent. So I pay close attention and am at the ready on tight turns, cresting a big hill, etc. but can relax pretty well during the long stretches of straight-ish freeway or in slow moving traffic. That might seem somewhat limited, but it provides good coverage for most of the times where I'd want such an ability to relax. I generally don't drive wide open, curvy roads for hours at a time.
So how many miles and hours of driving do you feel you require to learn all the idiosyncrasies, shortcomings and failure modes of each new software update? During the time you are learning/re-testing the latest software, do you also pay less attention than usual driving? If yes, you are taking risks (no, not Russian Roulette risks ;)) - even though chances are you may never have a serious problem. I used to be like you, as an engineer I thought I could figure out what the AP1 limitation were, and I was right 99% of the time. Worse yet, AP1 works really well 99.9% of the time if driven in areas where a lot of Tesla's drive (I drive in such an area, when I went of a cross country trip last year though I realized how badly it drives in rural areas off the main highways where there are no Tesla's, even on clearly marked roads) which lead me to trust it and actually pay less attention that I should have (hard to keep the brain engaged when the car drives itself so well, AP2 not there yet from what I hear but it will get there). All was well until one time I ran into a new behavior while thinking AP should be ok so paying less attention and almost ended up colliding with a semi. It was then I decided I like my health and life more than the convenience of driving on AP. I didn't even get EAP on our AP2. Btw, even when I thought I had AP all figured out, I always felt that unless one has a technical mind, one would have a hard time predicting where AP will lose its way and do something dangerous.

Also, "Lots if people do it and some eventually end up in accidents" can be said for every possible activity any human does.
True, unfortunately I don't have stats (Tesla does not release those). Insurance companies do, but it takes time before they are public. We do know a number of insurance companies have been hiking the Tesla rates recently, read into this what you will.

How important that observation is depends a lot on the relative values of "lots" and "some". In this case, so far as I know, the number of fatalities due to autopilot, world wide, is 1.
Accident does not equal fatality. I would prefer to avoid accidents, even if they don't kill me.

Comparing that with Russian Roulette, where the chances of fatality are 1:6 is pretty misleading.
I never said driving with AP is like playing Russian Roulette. I was using the game to illustrate the fallacy of "lots of people do something and they are fine, so must be safe " used by a lot of people in this thread - 5 out of 6 walk away unharmed from a game of Russian Roulette, yet nobody will claim it's a safe activity to engage in. Kind of like most people get lost with "if A then B" doesn't mean "if B, then A" explanation, but they quickly get "if it's a dog it has 4 legs" doesn't mean "if it has 4 legs it's a dog".:)
 
  • Like
Reactions: NerdUno
what version were u on? I've had it do fairly well thru that segment on AP2 17.17.17 but there are still times I have to intervene.
The version is 17.17.17. The problem was the sunset and how the sunlight made it difficult to see the lane markers. The old lane markers leftover from construction weave a pattern of arcs across the west bound express lanes. The best solution would be to pave the concrete.
 
  • Like
Reactions: cybergates
So how many miles and hours of driving do you feel you require to learn all the idiosyncrasies, shortcomings and failure modes of each new software update? During the time you are learning/re-testing the latest software, do you also pay less attention than usual driving? If yes, you are taking risks (no, not Russian Roulette risks ;)) - even though chances are you may never have a serious problem. I used to be like you, as an engineer I thought I could figure out what the AP1 limitation were, and I was right 99% of the time. Worse yet, AP1 works really well 99.9% of the time if driven in areas where a lot of Tesla's drive (I drive in such an area, when I went of a cross country trip last year though I realized how badly it drives in rural areas off the main highways where there are no Tesla's, even on clearly marked roads) which lead me to trust it and actually pay less attention that I should have (hard to keep the brain engaged when the car drives itself so well, AP2 not there yet from what I hear but it will get there). All was well until one time I ran into a new behavior while thinking AP should be ok so paying less attention and almost ended up colliding with a semi. It was then I decided I like my health and life more than the convenience of driving on AP. I didn't even get EAP on our AP2. Btw, even when I thought I had AP all figured out, I always felt that unless one has a technical mind, one would have a hard time predicting where AP will lose its way and do something dangerous.


True, unfortunately I don't have stats (Tesla does not release those). Insurance companies do, but it takes time before they are public. We do know a number of insurance companies have been hiking the Tesla rates recently, read into this what you will.


Accident does not equal fatality. I would prefer to avoid accidents, even if they don't kill me.


I never said driving with AP is like playing Russian Roulette. I was using the game to illustrate the fallacy of "lots of people do something and they are fine, so must be safe " used by a lot of people in this thread - 5 out of 6 walk away unharmed from a game of Russian Roulette, yet nobody will claim it's a safe activity to engage in. Kind of like most people get lost with "if A then B" doesn't mean "if B, then A" explanation, but they quickly get "if it's a dog it has 4 legs" doesn't mean "if it has 4 legs it's a dog".:)


Outside of the ramp up to autosteer actually functioning at freeway speeds, I haven't really needed to change my behavior release to release. I generally incorporate anything I've ever even heard could be problematic(I've never experienced truck lust, and it doesn't sound like it's a thing for AP 2 at this point, but I'm still cautious while passing semi's). But, from the beginning, long stretches of mostly-straight freeway(e.g. most of the 5 down to San Diego, a lot of the 101/880 up here, etc) have been pretty safe. I'll still pay attention at a high level for anything crazy(a good example happened yesterday, with a car stopped on the left shoulder, sticking out halfway into the fast lane. I took over before even getting to it), but such events tend to be crazy enough that I can see them coming well ahead of time, especially if I just stay in one of the interior lanes on a 3+ lane freeway.
 
  • Like
  • Informative
Reactions: NerdUno and VikH