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

New Tesla/EV owner and first-time FSD beta user - is this a joke?

This site may earn commission on affiliate links.
Many were surprised to see an older version of FSDb (11.3.6) included with the 2023.12.10 update this week that made FSDb access available to the masses instead of 11.4.x. While we're all guessing, I have a conspiracy theory: Tesla may have intentionally pushed out a bad version so that those of us who are new to FSDb will be disappointed, frustrated, and frightened and have our expecations of the feature lowered. Then, 11.4.x will be pushed out in a matter of weeks and, even though far from perfect, will seem like "magic" compared to what we started with and we'll all be praising it far more than we would had we started with that version. :)

DISCLAIMER: This post is just for fun - I don't actually believe that Tesla would do something like that...or would they? ;)
11.4 is a minor upgrade. If you are having issues in your car or area, they will likely persist until you drive your area with Beta on, often.

Again, every drive you send additional data to Tesla and it will improve in your area.
 
  • Informative
  • Like
Reactions: zazim and navguy12
11.4 is a minor upgrade. If you are having issues in your car or area, they will likely persist until you drive your area with Beta on, often.

Again, every drive you send additional data to Tesla and it will improve in your area.
Is that usually with updates or on the same update it will learn and improve? I actually live a block from where MKBHD did his video in Rutherford NJ of full self driving so I take that round about from his video every day for work it's down the street from me, It did that surprisingly well.
 
Recalibrate cameras? I take traffic circles on my commute and it’s fine.
All a recalibration of cameras will do is eliminate all the “fine tuning bias” that has taken place in the past.

A recalibration means the calibration data is wiped clean and a “coarse alignment“ takes place to allow one to re-engage the ADAS sooner than later, but it takes an accumulation of a few 100 km of straight hwy type driving to get all the cameras into a “final fine alignment“ position (all via software of course, the physical cameras are locked in a stationary position).

The analog to this would be the alignment of an old fashioned inertial nav unit, where one can navigate after a coarse alignment of about 90 seconds, but keeping the unit in “align” for a longer period of time results in more accuracy over a greater distance.

About two years ago a mobile tech showed me the data on my car (at the time) where each camera had a percentage showing (IIRC one of the cameras on the pax side was lower than the rest, at about 70ish percent).
 
  • Informative
Reactions: APotatoGod
I actually learned a lot when I was using FSD without the beta. One unintended benefit was that it limited me to 5 miles over the speed limit on city streets. It was annoying as hell, but it made me a much safer driver. Of course, the beta removes that limitation, best I can tell.

The best advice I received was to just let it do its thing, even when it appears to be off track absent an extreme emergency situation. it usually does what it is supposed to do, albeit awkwardly. I did preemptively override it last night at a difficult left turn when it looked like it was bound and determined to either ram or, at the very least, play chicken with a police cruiser.
 
I think FSDbeta is pretty awesome. Amazing when you think about what it is doing with only Camera for input.
still lots of room for improvement.

I have learned.. FSD... for the beta phase.
It can do 80-90% of the easy driving. and I accept that I will have to take over when it looks a little complicated, I will take over for that situation.
It is a team effort... tag team. less stress.

I am an odd duck... no matter what.. after 30 mins of driving (especially on highways)... i get sleepy.
The lines on the road passing by... causes me to get tired. And I have to keep my eyes on the road and look at the lines.
but with FSDbeta.. I can look around.. look at other things on the road. Focus around. but still pay attention.
I am more aware of everything else. I dont sleepy.. more awake .. more aware...
 
What are the biggest issues with the current FSD beta is that it cannot be told not to automatically change lanes. That functionalities been removed.

And frequently, the car attempts to kill you by changing lanes would it shouldn’t do it…

Similar to the whole issue I have with them removing the ability to change regenerative braking from 100%, this is another one of those things where they are preventing you from being able to turn off a key piece of functionality that could be seriously dangerous for you.
 
  • Like
Reactions: navguy12
What are the biggest issues with the current FSD beta is that it cannot be told not to automatically change lanes. That functionalities been removed.

.....So a Full (automatic) Self Driving (L2 mimicking L3) car is supposed to ask you if it can make a driving maneuver? 🤔 🤣

Also NO functionality has been removed from FSD Beta. FSD Beta has always controlled lane change decisions.

If you feel it is "seriously dangerous" (hyperbole) and want to confirm lane changes then switch to and use NoA. You can't have FSD that is not FSD or it would just be NoA, which you can use at any time.

EDIT: You can also Limit lane changes by using "Minimize Lane Changes" (right steering wheel button also).

IMG_2739 (1).jpeg


IMG_2737.jpeg
 
Last edited:
That's kind of what they are doing right now, no?

The fleet is reporting signs, poor road conditions, new roads, obstacles, wrecks, traffic, etc.
Not the same thing - Tesla needs to build a process by which we can ask them to update the map and make corrections. But that can’t all be just a website - someone at Tesla needs to verify before making map changes.
 
Well, a couple of days in, I've found that FSD struggles a lot in roundabouts when there are other cars around (it does fine when there are no other cars). Also this morning on the way to work, it decided to run a red light after decelerating fairly hard while approaching a yellow light. So yes, it needs a lot of work. But I knew that coming in. I think it'll be good for the highway portion of our upcoming road trip though.
 
  • Like
Reactions: navguy12
Finally I got it last night and the phantom breaking is insane, Since it follows about 4 car lengths back tons of cars will cut into the big opening then my car will slam on the brakes as if it was about to smack into the back of it but realistically it's 2 car lengths behind.
Well, that's not phantom 'braking', the car is doing it for a reason. That said, it could certainly be less aggressive.
 
Well, a couple of days in, I've found that FSD struggles a lot in roundabouts when there are other cars around (it does fine when there are no other cars). Also this morning on the way to work, it decided to run a red light after decelerating fairly hard while approaching a yellow light. So yes, it needs a lot of work. But I knew that coming in. I think it'll be good for the highway portion of our upcoming road trip though.
round-abouts.. for a normal driver, it is still a little confusing. little scary when it is a busy one.
FSD can do it most of the times.. but it also still has issues. much better than 4 months ago.

So, I always take over when I see a round-about with other cars in it.
 
I purchased my first EV last month - a Model Y Long Range. I've had several vehicles with adaptive cruise control and lane keeping assist. I used the adaptive cruise control often, but the lane keeping assist was a parlor trick at best. In the Hondas and Mazdas I've had with that feature, it slowly oscillated from one side of the lane to the other. I was very impressed with the Model Y's ability to stay well-centered in the lane using the standard Autopilot.

Yesterday, I began using FSD beta 11.3.6 after updating to 2023.12.10. This was my first time experiencing FSD. Yesterday evening, I let it drive me from home to a local trail and back. Immediately, I noticed how jerky and uncertain it was when pulling out of my driveway onto the main road. A mile later, I had to slam on the brakes to an ABS-controlled stop after the car appeared as though it was going to drive right through a 4-way stop with a stop sign and flashing red lights at 55 MPH.

The next time I had to intervene was when I was sitting at a major intersection waiting to turn left at a red light. When the traffic lights to the right of me turned green, my car turned on the right turn signal and began turning into traffic coming up from behind, so I had to step on the brake and yank the steering wheel. The whole time, the projected path was still for a left turn from the left turn lane - I have no idea why or what it would have done had I let it (aside from getting rear-ended by a vehicle to my right). :) The drive home was slightly less dramatic.

This morning, I let it drive me from home to work. This time, it stopped at the 4-way stop, but I had a right turn instead of going straight ahead. Multiple times I had to press the accelerator pedal as the car appeared very "timid" for no apparent reason and I could tell that traffic behind me was confused and frustrated. I noticed that it would also wait until the last moment and brake hard when coming to a stop whereas I would anticipate stops and begin slowing down hundreds of feet earlier.

I've been a tech geek all my life - first among my friends and family to have a computer, cell phone, and now an EV. I've had years of experience with "driver assist" features. I'm an engineer by day and spent a few years as an automotive journalist.

Based on my first hour with Tesla's FSD beta, I can't imagine how this is even allowed on public roads. It seems nowhere near ready. I've watched hundreds of videos demonstrating FSD, but based on my preliminary experience, I'm convinced those are cherry-picked examples where little to no intervention is required. Just in my first hour, I had to intervene nearly a dozen times.

Perhaps it's because I travel in rural and suburban areas where Teslas are extremely rare, so the system has no "training" here? As it stands, it's not only useless, but downright dangerous. I'll pay for another month and see how it works for other drives, but I have a strong suspicion that FSD in its current form is not for me. At minimum, I'm embarrassed by how it drives and even fearful at times - and this is coming from someone who puts a lot of faith in technology. :)
The new version (11.4.2) is much better but it still needs to be watched closely. These are Beta versions and should be treated as such.
 
Fair enough, but my (limited) understanding is that TACC and NOA is now using a “single stack” of software and it still has phantom braking issues as well as slowing down to 100 kph whilst inside the interchange between HWY 17 and HWY 417 in the Ottawa area.
\

No, only the FSDb is using the single stack, meaning that the FSDb now uses its new code on highways. If you're not using FSDb it's still the old code which is getting little attention.

This change greatly improved FSDb on the highway---where it is now the reliable driver for me. On streets and complex intersections, it often needs human supervision and it's often less stressful to drive manually. Before this change FSDb wasn't a valuable option.

Once the FSDb highway code migrates to regular AP/NOA, it will be an excellent driver assist product at high level 2 and depending on legality maybe low L3.

The streets portion is not universally reliable---you may take routes frequently where you have experience it works well, but beware on any new route.

I'm also testing in California which probably has better performance than other areas because of the density of California in the training set.
 
  • Helpful
Reactions: navguy12
If it were up to me, I would have all Teslas - even those not subscribed to FSDb constantly capture training data.

For example, I take the exact same routes five days a week - home > work > trail > home. I made those driven dozens of times before subscribing to FSDb. The vehicle had plenty of opportunity to learn how to drive these routes. It would be nice if the vehicle had learned things like where to stop, how to turn, etc. from my manual driving so that when I turned on FSD, it would already know how to make those particular routes.

There's probably a reason I'm not a FSD software engineer at Tesla. :D
 
  • Like
Reactions: navguy12
One of my main complaints of FSD Beta when I had it (paid $6k in 2019 for it), was that I had the impression that the car doesn't look as far ahead as it should. Coming to a stop a few hundred feet ahead, as a human I would let off the throttle a bit and let it coast to a stop, but FSD tends to keep going at full speed while not noticing the stopped cars until it's kind of, uncomfortably close. Many times, as OP mentioned, I felt like it's driving like a teenager that just got its license and doesn't know how to look that far ahead or, is timid to proceed even when it's clear to proceed, or do something outright dangerous last second (like turning on a red light when traffic is coming).

Over a year, it was fascinating seeing the traffic stop light, stop sign recognition, plus getting the first FSD beta releases and letting the car make turns by itself. But, the novelty wore off within a year or so of watching the software degrade and kind of regress in its driving proficiency with the latest releases. It was always a step forward, then two back, then two ahead and 1 back and then 4 steps back. It got a little tiring where I didn't use FSD much. EAP on the other hand is a bit more useful, at least for me.

Managing expectations is important for oneself, as it's still beta software, and it'll probably be 10 years from now I suppose. And, we all want it all, and want it now. But that's not realistic with software development. Definitely it is a monumental task, from a machine learning and AI standpoint, to continue building petabytes of videos showing every single scenario to train the driving models, but something new will always come up. And there's also having to account different countries for training driving models. It's a never ending task, like remaining in beta.
 
  • Like
Reactions: zroger73
No, only the FSDb is using the single stack, meaning that the FSDb now uses its new code on highways. If you're not using FSDb it's still the old code which is getting little attention.

This change greatly improved FSDb on the highway---where it is now the reliable driver for me. On streets and complex intersections, it often needs human supervision and it's often less stressful to drive manually. Before this change FSDb wasn't a valuable option.

Once the FSDb highway code migrates to regular AP/NOA, it will be an excellent driver assist product at high level 2 and depending on legality maybe low L3.

The streets portion is not universally reliable---you may take routes frequently where you have experience it works well, but beware on any new route.

I'm also testing in California which probably has better performance than other areas because of the density of California in the training set.
Thanks for the info.

My car is still on 2022.45.15 fsdb 11.3.6 so I think I’m somewhat behind the times…
 
One of my main complaints of FSD Beta when I had it (paid $6k in 2019 for it), was that I had the impression that the car doesn't look as far ahead as it should. Coming to a stop a few hundred feet ahead, as a human I would let off the throttle a bit and let it coast to a stop, but FSD tends to keep going at full speed while not noticing the stopped cars until it's kind of, uncomfortably close. Many times, as OP mentioned, I felt like it's driving like a teenager that just got its license and doesn't know how to look that far ahead or, is timid to proceed even when it's clear to proceed, or do something outright dangerous last second (like turning on a red light when traffic is coming).

Over a year, it was fascinating seeing the traffic stop light, stop sign recognition, plus getting the first FSD beta releases and letting the car make turns by itself. But, the novelty wore off within a year or so of watching the software degrade and kind of regress in its driving proficiency with the latest releases. It was always a step forward, then two back, then two ahead and 1 back and then 4 steps back. It got a little tiring where I didn't use FSD much. EAP on the other hand is a bit more useful, at least for me.

Managing expectations is important for oneself, as it's still beta software, and it'll probably be 10 years from now I suppose. And, we all want it all, and want it now. But that's not realistic with software development. Definitely it is a monumental task, from a machine learning and AI standpoint, to continue building petabytes of videos showing every single scenario to train the driving models, but something new will always come up. And there's also having to account different countries for training driving models. It's a never ending task, like remaining in beta.
This has been my big reservation about all of this: the car simply doesn’t look far enough ahead to avoid incipient situations.

I have no issue that the software will be safer/faster than I can be when in a close situation, but in another life I was taught that to avoid a threat was superior to using top level skills to escape a threat once caught by it.
 
This has been my big reservation about all of this: the car simply doesn’t look far enough ahead to avoid incipient situations.
This is likely because of insufficient resolution in the current camera set 1280x960, which is much too low compared to what a human fovea can do (though humans have to focus on specific regions to get that good resolution), and what other autonomous driving systems do.

Higher resolution means detecting and classifying accurately at a longer distance. It costs more and the data handling requirements go up very substantially and so does the compute cost.
 
  • Informative
Reactions: navguy12