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

FSD Beta 10.11

This site may earn commission on affiliate links.
How'd a car in Moscow get 10.11 at all? I thought FSD was still limited to the U.S.?

Once the cameras are calibrated, it should be a relatively straightforward transformation calculation to merge the two camera images.
I remember seeing somewhere that fusion of the cameras is still in the works and isn't employed right now.
 
I remember seeing somewhere that fusion of the cameras is still in the works and isn't employed right now.
Yes, that 8 camera fused image is what's coming in v11, according to Elon Musk. Expect for FSD performance to take a step backward as the vision processing neural networks will have to be completely retrained. In addition, the output of the vision processing NN will be the 3D vector space, instead of being "a bag of points," so the algorithmic code (or driving NNs) on top of the vector space may have to be rewritten/retrained as well.
 
Well, this is not a fun weekend so far. I just received the 2022.8.2 update on my '22 S Plaid this morning. For a second, I thought I was an FSD Beta enrollment push. And yet, my '21 M3 is still stuck at 10.10.2 with lots of data contributed to the FSD Beta program.

I have reserved the S P as my weekend driver since I don't have to drive to work and risk breaking my 100 SS. The M3's last FSD has a lot of bugs, for example, it jumps into turn lanes when trying to make a simple lane change. That's even with the updated map that clearly shows it's a turn lane and not part of the routed driving lane.

I'm not a happy camper, to say the least. Come on Tesla, make me a fanboy again.
 
  • Funny
Reactions: EVNow
Well, this is not a fun weekend so far. I just received the 2022.8.2 update on my '22 S Plaid this morning. For a second, I thought I was an FSD Beta enrollment push. And yet, my '21 M3 is still stuck at 10.10.2 with lots of data contributed to the FSD Beta program.

You have 1 car in FSD beta, and want more. Many of us have zero. Count yourself lucky, and get in line. :)
 
I remember seeing somewhere that fusion of the cameras is still in the works and isn't employed right now.
Fusion or not I don't see how Tesla can get over the 30" difference between what I can see leaning forward and the B Pillar camera especially if the car is turning right since the front camera is then not perpendicular to the road which reduces it's vision. I think it's for this reason the car likes to approach intersections straight on when a human turns the car slightly left or right.
 
Fusion or not I don't see how Tesla can get over the 30" difference between what I can see leaning forward and the B Pillar camera especially if the car is turning right since the front camera is then not perpendicular to the road which reduces it's vision. I think it's for this reason the car likes to approach intersections straight on when a human turns the car slightly left or right.

I'm also concerned about this. However, they had at least two full redesign opportunities to fix this (HW 1->2, and 2->3), so perhaps it's less a concern than we think? Contrariwise, the unsubstantiated rumors of HW4, if true, indicate that more cameras are coming -- though we don't know where.

Hopefully they'll also use this opportunity to also create a real birds eye view. I really miss this from my 2011 (!) BMW.
 
Yes, that 8 camera fused image is what's coming in v11, according to Elon Musk. Expect for FSD performance to take a step backward as the vision processing neural networks will have to be completely retrained. In addition, the output of the vision processing NN will be the 3D vector space, instead of being "a bag of points," so the algorithmic code (or driving NNs) on top of the vector space may have to be rewritten/retrained as well.
See the Beta 11 thread. They are already releasing some of those features in 10.x releases.
 
I tend to think it's that Tesla is not able to blend front and B-pillar camera data into a cohesive picture (so far), and it's possible that when the cameras individually don't agree with each other, for safety, the car doesn't go. For example, if in my example where I'm turning RIGHT at a T, and the car is already facing toward the right, the B-pillar data shows OK to go, but front camera is saying I can't see sh*t, the car doesn't go.
That's not quite how it works, as the cameras all contribute to a single unified view (the BEV), and the car then uses this to determine car positions/speeds (and probabilities). However, the new result is similar, as both the cameras will contribute input to that determination.
 
  • Like
Reactions: FSDtester#1
I'm also concerned about this. However, they had at least two full redesign opportunities to fix this (HW 1->2, and 2->3), so perhaps it's less a concern than we think? Contrariwise, the unsubstantiated rumors of HW4, if true, indicate that more cameras are coming -- though we don't know where.
My money is on cameras close to headlights that have a field of view that covers sides and forward looking over short distances (for short-distance stereoscopic distance calculations).
 
Tesla Fi now showing update from 2022.4.5.3 to 2022.4.5.15 ( which is FSD beta 10.11). It looks like Tesla is now starting to open up to new beta testers.
That is what confused me the other day. I had noticed there were a couple FSD Beta entries as well. But those may be unique situations as another member pointed put to me ealier. I would think all existing members would get first dibs before new enrollees.
 
My money is on cameras close to headlights that have a field of view that covers sides and forward looking over short distances (for short-distance stereoscopic distance calculations).
I think the entire body should be one big damn plastic bubble, the entire surface of which is a "camera". Redesign that whole camera concept and reality. And for good measure throw in a drone, you know, to predict the future...........
 
  • Like
Reactions: edseloh