Camera miscalibrated

[solarev]

I noticed that the on-screen display that shows our Teslas and the ghosts of the other vehicles next to it does not show proper relative positioning of the adjacent cars. For example, if there is a car on my driver side blind spot, and the front of the car is where my side view mirror is, the display actually shows the front of the car as being further back. This happens both when stopped and driving and on both sides. This is very concerning as I am assuming the car uses the camera locations to make its autopilot decisions and also guidance for my lane changes etc. I wonder if this is a calibration issue.

I ask all of you to take a closer look at this next time you drive and see if this is happening with your car. Test it out both when you are driving and when you are stopped.

I will wait to get my informal survey/poll results from you guys/gals before I call service, since these forums are incredibly helpful and efficient for answers.

Lastly, I did do a search for this issue but didn't find topics quite like mine.

Thanks.

 

[coconutboy84]

It’s fine

 

[Hal420]

I've noticed some oddities with mine as well. i just got it a couple of weeks ago, and it does not seem to show cars in my blindspot very well. It also does not show bicycles at all and rarely shows pedestrians. Thinking about having it checked out at the dealer.

 

[Mr. Big]

I find the blind spot feature kinda useless. By the time the camera sees the car it’s too late to change lanes without running into it.

 

[Knightshade]

The display is not intended to be a 100% point-correct display of what the cameras see.

If it were it'd just be camera feeds pasted together.

See all the previous threads about "jumping" cars in adjacent lanes for example (though this is a lot better than it used to be on older software)

 

[DougInCT]

I have the same issues, and my car shows itself farther left in the lane than it actually is. That said, when on autopilot, it is mostly centered in the lane (physically), and I have not had any issues with it trying to merge into cars. It still puts the blinker on at incredibly annoying times, but it shows the car as red and changes lanes safely.

I am hoping that this gets tuned in better soon, but it's just a novelty, really. If I am piloting the car (not on AP), then I am not looking down to see if there is a car next to me. I am still a big fan of properly adjusted mirrors.

 

[sduck]

It's not a calibration issue. It's an issue with how the car is displaying things on the screen, which is still pretty much useless.

 

[ZappCatt]

Are you talking about the display of cars around you in the "car card" or an actual camera view that you are looking at?

I find that the car display is super scary, and would do more to push me away from FSD than push me to buy it.

While at a busy intersection, our entertainment is to watch the cars displayed appear, disappear and move around(while they are of course totally still).

I keep telling myself this is just the "visualization" part which is so wonky, and the actual analysis is rock solid/perfect.

 

[lolder]

The driving display is purely advisory and is not what the AP EAP FSD is using. It has a little lag and doesn't begin to show all the objects the car is identifying and tracking.

 

[DrMrEd]

I noted the other day, while driving, the display appears to be delayed about 1 second.

Not scientific at all, but while on AP and cruising in the left lane, I would kind of watch when my car front/rear aligned with vehicles being passed and it seemed to be consistently one second behind. Conceptually, this makes some sense, as the data has to be interpreted and rendered, which takes at least a little time - and it may be more like 1/2 or 3/4 of a second, tough to gauge those time scales.

So if anyone else is paying attention as OP asked, can you see if this is consistent?

It's all a little moot, as I think this screen is all secondary info - highlighting lanes, showing red cars that caused an alert, etc. and it's not meant to drive by. And AP is surely using the raw data rather than what's being rendered.

 

[Kitfox]

The screen being .75 seconds behind is consistent and not concerning to me. 10.2 improved it a touch too.

What concerned me* and I have not yet tested on 10.2 is lane line detection being the same delay, which when going into turns and coming out of turns on the highway causes the car to first understeer and then oversteer in the turn, creating some movement to the outside of the curve at the start of the turn and the inside of the curve at the end of it.

That being said, what shows on the car display is not the only thing used for blind spot detection. The car will alarm at you for blind spot worries for vehicles it detects that it doesn't show on the display. I've learned:
If the car is fine with the maneuver, look anyway.
If the car is worried, even if it doesn't seem to be showing anything, 99% of the time it's a legitimate worry.

* This only concerns me because when driving, I try to anticipate turns and take action in advance so I don't do that small swing at the start or the end. However, not only is this not a norm for most drivers, but I also know I have performance bias and if I review a video of manual curve driving, I do the same thing as the car. I'm just in control of the behavior and thinking (mistakenly) that I'm not doing it.