Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
The next big milestone for FSD is 11. It is a significant upgrade and fundamental changes to several parts of the FSD stack including totally new way to train the perception NN.

From AI day and Lex Fridman interview we have a good sense of what might be included.

- Object permanence both temporal and spatial
- Moving from “bag of points” to objects in NN
- Creating a 3D vector representation of the environment all in NN
- Planner optimization using NN / Monte Carlo Tree Search (MCTS)
- Change from processed images to “photon count” / raw image
- Change from single image perception to surround video
- Merging of city, highway and parking lot stacks a.k.a. Single Stack

Lex Fridman Interview of Elon. Starting with FSD related topics.


Here is a detailed explanation of Beta 11 in "layman's language" by James Douma, interview done after Lex Podcast.


Here is the AI Day explanation by in 4 parts.


screenshot-teslamotorsclub.com-2022.01.26-21_30_17.png


Here is a useful blog post asking a few questions to Tesla about AI day. The useful part comes in comparison of Tesla's methods with Waymo and others (detailed papers linked).

 
Last edited:
Do we know if the updates to FSD 11.3.2 are still rolling out or if Tesla has paused the rollout? I haven't seen any posts of people having received it in the last couple of days. My Model 3 is still on 10.69.25.2. Patience is hard...
Pending download count on TeslaFi hasn’t moved since Tuesday, though I wouldn’t call it “paused” - phased rollout like 1%, 5%, 10%, 50%, 100% are pretty common to ensure they don’t push a completely broken build to the wider fleet.

I expect by this weekend they will push additional waves unless they identify something truly critical that’s broken.

I’m also eagerly awaiting the update from 10.69.25.2.
 
Zipper Merge is controversial: link . It works great in animated simulations and simplistic explanations, but actual real world evidence is rather sparse and mixed. I have yet to find a single definitive research study on the subject. Here is one study which searched for other studies and found that "...neither case study offers definitive evidence that the zipper merge is significantly more effective...", but recommended zippering anyway.

My own (twisted humor) suggestion when I encountered a clogged merge in San Francisco many years ago was that they should put up signs well in advance of the merge saying "Space Out!" That driving technique, by the way, is how those animated simulations make it look easy.

Clearly, FSD is not going to eliminate traffic slowdowns. At least not until there are enough FSD cars to enable platoon driving, which should happen by the end of the week. ;-)
there are two "problems" first is that many folks merge right at the start a leave a great long empty lane, these are the folks who cause the the onramp to backup.
Second is that the lane is empty so some drivers just speed past everyone patiently waiting in line, which pisses people off.
The solution is to learn how to zipper - that's how the roads are designed to work, just keep pace with the traffic and merge at the end. Don't "jump the line" but also don't stop dead in onramp.
Looks like FSD executes a proper zipper.
 
  • Like
Reactions: willow_hiller
there are two "problems" first is that many folks merge right at the start a leave a great long empty lane, these are the folks who cause the the onramp to backup.
Second is that the lane is empty so some drivers just speed past everyone patiently waiting in line, which pisses people off.
The solution is to learn how to zipper - that's how the roads are designed to work, just keep pace with the traffic and merge at the end. Don't "jump the line" but also don't stop dead in onramp.
Looks like FSD executes a proper zipper.
I giggle a little when I see a semi truck driving in blocked lane at the same speed as the traffic in the through lane, blocking all the line cutters. And when road raging blocked lane drivers try to cut around the truck in the emergency - the truck then splits the emergency and blocked lanes

It really is kind of fascinating how humans receive (or perceive to receive) slights in traffic from other drivers and flip the F out, some occasionally to the point of violence. I can't say I haven't had my 'flip the F out' moments, though very rare (my wife gets angry that I don't get angry more often lol) - but it could make some interesting material for scientific studies. Very likely already done, might have to go hunting for those
 
Last edited:
The solution is to learn how to zipper -
It seems clear to me that zippering works well if people do it. I had strongly favorable direct experience of it in 1969-70 as a summer co-op student employed at Bell Labs Holmdel. It was a large site with two circumferential road with LOTS of cross-overs so people could get in and out of their parking. The employees did zippering (which I called "alternate merge" at the time) at 35 mph flawlessly, every day.

1679594814901.png

At the time I was in awe--not at the concept, which seemed obvious, but at the success of the employees in implementing it.

By the way, the location has not been a Bell Labs location for quite a long time now and was quite new when I worked there.
 
there are two "problems" first is that many folks merge right at the start a leave a great long empty lane, these are the folks who cause the the onramp to backup.
Second is that the lane is empty so some drivers just speed past everyone patiently waiting in line, which pisses people off.
The solution is to learn how to zipper - that's how the roads are designed to work, just keep pace with the traffic and merge at the end. Don't "jump the line" but also don't stop dead in onramp.
Looks like FSD executes a proper zipper.
If Tesla is using zipper merge if sure would be helpful if that was communicated to everyone. I would "supervise" FSD differently if I knew that. Shame on me I guess but frankly until this thread I've never heard of this merge approach. Just highlights the lack of any communication by Tesla to testers. For example, how many new testers are completely surprised when FSD turns on a red light or doesn't stop for a school bus with flashing lights?
 
  • Like
Reactions: VanFriscia
Installed 11.3.2 FSD beta….

Seems that Tesla has gone out of their was to attempt to kill you with this version.

Notably:

- You cannot turn off auto lane change while in FSD
- it appears that the car now likes to change into lanes that aren’t actually lanes. The resulting issue is either hitting a barrier in the center of the road, smashing into another car, or driving off the road.

Woot. Awesome. Why do they screw with things? Seriously for what good purpose is it to not allow you to turn off auto lane change?
 
Woot. Awesome. Why do they screw with things? Seriously for what good purpose is it to not allow you to turn off auto lane change?
You can minimize them. (Every drive.)

They do this because they have ~350k people beta testing the Robotaxi software, so they urgently need to gather data on public roads at minimal expense with volunteer labor. The interventions will allow them to highlight any shortcomings and get the Robotaxi software released sooner.

There is no requirement to test the robotaxi software if you do not want to.
 
Last edited:
Just downloaded 2022.46.11 V11 and took a 160 road trip. Had a lot of breaks for no reason also didn’t read to speed limits signs and much
phantom breaking again. Any body else have these issue.
Nope, no issues as I still haven't gotten the stupid thing. Tesla's 'wide rollout' is not very wide at the moment. My inner 5 year old is getting impatient 🙃
 
The above is incorrect. It can see other aircraft that are Mode S or Mode C-equipped. As stated in the wiki you linked.
Right, TCAS does not use GPS. ADS-B does, and can augment TCAS, but both systems only help the pilot to avoid aircraft, and only avoid other aircraft which are similarly equipped.

My comments were to a poster who thought aircraft collision avoidance was a good model for future autonomous cars. My point was that modern manual air traffic control is largely how aircraft collisions are prevented, but it is labor intensive, requires pilot reaction, and is neither universal nor perfect - really not a good model for FSD. "Oakland car traffic control, Tesla Y number 1974629 requesting clearance to depart my driveway enroute to Starbucks on MacArthur." Sorry, not very automated.

For the non-pilots among us, aircraft autopilots have been able to steer, turn, and navigate for years. But they use ground based or satellite radio signals, and do not avoid other aircraft. The systems for collision avoidance are patched together overlays designed to and succesful in making flying safer. Commercial flight today is far, far safer than cars, so well done! For me, Tesla's vision based FSD is a huge technology leap.

Automobile safety is atrocious. On U.S. roads we kill as many people every month as were killed in the 9/11 attacks. It seems very plausible that FSD like cars will help reduce this awful toll by using 360º sensing, quick responses, and safe driving styles.

I bought FSD to watch and to contribute to this literally vital project. Robotaxi, if it happens, will be icing on the cake in my view. When I'm too old (or too drunk) to drive myself, I'll probably feel differently. ;-)

Optimus, enabled in large part by machine vision, is probably going to have a much bigger impact, but that is a whole other conversation.
 
If Tesla is using zipper merge if sure would be helpful if that was communicated to everyone. I would "supervise" FSD differently if I knew that. Shame on me I guess but frankly until this thread I've never heard of this merge approach. Just highlights the lack of any communication by Tesla to testers. For example, how many new testers are completely surprised when FSD turns on a red light or doesn't stop for a school bus with flashing lights?
not shame on you when it isn't properly communicated to any drivers. Roads are designed to use it but in the US they don't bother telling people or have signage to help inform.
They have signs for roundabouts, but don't bother for onramps.
 
Zipper seems the most logical method for merging. What other methods should Tesla implement?

Here are a few I found:

Sidewalk Method: Hop the curb and roll. Whisk by pedestrians as construction workers gape in astonishment at your audacity. Almost as irritating as it is illegal, its proponents consider the sidewalk method as a sort of “relief zipper.”

Bike Lane Method: No lane? No problem! Just squeeze your car into the lane reserved for cyclists and put the pedal to the metal. The key here is to harbour a deep-seated disregard for human life. That, and a willingness to put this deep-seated disregard into practical application.

The Turn-Right-then-Make-an-illegal-U-turn-then-Drive-Forward Method: Drivers employ this trick when confronted by a long line of cars waiting to turn left. Rather than wait their turn (“turn” – get it?) these drivers execute this convoluted maneouvre. It’s contagious. Once one motorist pulls it off successfully, many more will follow, creating a jam in the previously empty right lane. Problem solved.

The Chicken Method: Ever wonder how the car in the next lane got all those dents? Well, you’re about to find out. He’s nosing into your lane, and if you don’t make way, you’re going to add to the collection. A time-tested means of merging. It needs no signal, no wave, no pleading look – just a whole lot of ruthless determination.

Ambulance Method: Normally restricted for use on highways, this one’s for the folks who, when life gives them lemons, make lemonade. When life causes a horrible accident, they tailgate the ambulance rushing to the scene so they can pass traffic. While it may seem dangerous, it also reveals a stunning absence of character.

The “Cone? What Cone?” Method: You drive over the orange construction cones.

Mindful Merging Method: Stay in the present. Pay attention to the inconvenience you are causing others. Needlessly stop traffic two blocks from the where the lanes combine and merge. As you poke into the lane, be aware of what you have done. Think “jerk move,” and then let it go. You may have other thoughts condemning your driving. Acknowledge each one, but do nothing about them. Be fully engaged and in your moment of automotive transgression.
 
If Tesla is using zipper merge if sure would be helpful if that was communicated to everyone. I would "supervise" FSD differently if I knew that. Shame on me I guess but frankly until this thread I've never heard of this merge approach. Just highlights the lack of any communication by Tesla to testers. For example, how many new testers are completely surprised when FSD turns on a red light or doesn't stop for a school bus with flashing lights?

Wait what? you've never heard of a zipper merge? What do you do in the TONS of situations around here were 2 lanes suddenly become one?

In my town, they used to just have a sign that represented "right lane ending." Now they replaced it with an "alternate merge" sign, which is the zipper.

The sign replacement was a godsend because now people use both lanes equally; whereas before, people treated the right lane as the cheat lane, making the left lane people tailgate to prevent the right side from merging in.
 
  • Like
Reactions: DarkForest
Just downloaded 2022.46.11 V11 and took a 160 road trip. Had a lot of breaks for no reason also didn’t read to speed limits signs and much
phantom breaking again. Any body else have these issue.
Yep, there’s an issue with speed limit signs now. Had pretty great drives all week, this afternoon I traveled some rural roads, with 45mph posted speed limit signs, sometimes the car would pick them up, other times the car would keep slowing down to 25mph, and the speed limit sign displayed was a 25mph (default) sign. Again, it was raining heavy during all my drives today, maybe that has something to do with it, who knows? Otherwise, pretty decent all week except today.
 
How’s the jerk profile on slowdowns due to limit changes?

Is it the standard extremely high jerk Tesla response from AP team, or do they actually phase the deceleration in (and out) like they are easily capable of doing, as demonstrated in other scenarios?

I’m hoping the slowdowns are substantial prompt and minimally perceptible. A very reasonable ask and fairly trivial I think.