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:
Did they actually implement any merging control for highway? We were merging at stop-and-go speeds in rush hour traffic, the lead car merges into traffic, and FSD decides to swerve around them and everyone else like a complete jerk. I feel like it doesn't attempting merging at all, it just treats the whole process as an obstruction to pathfind around.
There was a fair amount of discussion up-thread ( just two or three days ago I think) about merging policy - you can search for "zipper merge" discussions.

I'll try to summarize what I learned from reading those discussions; please note that a lot of this was new to me and I may not be explaining it quite correctly:

It seems that in countries like Germany, and some US localities where zipper merge practices have been pushed by local traffic authorities, the expected correct behavior is not to merge at the earliest opportunity, but to utilize all of the available entrance lane, merge near the end of its run, and expect drivers in the through lane to cooperate with alternating zipper merge practices. Supposedly this is the theoretically most efficient way to execute a merging of two lanes.

However, if this technique hasn't been officially publicized, promoted and taught, then it may come across as rude behavior, where the merging car is thought to be selfishly jumping ahead to pass up other traffic that's been patiently puttering in the slow through lane.

So, to people who believe in the zipper merge, you're a bad driver if you don't scoot forward to the last possible merging point, but to people who are unaware or disagree with it, you're a selfish jackass if you do. I guess that the very poorest behavior would be to jump out of the through lane, into the incoming merge lane and then scoot forward past drivers who had just been directly in front of you!

I don't remember being taught any official and universal method, so until now I've tended to think it's more polite to merge a bit earlier and not try to pass too many cars in the process - but sometimes I move farther up the line if it's clear that the driver(s) immediately next to me aren't interested in making room. It does seem that overall stress and conflicts would be reduced if everyone understood a clear strategy. Tesla FSD is currently very far from any ability to interpret adjacent drivers' attitudes from their vehicle-jockeying behavior, much less their facial expressions or hand gestures.

Bottom line, FSD may actually be following the "advice of experts" even though it's not obvious to many of us.
 
Last edited:
There was a fair amount of discussion up-thread ( just two or three days ago I think) about merging policy - you can search for "zipper merge" discussions.

I'll try to summarize what I learned from reading those discussions; please note that a lot of this was new to me and I may not be explaining it quite correctly:

It seems that in countries like Germany, and some US localities where zipper merge practices have been pushed by local traffic authorities, the expected correct behavior is not to merge at the earliest opportunity, but to utilize all of the available entrance lane, merge near the end of its run, and expect drivers in the through lane to cooperate with alternating zipper merge practices. Supposedly this is the theoretically most efficient way to execute a merging of two lanes.

However, if this technique hasn't been officially publicized, promoted and taught, then it may come across as rude behavior, where the merging car is thought to be selfishly jumping ahead to pass up other traffic that's been patiently puttering in the slow through lane.

So, to people who believe in the zipper merge, you're a bad driver if you don't scoot forward to the last possible merging point, but to people who are unaware or disagree with it, you're a selfish jackass if you do. I guess that the very poorest behavior would be to jump out of the through lane, into the incoming merge lane and then scoot forward past drivers who had just been directly in front of you!

I don't remember being taught any official and universal method, so until now I've tended to think it's more polite to merge a bit earlier and not try to pass too many cars in the process - but sometimes I move farther up the line if it's clear that the driver(s) immediately next to me aren't interested in making room. It does seem that overall stress and conflicts would be reduced if everyone understood a clear strategy. Tesla FSD is currently very far from any ability to interpret adjacent drivers' attitudes from their vehicle-jockeying behavior, much less their facial expressions or hand gestures.

Bottom line, FSD may actually be following the "advice of experts" even though it's not obvious to many of us.

"Zipper in" is the correct interpretation in the UK of its highway code, although not many have the comprehension skills for the rules so yes, people get excited by the behaviour.
 
  • Like
Reactions: JHCCAZ
There was a fair amount of discussion up-thread ( just two or three days ago I think) about merging policy - you can search for "zipper merge" discussions.

I'll try to summarize what I learned from reading those discussions; please note that a lot of this was new to me and I may not be explaining it quite correctly:

It seems that in countries like Germany, and some US localities where zipper merge practices have been pushed by local traffic authorities, the expected correct behavior is not to merge at the earliest opportunity, but to utilize all of the available entrance lane, merge near the end of its run, and expect drivers in the through lane to cooperate with alternating zipper merge practices. Supposedly this is the theoretically most efficient way to execute a merging of two lanes.

However, if this technique hasn't been officially publicized, promoted and taught, then it may come across as rude behavior, where the merging car is thought to be selfishly jumping ahead to pass up other traffic that's been patiently puttering in the slow through lane.

So, to people who believe in the zipper merge, you're a bad driver if you don't scoot forward to the last possible merging point, but to people who are unaware or disagree with it, you're a selfish jackass if you do. I guess that the very poorest behavior would be to jump out of the through lane, into the incoming merge lane and then scoot forward past drivers who had just been directly in front of you!

I don't remember being taught any official and universal method, so until now I've tended to think it's more polite to merge a bit earlier and not try to pass too many cars in the process - but sometimes I move farther up the line if it's clear that the driver(s) immediately next to me aren't interested in making room. It does seem that overall stress and conflicts would be reduced if everyone understood a clear strategy. Tesla FSD is currently very far from any ability to interpret adjacent drivers' attitudes from their vehicle-jockeying behavior, much less their facial expressions or hand gestures.

Bottom line, FSD may actually be following the "advice of experts" even though it's not obvious to many of us.

Zipper merge is recommended (in Germany) when two lanes with equal rights merge. Highway on-ramps and similar constructs don't have right-of-way. A car on the ramp has to give way and find a gap to merge. In other words: It's expected to merge from a ramp as soon as it's safe to do so. With a zipper merge, it is most efficient to merge late.
 
Zipper merge is recommended (in Germany) when two lanes with equal rights merge. Highway on-ramps and similar constructs don't have right-of-way. A car on the ramp has to give way and find a gap to merge. In other words: It's expected to merge from a ramp as soon as it's safe to do so. With a zipper merge, it is most efficient to merge late.
Thanks, that makes sense. However, perhaps you could comment on this (English language) webpage. From the drawings and the explanation, it looks like they're claiming that zipper merge is used even on Autobahn merges:
Back to USA attempts to encourage zipper merging, including highway entrance merging:
I found this relatively short and easy to watch YouTube video. Note that two or three times in here, they emphasize that you should use all of the entrance lane, and everyone should use the merge point where the lane really ends.
 
  • Like
Reactions: Mike1080i
Day three drive to work. Flawless trip from the moment I engaged from my driveway to the center entrance to my building. I was the worthless human being sitting in the drivers seat torquing the steering wheel.

The car made an impressive move last night. turn right and jumped four lanes into a left turn lane by the next block.

the new blue images imply what the car is thinking about or fixating on. The new comments are good when the car is trying to decide if the coast is clear to go. Also moved them back to the top of the screen where I can see them better. Now I need a steering yoke.

We have turn signals! what will the FSD team think of next.

This update reminds me of when we got 9.0 Auto Pilot. Really cleaned things up into a nice presentable package. Smooth and flowing operation. There polishing what it has to offer.

I do wish we could drop a pin for the destination. my car will take me to the from door as if it was a Robo taxi, but my entrance is down the side of the building complex. Then I could have it autopark. Next version right.

I have only had two disengagements so far, both on left turns failing to maintain its lane. I drifted from the left turn lane into the right lane half way through the turn.
 
The prosecution rests...

While it appears it MAY be true that is a PURE supposition article based on the same Musk Tweet we have all read from 3 days ago that is somewhat ambiguous (as many are). And the title is clickbait inaccurate since Musk did NOT announce 11.3.4 (as of now).

There is no evidence of 11.3.4 being offered other than speculation and the obvious statement ...The next release, version 11.3.4, will undoubtedly bring even more improvements and advancements to Tesla's FSD system...." That "No Shi$ Sherlock" statement covers ALL next releases.

Again 11.3.3 is looking more doubtful but we are all still in the dark about the next steps until more info surfaces.
 
Last edited:
I do wish we could drop a pin for the destination. my car will take me to the from door as if it was a Robo taxi, but my entrance is down the side of the building complex. Then I could have it autopark. Next version right.
I have a couple of cases in my daily commutes, where I wanted the car to approach the destination in a different way than the navigation would choose.

In one case, there's a parking lot entrance that the map routing never chooses if you just put in the published address of any business in that shopping center.

I solved this by getting on the desktop version of Google Maps on my PC, and playing with destination addresses. I tried various street numbers that didn't really exist for any business, but by trial and error I could find the exact street number that corresponded to the parking lot entrance I wanted to use.
(Sometimes, you can grab the little target on the desktop app and move it with the mouse to get a real-time changing display of the corresponding street address, other times you see only GPS coordinates and then you have to just type in different street numbers until you nail it.)​

By programming those pseudo-address into the Tesla nav, I was able to get it to turn exactly where I wanted. Note though, that in one case I had to tweak the pseudo-address street number a little bit more from what the Google Maps session had suggested.

After capturing the needed pseudo addresses, I programmed them as named Favorites in in the Tesla nav, and now they work every time.

This meth9d is obviously not too helpful if you just want to program a quick route to a new destination one time - it takes some effort and only makes sense as a workaround for destinations that you visit frequently.

By the way, thanks for all your reports and it's encouraging that you're having such good results. I wish I could get 11.3.3 myself!
 
I do wish we could drop a pin for the destination.

You can drop pins for destinations. Press and hold on a point on the map and it will drop an arbitrary pin. From there you can tap the empty star icon in the corner of the popup to favorite it, and give it a name.

This is the technique I used to try and get FSDb to pull around to the kiss-and-ride of a local train station. It would have worked, except there was a bus lane running parallel across a physical median (Beta assumes the median means my desired lane is for on-coming traffic, and tries to take the bus lane every time).
 
There was a fair amount of discussion up-thread ( just two or three days ago I think) about merging policy - you can search for "zipper merge" discussions.

Bottom line, FSD may actually be following the "advice of experts" even though it's not obvious to many of us.
You're exactly right. I am someone who wasn't familiar with zip merging and after the discussion in this thread now understand better what FSD is trying to do. For me it's been a mixed bag since FSD still doesn't adjust speed properly by either slowing or speeding up to merge. Last night for example I late the merge play out and had to disengage when it was clear FSD was going to run out of merge lane and would start to drive in the break down lane. We were driving along side a car that could have adjusted it's speed for us but clearly was not going to. This is especially problematic during rush hour when traffic is slow. Merging still needs a lot of work.
 
I have a 2020 model 3 and have had a blast w FSD beta since. After upgrading to 2022.45.12, FSD beta is no longer available. My safety score is a 90 so not sure if that is too low. Also, I got a weird message about FSD not available in my region but I’m in Texas and have had it for 3 years. Any ideas?
 
  • Informative
Reactions: pilotSteve
Thanks, that makes sense. However, perhaps you could comment on this (English language) webpage. From the drawings and the explanation, it looks like they're claiming that zipper merge is used even on Autobahn merges:
Back to USA attempts to encourage zipper merging, including highway entrance merging:
I found this relatively short and easy to watch YouTube video. Note that two or three times in here, they emphasize that you should use all of the entrance lane, and everyone should use the merge point where the lane really ends.

This is the machine-translated wording of §7 (4) StVO:
If, on roads with several lanes, continuous use of one lane is not possible for one direction or if one lane ends, vehicles prevented from continuing shall be allowed to cross to the adjacent lane in such a way that these vehicles can alternately line up after a vehicle travelling on the continuous lane immediately before the start of the narrowing (zipper method).
IANAL but my read of "roads with several lanes" is that it implies that all lanes have equal rights and therefore does not apply to on-ramps.
The zipper merge minimizes congestion when a lane ends or is blocked. With light traffic, it doesn't matter whether cars merge early or late. Likewise, traffic merging from an on-ramp is usually sparse unless the highway is already congested. In which case it's not a matter of getting up to the same speed as traffic on the highway but finding a gap and yes, people tend to merge late in such cases. The OP was about normal traffic flow - if I remember it right, it's several pages back.
 
  • Informative
Reactions: JHCCAZ
While we're still beta testing the thing, voice announcements would be nice, as with the navigation system. Mostly just "proceeding", "stopping", "turning left", "turning right", "changing lane left", "changing lane right", "freaking out". Stuff like that.
Careful for you wish for. I remember as a kid my dad had a 1980 Chrysler LaBaron. It talked to him no stop. "put you seat belt on" it was a gimmick and got annoying real quick. He could not turn it off. At least with Tesla we could shut it off.
 
Where I live 90% of the turns you're not supposed to turn on have a simple "No Turn on Red" sign which is very visible. Of course there will be exceptions but I suspect the majority of intersections nationally could be handled by keeping sign recognition simple and start with just the most common signs.

Generally when the car turns on red you're going to be ok since FSD is checking for cross traffic. However, we have a large and very busy rouindabout where each entry point is controlled by lights. When FSD stops on red and immediately checks for cars coming it doesn't realize that multiple lanes of traffic that are stopped are going to come suddenly and way before FSD makes it's decision. Disaster time. I have to disengage every time including on 11.3.3 today.
AT each entry point the "No Turn on Red" is attached to the light hanging directly in front of the car. Super obvious.

Google Maps

haha before i clicked your link, I was thinking, hmmm sounds like Drum Hill Rotary!
 
There was a fair amount of discussion up-thread ( just two or three days ago I think) about merging policy - you can search for "zipper merge" discussions.

I'll try to summarize what I learned from reading those discussions; please note that a lot of this was new to me and I may not be explaining it quite correctly:

It seems that in countries like Germany, and some US localities where zipper merge practices have been pushed by local traffic authorities, the expected correct behavior is not to merge at the earliest opportunity, but to utilize all of the available entrance lane, merge near the end of its run, and expect drivers in the through lane to cooperate with alternating zipper merge practices. Supposedly this is the theoretically most efficient way to execute a merging of two lanes.

However, if this technique hasn't been officially publicized, promoted and taught, then it may come across as rude behavior, where the merging car is thought to be selfishly jumping ahead to pass up other traffic that's been patiently puttering in the slow through lane.

So, to people who believe in the zipper merge, you're a bad driver if you don't scoot forward to the last possible merging point, but to people who are unaware or disagree with it, you're a selfish jackass if you do. I guess that the very poorest behavior would be to jump out of the through lane, into the incoming merge lane and then scoot forward past drivers who had just been directly in front of you!

I don't remember being taught any official and universal method, so until now I've tended to think it's more polite to merge a bit earlier and not try to pass too many cars in the process - but sometimes I move farther up the line if it's clear that the driver(s) immediately next to me aren't interested in making room. It does seem that overall stress and conflicts would be reduced if everyone understood a clear strategy. Tesla FSD is currently very far from any ability to interpret adjacent drivers' attitudes from their vehicle-jockeying behavior, much less their facial expressions or hand gestures.

Bottom line, FSD may actually be following the "advice of experts" even though it's not obvious to many of us.

In aggressive-driving cities (I'm from Boston), you'd be expected to take up all the free space in the merge lane and merge in last-minute. If you don't do it, there's a horde of other cars behind you that will once you merge in early. It's part of the culture, and zipper merging becomes very natural. And there are plenty of drivers that will exploit a large enough gap to squeeze themselves in as an extra car in the typical 1:1 zipper merge. So it's become standard to tailgate as much as possible leading up to the zipper merge so people can't take advantage of gaps. The end result is that the two lanes have long become one, but the cars are still side-by-side because everyone is tailgating and waiting for evidence that the next person behind isn't attempting to violate the 1:1 rule.

Now, you take that kind of driving culture, and then you watch a Dirty Tesla video in calm, serene Ann Arbor MI, where going exactly the speed limit is acceptable, and suddenly FSDb performance there is completely irrelevant to my driving locale. It's great to see progress there, but it's nowhere near ready to perform here.