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

Annoying FSD/autopilot behavior with on-ramps

This site may earn commission on affiliate links.
Simply because Tesla is designed to stay in the center of two white lines ie the blue line that you drew.
Yep, they try to keep it simple in spite of how confusing it is to anyone on the roadways. The NN ain't getting it done when it makes so many irrational generalizations. It makes one wonder why not fix all the little things that cause so much driver frustration? Or do they think the current design hasn't proven itself worthy of refinement? Things that make you go hmmm.
 
Last edited:
  • Like
Reactions: enemji
I get that the focus is on FSD, but there have been MANY AP updates pushed over the "years". One would think this would be easy to solve...

So this isnt a problem AT ALL for FSD Beta? Its been totally solved?

Yes, this is completely solved in FSD Beta. I drive down an "expressway" which has similar merge lanes every day and it stays centered in its lane every time.
 
  • Disagree
  • Like
Reactions: enemji and rpo
A.k.a. yet another bug.
No, it's more an example of how even 'simple' aspects of driving are more complex and nuanced than they appear at first glance.
Yep, they try to keep it simple in spite of how confusing it is to anyone on the roadways. The NN ain't getting it done when it makes so many irrational generalizations. It makes one wonder why not fix all the little things that cause so much driver frustration? Or do they think the current design hasn't proven itself worthy of refinement? Things that make you go hmmm.
This is part of AutoPilot. As I understand it, the neural net isn't involved. There are currently 2 different programs running simultaneously - AutoPilot that's used on limited access highways and FSDb that's used everywhere else.
 
  • Like
Reactions: 2101Guy
Like I said before, nobody told my M3LR that because it almost always centers itself both on freeways and on city streets. Which version of FSDb you on?
I'm on the latest version (10.69.2.3), but city streets code has been working fine for many months with merge ramps. On the freeway, it always bobs to the right like OP mentions. Maybe i'll try to make a video. I drive by merges like this all the time.
 
I'm on the latest version (10.69.2.3), but city streets code has been working fine for many months with merge ramps. On the freeway, it always bobs to the right like OP mentions. Maybe i'll try to make a video. I drive by merges like this all the time.
Is there a seam in the roadway on the city streets with these merge lanes?

The neural net does interpret road seams as lane separators (thankfully) if prevalent enough, so even on the highway it doesn't always do that.

And what is the length of the unmarked merge lane on the "city" streets? Is there enough time for it to actually move to the right?

If this is actually the case, I'd be even more disappointed since apparently they have the logic working to know when to center and when not to, and after all this time they are not willing to put that into the AP stack. I get that they are working towards the eventual goal of merging the stacks, and maybe this fix is not worthy of all the "researchy" things they probably like to be working on (and Elon hired them, and likely compensates them for), but this is a huge detractor for the brand and all those that do not have FSD, and unless the stack merge is imminent (which it's not), it should be addessed.
 
A.k.a. yet another bug.
No, it's more an example of how even 'simple' aspects of driving are more complex and nuanced than they appear at first glance

Bug: A bug is an unexpected problem with software or hardware. Typical problems are often the result of external interference with the program's performance that was not anticipated by the developer. Minor bugs can cause small problems like frozen screens or unexplained error messages that do not significantly affect usage.

Of course this is a bug. Didn't say it was major, easy, or a priority, or even a large safety issue. But by definition it is a failure of the software to act properly under the specified circumstances, as nearly all of the comments here would agree.
 
Bug: A bug is an unexpected problem with software or hardware. Typical problems are often the result of external interference with the program's performance that was not anticipated by the developer. Minor bugs can cause small problems like frozen screens or unexplained error messages that do not significantly affect usage.

Of course this is a bug. Didn't say it was major, easy, or a priority, or even a large safety issue. But by definition it is a failure of the software to act properly under the specified circumstances, as nearly all of the comments here would agree.
1_UH1ZkAzN4lLeWRDSggFq6Q.jpeg
 
No, it's more an example of how even 'simple' aspects of driving are more complex and nuanced than they appear at first glance.

This is part of AutoPilot. As I understand it, the neural net isn't involved. There are currently 2 different programs running simultaneously - AutoPilot that's used on limited access highways and FSDb that's used everywhere else.
Good point but I experience it whenever the road widens even on surface streets.

No doubt it's far more complex after peeling the onion. For now it feels like FSDb lacks depth to retain sufficient scenario details to make predictable rational decisions.

There seems to be more than a few things lacking. Hopefully the training computer (Dojo) will show huge improvement in training and or shine the light on possible NN architecture/Tesla pure vision shortcomings.
 
Bug: A bug is an unexpected problem with software or hardware. Typical problems are often the result of external interference with the program's performance that was not anticipated by the developer. Minor bugs can cause small problems like frozen screens or unexplained error messages that do not significantly affect usage.

Of course this is a bug. Didn't say it was major, easy, or a priority, or even a large safety issue. But by definition it is a failure of the software to act properly under the specified circumstances, as nearly all of the comments here would agree.
In this case the software (and hardware) are behaving exactly as designed - they are programmed to center within the lines. The problem is that the design apparently didn’t account for these scenarios. If you classify poor/inadequate design as a bug then yes, it’s a bug. I consider bugs to be cases where the software does something it wasn’t intentionally programmed to do.
 
Is there a seam in the roadway on the city streets with these merge lanes?

The neural net does interpret road seams as lane separators (thankfully) if prevalent enough, so even on the highway it doesn't always do that.

And what is the length of the unmarked merge lane on the "city" streets? Is there enough time for it to actually move to the right?

If this is actually the case, I'd be even more disappointed since apparently they have the logic working to know when to center and when not to, and after all this time they are not willing to put that into the AP stack. I get that they are working towards the eventual goal of merging the stacks, and maybe this fix is not worthy of all the "researchy" things they probably like to be working on (and Elon hired them, and likely compensates them for), but this is a huge detractor for the brand and all those that do not have FSD, and unless the stack merge is imminent (which it's not), it should be addessed.
No seams, but maybe shorter merges. Regardless, you can watch the blue line, it doesn't move right at all. Follows the left line straight through confidently. Just uploaded a video. I go through 3 merges here