FSD v13 rolling out!

[2DVS4U]

I have noticed on my Cybertruck lately that there are no pictures of traffic cones and/or barricades like the used to be on autopilot in the beginning and I even used to see mailboxes and garbage cans on the screen (no more), the only reason I am concerned about this is because the amount of road construction that is slated in my area for 5 years and it has been a pita to keep intervening when it doesn't see them or identify them, I have noticed on the videos I watched this morning from Chuck Cook and Whole Mars Catalogue that there are no drawing/picture either, am I missing something or has this issue been addressed in the past?

Sponsored

 

[Dmayo305]

I have noticed on my Cybertruck lately that there are no pictures of traffic cones and/or barricades like the used to be on autopilot in the beginning and I even used to see mailboxes and garbage cans on the screen (no more), the only reason I am concerned about this is because the amount of road construction that is slated in my area for 5 years and it has been a pita to keep intervening when it doesn't see them or identify them, I have noticed on the videos I watched this morning from Chuck Cook and Whole Mars Catalogue that there are no drawing/picture either, am I missing something or has this issue been addressed in the past?

In the autopilot section of the settings menu, make sure to enable the enhanced display. I'll try to remember to take a picture of it later.

 

[stumby]

Tesla FSD 13 navigates parking lot then BACKS ITSELF INTO SUPERCHARGER STALL





Tesla FSD V13 Goes in Reverse! First Drive and Initial Impressions

 

[kappaknight]

I have noticed on my Cybertruck lately that there are no pictures of traffic cones and/or barricades like the used to be on autopilot in the beginning and I even used to see mailboxes and garbage cans on the screen (no more), the only reason I am concerned about this is because the amount of road construction that is slated in my area for 5 years and it has been a pita to keep intervening when it doesn't see them or identify them, I have noticed on the videos I watched this morning from Chuck Cook and Whole Mars Catalogue that there are no drawing/picture either, am I missing something or has this issue been addressed in the past?

It still sees them even if it doesn't show them.

 

[CTSoFL]

 

[2DVS4U]

In the autopilot section of the settings menu, make sure to enable the enhanced display. I'll try to remember to take a picture of it later.



I do see the Expanded full self driving visualization but it doesn’t seem to change it or bring back what we used to have.

 

[CYBERUN]

Looks like Cybertruck is not included in FSD v13.2 upgrade yet. We will have to wait for a while.

 

[2DVS4U]

It still sees them even if it doesn't show them.

I am sure it sees things but why did they remove it from the screen and why does my Cybertruck ignore the merging cones that it used to display and doesn't anymore is my concern.

 

[Dmayo305]

I am sure it sees things but why did they remove it from the screen and why does my Cybertruck ignore the merging cones that it used to display and doesn't anymore is my concern.

Have you done a scroll wheel reset? Hold down both scroll wheels until the screen shuts off and then it'll reboot.

 

[Crissa]

I am sure it sees things but why did they remove it from the screen and why does my Cybertruck ignore the merging cones that it used to display and doesn't anymore is my concern.

The visualization is run by a simpler (earlier) stack than the self-driving, they run in tandem but don't interact in the car as far as I know.

-Crissa

 

[CyberGus]

I am sure it sees things but why did they remove it from the screen and why does my Cybertruck ignore the merging cones that it used to display and doesn't anymore is my concern.

Tesla's FSD team has talked about implementing the "Occupancy Network", which is basically a translation of the visual input into a 3D grid representation of occupied space. Once an area is deemed "occupied" there's no point in wasting more CPU cycles to determine what it is, so they're not really doing object-detection any more.

Of greater interest are things that can move, like vehicles and pedestrians. Those must be closely tracked to project if they will intersect with the intended direction of travel.

But for immovable objects, it doesn't matter if it's a cone, or barrel, or trash can. The computer just needs to know that the space is occupied, and thus be avoided.

 

[Darthamerica]

Tesla's FSD team has talked about implementing the "Occupancy Network", which is basically a translation of the visual input into a 3D grid representation of occupied space. Once an area is deemed "occupied" there's no point in wasting more CPU cycles to determine what it is, so they're not really doing object-detection any more.

Of greater interest are things that can move, like vehicles and pedestrians. Those must be closely tracked to project if they will intersect with the intended direction of travel.

But for immovable objects, it doesn't matter if it's a cone, or barrel, or trash can. The computer just needs to know that the space is occupied, and thus be avoided.

The problem is that it hurts user confidence in the system. We want to know the vehicle “sees” what we see in a supervised system. I suspect the real reason is that rendering graphics is consumes more compute that they need for something else. They may need to optimize the SW a bit to bring it back.

 

[CyberGus]

The problem is that it hurts user confidence in the system. We want to know the vehicle “sees” what we see in a supervised system. I suspect the real reason is that rendering graphics is consumes more compute that they need for something else. They may need to optimize the SW a bit to bring it back.

The visualization does show that something is there, but only minimally. ?‍

 

[Darthamerica]

There aren’t many CTs in the market compared to the rest of the Tesla fleet. Consequently, we might have to accept being the last ones to adopt FSD until more CTs become available to customers.

I believe the Tesla fleet covers approximately 20 million miles daily. However, only a small fraction of those miles are suitable for training the neural network. Moreover, if we consider vehicle-specific miles for CT, the number becomes even smaller.

I wouldn’t be surprised if Tesla prioritizes Model 3/Y to achieve Level 3 autonomy in geofenced areas next year. This move aligns with Elon Musk’s target of 2025. Consequently, the HW4 Model S/X and CT may have to wait a bit longer. From a business perspective, this strategy seems prudent.

 

[Darthamerica]

The visualization does show that something is there, but only minimally. ?‍

Yes it’s a grey mess of pixels. I hate it!

Sponsored