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

Dashcam - The Missing 3 Seconds

This site may earn commission on affiliate links.
I've complained about his before and I'm about to do it again...

Here is a frame from the dashcam as I approach a turn lane:
Pic1.jpg


In this frame the car is starting to enter the turn lane. It is the last frame in a 1-minute clip.
Pic2.jpg


It gets almost half way into the turn lane and then swerves back. But the first frame in the next clip is this:
Pic3.jpg


So I can't show you (or Tesla) how it mistakenly tried to enter the turn lane because there is a 3-second gap in all the 1-minute clips saved by the dashcam and this event just happened to coincide with a gap. I'm sure the reason isn't with the SD card I'm using as I've had this problem with a USB drive and with an SSD.

With my luck, if the day ever comes where I really really need a dashcam, it will be during those 3 seconds. And if your clips do not have this 3-second gap, please speak up. I doubt if some Teslas do have it and some don't but I can only hope that there is *something* I can do to fix it.
 
I installed a 2 channel Blackvue 900 UHD/4K in my car. The quality of video with the Tesla cams is not the best and you often can’t even read the license plates. Just look at your photos above. No way to read the trucks license plate. With the Blackvue it is crystal clear. Also no skips in between files and you can change to 3 minute clips. EVTuning has these on sale for $399 for Black Friday at the moment if interested.
 
That truck might be a little too far to get a plate number but you can't tell from my picture. It's a screen grab from a reduced size video. Far less resolution than an extracted image direct from the video file.

I've tried to work around this 3-minute gap by installing a VanTrue Pro dashcam to the overhead glass, right next to the dome light. It has both a forward and rear facing camera built into the one unit. I wanted it there so it had a view out the front & rear glass and at the same time record what is on the LCD. It would have worked just fine if only that camera had a decent dynamic range. As it is the view out the windshield is either too bright to see anything or if adjusted ok for that then you cant see what's inside the car, too dim.
 
So I can't show you (or Tesla) how it mistakenly tried to enter the turn lane

No mistakes are possible in this sort of situation.

For safety, I’d recommend that you master the use of AP on freeways before “graduating” to attempting to use AP outside of its intended design domain (where “unexpected” behavior is, because of its lack of any ability to handle anything but the simplest cases, always expected).
 
  • Like
Reactions: Phlier
You guys can give Tesla a pass on this but I won't. My complaint wasn't so much about it trying to enter the turn lane and swerving back (like it occasionally does on "approved" highways), but the fact that there is a missing 3-second gap in the dashcam recordings. So how about trying to not be so quick to preach to me about where I should use AP and where not, and pay attention to the primary subject.
 
You guys can give Tesla a pass on this but I won't. My complaint wasn't so much about it trying to enter the turn lane and swerving back (like it occasionally does on "approved" highways), but the fact that there is a missing 3-second gap in the dashcam recordings. So how about trying to not be so quick to preach to me about where I should use AP and where not, and pay attention to the primary subject.

I agree there should not be a three second gap in recordings.
 
You guys can give Tesla a pass on this but I won't.

To be clear there's 2 entirely unrelated problems that came up in your original post.

The first is you're using AP incorrectly and (just as incorrectly) suggesting it's a Tesla problem that needs to be reported to them and thus are trying to capture footage of what turns out to actually be user error.

Pointing that out isn't giving Tesla a pass- it's pointing out where the actual error is.



The second issue is the 3 second gap- on which I entirely agree with you that shouldn't happen and is an issue Tesla would need to address. Never noticed it on my footage but also don't think I ever looked frame by frame like that.

First thing to determine is is everyone seeing this, or just you... if it's everyone it's a Tesla software problem they'll need to address generally. If it's just you it could be a car-specific HW issue, or some SW glitching just on your car.

Some things you'd probably want to know when reporting the issue to Tesla though (in addition to "is everyone else getting this or just you") would be if it's happening on all 4 camera streams or just one, does it happen on every single clip for that (or all) cameras, and does it happen both in motion and when parked?
 
  • Like
Reactions: jjrandorin
I've complained about his before and I'm about to do it again...

Here is a frame from the dashcam as I approach a turn lane:
View attachment 478223

In this frame the car is starting to enter the turn lane. It is the last frame in a 1-minute clip.
View attachment 478225

It gets almost half way into the turn lane and then swerves back. But the first frame in the next clip is this:
View attachment 478226

So I can't show you (or Tesla) how it mistakenly tried to enter the turn lane because there is a 3-second gap in all the 1-minute clips saved by the dashcam and this event just happened to coincide with a gap. I'm sure the reason isn't with the SD card I'm using as I've had this problem with a USB drive and with an SSD.

With my luck, if the day ever comes where I really really need a dashcam, it will be during those 3 seconds. And if your clips do not have this 3-second gap, please speak up. I doubt if some Teslas do have it and some don't but I can only hope that there is *something* I can do to fix it.
Okay, is the issue the dash cam, cause that's what's in the subject line, or is it the AP error/correction? If it's the dash cam, then I'd try a different USB stick, to see if that makes any difference.

As for the AP error, is it repeatable, in that spot? If you look at your display, what does it show when you pass this spot? Does it show the double yellow gradually curving way off ahead, because I get the sense that the car does some visual smoothing, in order to minimize jitters and having the car move too much in its lane making the passengers carsick. Also, that dashed line seems abnormal. Like the white dashes are too short. Maybe, they are too short, and so the computer hesitates, until it understands that that's a dashed white line.

Screenshot 2019-11-18 13.12.19.jpg


The dashed lines spacing is too far. Should be as long as the dashes themselves.
 
Last edited:
Okay, is the issue the dash cam, cause that's what's in the subject line, or is it the AP error/correction? If it's the dash cam, then I'd try a different USB stick, to see if that makes any difference.

OP already covered that not being the issue-

Original post in thread said:
I'm sure the reason isn't with the SD card I'm using as I've had this problem with a USB drive and with an SSD.


As for the AP error

That's not an AP error- it's a user error, expecting AP to work someplace explicitly outside its intended operational domain.
 
I'm interested in understanding why AP wouldn't detect the dashed line.

But it did detect the dashed line!

It's not really the topic at issue here, but it's not really surprising at all - it did detect the dashed line as the OP said. But AP would want to stick with the yellow line, as that's probably the right thing to do, when you're on the "freeway," like the OP was (as far as AP was concerned). So it tried and then decided maybe it wasn't the right thing. Or something like that. I'm sure there are any number of possible explanations for how the logic decided to do this. But it really doesn't matter; it's of no consequence.

I would imagine in freeway situations they may even have to label tricky similar situations (that are actually on a freeway) to prevent similar behavior, but it's hard to know what they are doing under the hood when using AP. I know they don't like to label things as it deviates from an "autonomous" approach, but they probably do.
 
I am not sure why it is a user error. I use AP all the time on non-restricted access roads and I typically don't get this behavior, although as someone noticed, the broken lines do not look standard, so was probably following the yellow. Still, not a reason to NOT use AP.

When I do use AP on local roads, I am constantly vigilant to the car doing unexpected things, but I also like testing the limits of what situations the AP can handle, especially on some of the edge cases (without my wife in the car). I'd like to believe that Tesla is incorporating this into its neural network so the algorithms can be improved in the future.

There is soon going to be the ability where AP will stop at stop signs and lights on its own, so it better be able to handle a lot of those types of situations without the swerving.
 
I am not sure why it is a user error.

Because the system is explicitly not intended for use on non-restricted roads like that.

I use AP all the time on non-restricted access roads

People drive drunk all the time too-most of them don't even crash! That doesn't mean it's intended behavior.


When I do use AP on local roads, I am constantly vigilant to the car doing unexpected things, but I also like testing the limits of what situations the AP can handle, especially on some of the edge cases

Those aren't "edge cases"

You aren't "testing the limits" of the system

You're using the system someplace it's entirely not intended to work. At all.

The fact it usually manages to not screw up in places that it's not even meant to work is neat, but any errors you do run into are entirely user error


AP fundamentally operates today with these assumptions:

Road access is restricted by things like on/off ramps. So it has no idea what a turn lane is

All traffic is going the same direction, and there's no cross traffic. Which is why folks complaining about "phantom braking" when oncoming cars turn across their lane are folks who don't understand the operational domain of the system and that the problem is user error, not APs error.



Model 3 owners manual said:
Warning: Autosteer is intended for use only on highways and limited-access roads with a fully attentive driver... Do not use Autosteer on city streets, in construction zones, or in areas where bicyclists or pedestrians may be present



Every objection beyond that amounts to asking "Why doesn't it work perfectly in places it's explicitly told me it's not supposed to??"




There is soon going to be the ability where AP will stop at stop signs and lights on its own,


No, there isn't.

Only FSD vehicles will do that. And will require both HW3 and a completely different set of software than what is running today (much larger neural nets).

Nothing you see AP doing today on local roads tells you anything useful about how it'll work once it's actually intended to using an entirely different and much more complex software load.
 
Last edited:
I've complained about his before and I'm about to do it again...

Here is a frame from the dashcam as I approach a turn lane:
View attachment 478223

In this frame the car is starting to enter the turn lane. It is the last frame in a 1-minute clip.
View attachment 478225

It gets almost half way into the turn lane and then swerves back. But the first frame in the next clip is this:
View attachment 478226

So I can't show you (or Tesla) how it mistakenly tried to enter the turn lane because there is a 3-second gap in all the 1-minute clips saved by the dashcam and this event just happened to coincide with a gap. I'm sure the reason isn't with the SD card I'm using as I've had this problem with a USB drive and with an SSD.

With my luck, if the day ever comes where I really really need a dashcam, it will be during those 3 seconds. And if your clips do not have this 3-second gap, please speak up. I doubt if some Teslas do have it and some don't but I can only hope that there is *something* I can do to fix it.

I'm using a 350 megabyte per second ssd, and I have this issue on a raven model s. When I looked into it, it seems to be caused by the time it takes to write the four one-minute videos at the maximum speed of the usb2 interface(~30MB/s). Apparently they don't start recording the next set of videos until they've completed the writing of the current videos. Why?
 
Because the system is explicitly not intended for use on non-restricted roads like that.

Knightshade, I appreciate the detailed blow-by-blow response.

I am not saying that the system is approved for non-freeway driving and I am not expecting the system to be perfect, but let's be clear, despite Tesla providing a CYA statement for liability, I would hazard that almost 100% of owners have used AP on non-restricted access roads. It works, and in my opinion, AP works great in most situations. I would be surprised if even you have not used AP on other than restricted access roads.

People drive drunk all the time too-most of them don't even crash! That doesn't mean it's intended behavior.

That's nice - I assume you drive the exact speed limit as well.

Those aren't "edge cases"

You aren't "testing the limits" of the system

You're using the system someplace it's entirely not intended to work. At all.

We will have to agree to disagree. The system is learning, and in my view, the oddball situation IS an edge case. As I said, Tesla is not going to give the green light, but the system IS designed to work in those situations. Otherwise, the car would not be able to operate well at all.

Now maybe YMMV with your roads. I don't know where you live, but I know the roads I am driving, and when they are well marked, the car handles almost all situations perfectly. When they are not or if there are some deviations, I will either take control or be ready to take control. I get your point that it won't handle everything as well, but it is the same on the restricted access roads when there is construction or bad or double markings. The environmental situations can be similar.

Every objection beyond that amounts to asking "Why doesn't it work perfectly in places it's explicitly told me it's not supposed to??"

I do not hold the AP to be perfect, even on the freeway, but I can see the improvements and it is WAY better than it was six months ago. I am not saying that we can expect AP to work flawlessly yet, but my point is that it is not USER error to use autopilot. It is wrong to have expected it to work 100% in all cases, but it is worth examining what it can do well and what it can't do so well, yet.

Only FSD vehicles will do that [stop at lights and signs]. And will require both HW3 and a completely different set of software than what is running today (much larger neural nets).

I am not sure what you are arguing - Elon has said that there will be early access this year for that functionality. Yes, it will require new software of course, but it is not going to be night and day from most of the current AP functionality. I have HW3 and while I won't be in the first launch, I hope to be soon after and I will be testing it as much as I can.
 
If they don't expect us to use it on non-limited access roads, why do they provide traffic light warnings when on autopilot? I haven't seen them, but are there traffic lights on limited access highways? Not on the on or off ramps but the highways themselves.
 
If they don't expect us to use it on non-limited access roads, why do they provide traffic light warnings when on autopilot?

There's traffic lights on the on-ramps of quite a few limited access roads FWIW- remember nav on autopilot is explicitly intended for ON RAMP to off ramp...so those would absolutely be in the operational domain of the system unlike local roads.

ramplight.jpg
 
Knightshade, I appreciate the detailed blow-by-blow response.

I am not saying that the system is approved for non-freeway driving and I am not expecting the system to be perfect, but let's be clear, despite Tesla providing a CYA statement for liability, I would hazard that almost 100% of owners have used AP on non-restricted access roads. It works, and in my opinion, AP works great in most situations.

It fails to fail as often as you might expect certainly. And if you're paying close enough attention that doesn't usually cause an issue.

That doesn't change the fact it's not supposed to be or intended to be used there...and when it does fail that's the users fault, not Teslas.

So the OP suggesting this is something he wants to "report" to Tesla is nonsensical, since he's basically going to be telling them "Hey- this didn't work in a situation you already told me it isn't meant to work..."


Imagine you contacted Michelin and said "Hey, I want to report that these summer performance tires didn't work very well when I drove them in the snow..."

They'd laugh you off the call.

Same deal here.

We will have to agree to disagree.

You're not disagreeing with me.

You're disagreeing with the literal text of the owners manual from the folks who built the system.

The system is learning

Autopilot does not learn in real time. Behavior only changes with updates from Tesla.

And the current software is written for hardware that explicitly can not handle local driving (hence the reason HW3 even exists though the much more powerful software isn't rolled out to the public fleet yet).


As I said, Tesla is not going to give the green light, but the system IS designed to work in those situations.

Again- Tesla explicitly says you are wrong right in the manual.


Otherwise, the car would not be able to operate well at all.

"Works most of the time except all the cases where it doesn't" isn't really working well.

See again the myriad threads about "AP TRIED TO KILL ME WHEN I USED IT SOMEPLACE I WAS EXPLICITLY TOLD NOT TO!"

That's the drivers fault, not Teslas.


Now maybe YMMV with your roads. I don't know where you live, but I know the roads I am driving, and when they are well marked, the car handles almost all situations perfectly.

Well no....

It handles all situations that happen to be just like a divided highway perfectly.

Following well marked lines. Adjusting follow speed with TACC. Those work great because you're using them in a situation that looks like a highway to the car.

But come up on a turn lane and you get the original post.

Have oncoming traffic turn in front of you you get people bitching about WHY IS THE CAR BRAKING!

Easy- because it doesn't know WTF oncoming traffic is

It thinks there's suddenly a car you're about to slam into that just came into your lane.

In the worst cast, you're not paying attention, and you die from your user error. As at least 2 Tesla drivers have done by using the system in places with cross-traffic.

Where, again, it's explicitly not meant to be used


I do not hold the AP to be perfect, even on the freeway, but I can see the improvements and it is WAY better than it was six months ago.

Agreed... it handles a ton of highway situations either better (or at all) that it didn't 6 months ago.


I am not saying that we can expect AP to work flawlessly yet, but my point is that it is not USER error to use autopilot.

It's user error to use it outside its operational domain and be surprised if it doesn't "work correctly"

Because it's not intended to



I am not sure what you are arguing - Elon has said that there will be early access this year for that functionality. Yes, it will require new software of course, but it is not going to be night and day from most of the current AP functionality. .

Yes it absolutely will be.


Today all public fleet cars are running the 2.x code, even HW3 cars running it in emulation. HW3 native code will be MASSIVELY larger and more powerful.


but don't take my word for it-here's Andrej Karpathy, the guy in charge of all this stuff at Tesla, from autonomy day-I bolded the two relevant bits.


Andrej Karpathy said:
“The team is incredibly excited about the upcoming upgrade for the Autopilot computer. This upgrade allows us to not just run the current neural networks faster, but more importantly, it would allow us to deploy much larger computational and more expansive networks to the fleet. The reason this is important is that it is a common finding in the industry that as you make the networks bigger, the accuracy of their prediction increases with the added capacity. Now we’re currently at a place where we’ve trained large networks that work very well, but are not able to deploy them to the fleet due to computational constraints. So all of these will change with the next iteration of the hardware, and it’s a massive step improvement in computing capability, and the team is incredibly excited to get these networks out there.”


TLDR version- they have a vastly more capable and advanced software suite in the lab and have some time- that HW2.x simply can not run. HW3 (once they go to native HW3 software) will usher in massive improvements.

It will not be tiny fractional ones like you see month to month with current software updates written for the 2.x HW.
.

 
  • Like
Reactions: MP3Mike