Don't you think your evidence is a bit skewed if you are changing dev settings, disabling debug, etc? Where might one see this thread? If you know how to get around this "nag" wouldn't it be nice to share it? Maybe I'm just confused by how you aren't getting any nags. Not sure why some are getting the nag and others are not, also not sure what the percentage of yes and no's are. But there is defiantly some sort of new "timed nag" for some.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
7.1 AutoPilot Nag
- Thread starter boonedocks
- Start date
Ok thanks, will try doing some more testing. When you say back to normal, were your settings abnormal before and how so? Thought 1300 miles was a good amount of data, guess I need more. lol
By back to normal I mean disabling factory mode, re-enabling the +5 MPH restriction on back roads, turning off autopilot debug stats, turning off developer mode, etc.
Image from my other thread:
I've had the same result as wk057 (no timed nags) and my car has never been in dev mode.
Perhaps tesla introduced a way to update AP settings outside of the firmware update process?
I don't think I had a timed nag, and then a few days later I did. That's my perception, anyway.
I think I have it figured out.
<speculation>
It appears that the "nag" gets triggered on certain boundaries if the average confidence over that last interval was less than some threshold. Holding the wheel increases confidence. So, when you're driving along and the car is getting OK but below average input it's confidence level is low, and thus triggers the "nag." Anything could drive confidence down. From what I can tell from playing with dev mode, the amount of steering corrections needed over a short interval is factored in. So if the car makes a quick jerky correction, this lowers confidence. Shadows on the road lower confidence. Oncoming headlights or sun glare lowers confidence. And then the obvious things like crappy lane markings, bad maps, and higher speeds.
So, I think anyone who seems to be receiving a "timed" nag is actually experiencing lower confidence. It seems a big factor is speed relative to the speed limit. At +15 I got more nags than expected. At +10 I got none.
</speculation>
<speculation>
It appears that the "nag" gets triggered on certain boundaries if the average confidence over that last interval was less than some threshold. Holding the wheel increases confidence. So, when you're driving along and the car is getting OK but below average input it's confidence level is low, and thus triggers the "nag." Anything could drive confidence down. From what I can tell from playing with dev mode, the amount of steering corrections needed over a short interval is factored in. So if the car makes a quick jerky correction, this lowers confidence. Shadows on the road lower confidence. Oncoming headlights or sun glare lowers confidence. And then the obvious things like crappy lane markings, bad maps, and higher speeds.
So, I think anyone who seems to be receiving a "timed" nag is actually experiencing lower confidence. It seems a big factor is speed relative to the speed limit. At +15 I got more nags than expected. At +10 I got none.
</speculation>
SomeJoe7777
Marginally-Known Member
Hmmm ... perhaps not so fast. :smile:
This is a video I made on Sunday. Conditions:
- Bright sunny, but traveling northbound, so sun was not shining at the AP camera
- Ideal weather
- Well-marked lanes
- 4-lane divided highway
- 5 MPH under the speed limit (70 in a 75)
- No curves/turns at all
- No traffic
According to your analysis, this should be quite high confidence.
The video shows 5 cycles of the Hold Steering Wheel messages at exact intervals of 3 minutes and 2 seconds.
Now, I know for a fact that this doesn't always happen. I have another video I did on the same trip where I captured 2 cycles of the Hold Steering Wheel message at 5 minute intervals. (I have not edited and posted that one yet, will probably be done this weekend). In addition, I drove one segment of that trip where I had no messages at all for over 15 minutes. (I don't have that period on video).
There's a lot more going on here that we haven't discovered yet. The intervals of the Hold Steering Wheel message (or whether you will see them at all) seem to be dependent on multiple factors.
Last edited by a moderator:
Conditions up until right about where you disengaged by accident looked terrible. Look at your dash cam video. That's roughly what the autopilot camera can see as well. As a human I could barely make out any lane markings. Plus the huge contrast changes due to different pavement types all the time. Certainly killed that confidence meter there at the beginning. The one after you re-engaged... *shrugs* ... and the last one was back to terrible visibility for part of it.
Mythbusters would give it a "Plausible".
Try the same thing at night on the same stretch of road. $10 says it doesn't happen.
Last edited by a moderator:
I mentioned something similar on the day I got the update, just less scientifically
. When I drove what Tesla thought was a non-divided highway (it was divided), I had a nag every 60 seconds. After I held the wheel for 5 minutes, and then let go, my nag turned into 2 minutes, after that back down to 1 minute. I speculated that it "punishes bad behavior".
Not convinced it has to do with the +X speed setting. I was driving 55 in a 55, and I had timed nags on the 3-minute mark on a divided interstate highway.
SomeJoe7777
Marginally-Known Member
wk, relax. :smile: Don't get defensive, it's not a refutation of your analysis.
The concrete was lighter in color, and there was patchwork on the road surface with much darker-colored asphalt. Yes, this probably had an effect. How much? I have no idea ... we have no indication of the confidence level that AP is using internally. On the IC, there were no indications that AP was having any issues (e.g. no fading blue lines, etc.)
The point is that there is no simple answer to the conditions that the HSW message appears. This particular incident shows that under these conditions, a 3 minute timed HSW message was shown. Like I mentioned, I have another video I will post soon that shows a 5 minute HSW message cycle. And we've all experienced times where there is no message at all for quite long periods.
At least some of the factors that figure into this are:
- Whether you're following another vehicle or not.
- Conditions for the confidence level of AP.
- Speed.
- Road type? It appears that non-divided roads may have a shorter interval.
I don't know if I can try this particular stretch of road again any time soon. This is US Highway 77 going northbound out of Victoria, TX towards Hallettsville. A round trip from my house to this stretch of road and back is over 200 miles -- not something I can just go try real quick. But it would indeed be interesting to see AP's reaction at night where the lane markers would be more prominent.
Soolim
Member
To differentiate location lane marking triggered nag versus time nag, start your timing at different location to see if it is repeatable.
The nagging makes absolutely NO sense to me. My experience, driving on very bad roads for AP, is that when I drive on straight roads with clear markings, it tends to nag more often. I usually don't get nags when driving on horrible roads, with sharp turns and no lane markings. I don't get it - the car keeps going in AP when it completely loses any input. No car in front, and no lanes sidelines. It's driving completely blindly - but doesn't even hint that I should take over.
In my opinion the AP warnings and more importantly lack of warnings, are a disaster. When it works well, it nags you. But when it doesn't work? It won't tell you. It would me much more relaxing to use the system I knew I would get a warning immediately when it loses confidence. But it doesn't - when it loses track of cars and lanes, it just keeps going like nothing happening, while swerving onto oncoming traffic...
So my suggestion:
1. Completely remove the current implementation of the "Hold wheel" nag - it doesn't work properly, and makes no sense.
2. When it loses track of one side, show an alert. When it loses track of both sides, make an audible alert.
In my opinion the AP warnings and more importantly lack of warnings, are a disaster. When it works well, it nags you. But when it doesn't work? It won't tell you. It would me much more relaxing to use the system I knew I would get a warning immediately when it loses confidence. But it doesn't - when it loses track of cars and lanes, it just keeps going like nothing happening, while swerving onto oncoming traffic...
So my suggestion:
1. Completely remove the current implementation of the "Hold wheel" nag - it doesn't work properly, and makes no sense.
2. When it loses track of one side, show an alert. When it loses track of both sides, make an audible alert.
Super_Popular
Well-Known Member
I used AP for most of the day and I can honestly say that I love it!....the nag really is not that annoying and I am actually glad it has the nag....keeps me alert.
Mark Z
Active Member
Got 7.1 on Friday at the SC for Model X. Heard the nag for the first time. Now I know what it sounds like!
davidc18
Active Member
Have to agree that the 'new' system does not seem to follow any pattern that makes sense, unlike 7.0 which was very predictable.
Mark Z
Active Member
It seems to me that AP may be using the "hold steering wheel" nag to notify the driver of things that may have been ignored by AP in the past, but are now being "announced" using the nag. In other words, it's more sensitive to feedback from the sensors or the camera if the slightest thing is out of the ordinary.
The above comment is IMHO and has no basis of fact, only that more messages are occurring and I cannot imagine Tesla creating those messages without some "input" from the system.
UPDATE: I was completely off thinking about sensitivity. There is a way to keep the nag from happening! Read on for the Jan 24th drive report...
The above comment is IMHO and has no basis of fact, only that more messages are occurring and I cannot imagine Tesla creating those messages without some "input" from the system.
UPDATE: I was completely off thinking about sensitivity. There is a way to keep the nag from happening! Read on for the Jan 24th drive report...
Last edited:
Similar threads
