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:
ok, so we all agree the auto wipers dont work (though they USED to for me at least). Rather than complaining here, why dont we all get on Twitter X and tweet bomb Elon? As in, you know, try to do something rather than just complain to each other?

Suggested text: "@ElonMusk How about auto wipers that work? We know FSD is tricky but isnt it a bit embarrassing that auto wipers (that USED to work fine) are so broken right now? They come on during sunny days, and stay off during a rainstorm."
I’m game. Done.
 
***This new drug may cause blindness, cramps, diarrhea, and occasional black outs.
Well, yeah. Chemo can cause all of that and then some.
There are actually a lot of experimental drugs for cures not yet available ... or doing it differently but less accurately.

Good example is blood glucose sensor. Yes, you can prick your finger and get (somewhat accurate) numbers. But its a major pain ... ;)

So, we get all kinds of sensors you put on and get continuous readings with no pain / effort. But they are all less accurate ... should we wait until the sensors get as accurate as finger prick ? Why not ?

Similarly for simple reading of temperature. A mercury thermometer under the tongue may be the most accurate - but a IR scan of the forehead is so much easier, even if less accurate.

You have to only think a little to find such examples in the medical field.
The chemo example was partially in jest and actually isn't the best example but it served to illustrate a point.

In general, the FDA requires a drug be shown to be better than placebo to even get a approval and most physicians look critically to see if they are better than the current standard of care.

In your glucometer example, the traditional finger stick meters are required to pass accuracy and consistency tests before being marketed. The continuous monitor is not as good as a finger stick, *but* has the signifiant advantage of being continuous and linkable.

Consumer grade thermometers that do IR scans of the head are notoriously inaccurate. Any physician who wants an accurate temperature reading will use a digital thermometer that goes under the tongue. In the OR we use an esophageal temperature probe. There are medical grade skin temp scanners but they need to be used properly. More importantly, they are don't completely replace the gold standard.

Which brings us back to Tesla. In the above examples, there are clear clinical advantages that balance out the potential disadvantages of the devices/drugs and they never completely replace the gold standard, they simply supplement it in certain situations. Tesla took a system that worked and replaced it with one that was clearly inferior and never superior without any actual pressing need to do so, other than their own desires.
 
Well, yeah. Chemo can cause all of that and then some.

The chemo example was partially in jest and actually isn't the best example but it served to illustrate a point.

In general, the FDA requires a drug be shown to be better than placebo to even get a approval and most physicians look critically to see if they are better than the current standard of care.

In your glucometer example, the traditional finger stick meters are required to pass accuracy and consistency tests before being marketed. The continuous monitor is not as good as a finger stick, *but* has the signifiant advantage of being continuous and linkable.

Consumer grade thermometers that do IR scans of the head are notoriously inaccurate. Any physician who wants an accurate temperature reading will use a digital thermometer that goes under the tongue. In the OR we use an esophageal temperature probe. There are medical grade skin temp scanners but they need to be used properly. More importantly, they are don't completely replace the gold standard.

Which brings us back to Tesla. In the above examples, there are clear clinical advantages that balance out the potential disadvantages of the devices/drugs and they never completely replace the gold standard, they simply supplement it in certain situations. Tesla took a system that worked and replaced it with one that was clearly inferior and never superior without any actual pressing need to do so, other than their own desires.
And CGM sensors, overall, help manage the disease better than finger stick methods - as you have continuous monitoring and insulin/carb intake adjusting. Finger pricking sucks, and people would do it as little as possible, resulting in faster disease progression and complications from high or low blood sugar levels.
 
Again you are making the mistake of thinking I agree with what Tesla did. I'm just giving an explanation of "why" ...

If you go back and see my posts on ultrasonic (and aradar) removal you will see I put that down to cost cutting / parts shortage.
You’re right but the optics are still bad - adding frivolous and unnecessary features while core driving functions appear to languish.
 
Precisely. Around the time Tesla removed the USS, there were Fords piling up in parking lots because they couldn't adapt to the chip shortage: https://www.motortrend.com/news/2022-ford-pickup-trucks-f150-chip-shortage-kentucky-speedway/

When the alternative is ceasing selling vehicles entirely, removing the USS looks like a smart decision.
You use a false choice in your argument. There was another option - install the wiring harness without the sensors and have people return to the service center to have them installed once they became available.
 
  • Like
Reactions: jebinc
You use a false choice in your argument. There was another option - install the wiring harness without the sensors and have people return to the service center to have them installed once they became available.
Monty Python: I didn't come here for an argument. Yes you did!


Perhaps we come here for abuse... ;-)

Regarding the wipers, I think some folks here signed up for the testing branch only as a way to get FSD before Tesla was ready to release it to production. 11.4.8.1 is not a Tesla product for sale. 11.4.4 is what we paid for, 11.4.8 is what we volunteered to help develop. Feeling disappointed about 11.4.8.x is self inflicted injury, not proof of some failure on Tesla's part.

Tesla makes plenty of dumb mistakes, like slow repair parts, service making thing worse (they once bricked our MY for 2 months.) blaming software for hardware problems, etc, etc, etc. But FSD beta testing is testing, not product reviewing, IMO
 
Monty Python: I didn't come here for an argument. Yes you did!


Perhaps we come here for abuse... ;-)

Regarding the wipers, I think some folks here signed up for the testing branch only as a way to get FSD before Tesla was ready to release it to production. 11.4.8.1 is not a Tesla product for sale. 11.4.4 is what we paid for, 11.4.8 is what we volunteered to help develop. Feeling disappointed about 11.4.8.x is self inflicted injury, not proof of some failure on Tesla's part.

Tesla makes plenty of dumb mistakes, like slow repair parts, service making thing worse (they once bricked our MY for 2 months.) blaming software for hardware problems, etc, etc, etc. But FSD beta testing is testing, not product reviewing , IMO
But, but, but, that’s what I’ve been trying to get - what I paid for 11.4.4 and a production branch not 6 months old! Tesla refuses to give it to us, as discussed previously, ad nauseum.
 
In general, the FDA requires a drug be shown to be better than placebo to even get a approval.
Today's game is it has to be shown to be not inferior, in fake pharmaceutical studies, to that effective dirt-cheap medicine that has helped mankind survive since the dinosaurs went awol, but most importantly, it has to cost at least 100,000 times more..........
 
But, but, but, that’s what I’ve been trying to get - what I paid for 11.4.4 and a production branch not 6 months old! Tesla refuses to give it to us, as discussed previously, ad nauseum.
I hear you! We did opt in, but there is no known way to opt out.

FSDβ testing is similar the Hotel California: You can check in any time you like, but you can never leave.

(Hmmm. My understanding is that FSD does transfer when ownership is transferred? How about beta testing, that should not transfer. If this is correct, could maybe one gift the car to himself and register the "transfer" with Tesla, maybe DMV too?)
 
2. I'll switch profiles to my FSD not enabled profile and see if there's a difference. We are in winter driving now so I don't use FSD at all at the moment (I had ineffective Auto wipers on Sunday when it was needed but the car had FSD enabled, just not used, on that drive, ) It will take a few days to report back since the driving is currently bad enough that I am avoiding taking the car out at all. Even if I slow down and have winter tires on, and drive predictably, the other drivers on the road can't be relied upon to not plow into me, so I don't risk it unless I need to go somewhere, and errands with the car are batched.

Today was the first day I got to drive the car again. I'm the stereotypical little old lady who only drives to church on Sunday!

When we were out this week, the wipers were incredibly non-functional which made it easy to remember I had promised to do this test when I got my turn behind the wheel. Last night when we got home I set the profiles up so I'd start out in AP only (not EAP) even if I forgot about doing the test. And I set the wipers to auto.

I didn't have a chance to check dry swiping, it was raining today (and temp was 4C.) In Auto my wipers never swiped once. I double checked by doing voice command Wipers Auto and then Wipers Off and then Wipers Auto to make sure. They never worked. I eventually gave up and turned them on because I couldn't see to drive safely.

After church it was just a light mist that the RainX add-on from the car wash dealt with. They never turned on.

Sadly, the headlights were in the Off position, not Auto so I sadly drove to church in the rain with no taillights. That's not my normal way of defensive driving (avoid hitting things and avoid being hit by idiots.) And I learned that Headlights ON is not a recognized command (it said that there were no headlights commands at this time.)
 
. It gets to 68mph on Carl Parker and slows downs gently before ending up aggressively braking when it figures out it has a sharp right/left to get onto Rices Crossing and craps out on the left-hander part. I took over there and re-engaged for it to handle the right onto S Main. Most uncomfortable and another regression.

First, US speed limits freak me out. But that's beside the point.

When you drive that series of changes without AP/EAP or FSDb, do you attempt maintain the speed limit of Carlos Perkins (65mph) while on the off ramp or follow the 20mph suggestion that is CLEARLY POSTED TWICE?

We find FSDb doesn't read yellow signs so won't take ramps at the suggested speed. In this case, that means the car doesn't know it isn't supposed to be driving at 65mph for the hard right curve and immediate left turn. Obviously an FSD fail. (And I'm betting FSD (not b, but EAP) would pull the same bad exit strategy and hit those curves going way to fast.)
 
First, US speed limits freak me out. But that's beside the point.

When you drive that series of changes without AP/EAP or FSDb, do you attempt maintain the speed limit of Carlos Perkins (65mph) while on the off ramp or follow the 20mph suggestion that is CLEARLY POSTED TWICE?

We find FSDb doesn't read yellow signs so won't take ramps at the suggested speed. In this case, that means the car doesn't know it isn't supposed to be driving at 65mph for the hard right curve and immediate left turn. Obviously an FSD fail. (And I'm betting FSD (not b, but EAP) would pull the same bad exit strategy and hit those curves going way to fast.)
I’ve always noticed marked or not if my car enters an off ramp or aggressive turn it automatically reduces speed to maintain center lane.
 
  • Like
Reactions: Supcom
First, US speed limits freak me out. But that's beside the point.

When you drive that series of changes without AP/EAP or FSDb, do you attempt maintain the speed limit of Carlos Perkins (65mph) while on the off ramp or follow the 20mph suggestion that is CLEARLY POSTED TWICE?

We find FSDb doesn't read yellow signs so won't take ramps at the suggested speed. In this case, that means the car doesn't know it isn't supposed to be driving at 65mph for the hard right curve and immediate left turn. Obviously an FSD fail. (And I'm betting FSD (not b, but EAP) would pull the same bad exit strategy and hit those curves going way to fast.)
Yeah, old style AP can not handle curves at all. Here in Oakland Calif, FSDβ on the other hand, slows down in anticipation of sharp turns, far, far better than non-FSD autopilot. Actually it drives such pretty comfortably. Non-FSD does not slow until it is in the sharp turn, as you describe. The result is that it enters and exits the turn late as well, and often drifts over the dividing line in the process, much to the discomfort of approaching cars. FSD handles the curves much, much better. I do wish it would take a still wider, early entry, clip the inside apex and exit the turn not quite as wide. But I gather FSD never took a defensive driving class, nor was mentored by a race driver. Or a good, smooth driver for that matter. Still, it is far better than the old "try to stay at the set speed in the center of the lane" A/P, in spite of the far better line achieved by maximizing the turn radius and pre-adjusting the speed to match. Sigh.
 
Not our experience. Three users of them in this household, fall out aside. 65% fall out rate and that inaccuracy amount - something else was be going on as that is not the normal, but rather the exception.
Not our experience. Three users of them in this household, fall out aside. 65% fall out rate and that inaccuracy amount - something else was be going on as that is not the normal, but rather the exception.

I didn't say 65% fall out, I said 65% don't fall out. They are far too expensive to have >5% fallout.
 
First, US speed limits freak me out. But that's beside the point.

When you drive that series of changes without AP/EAP or FSDb, do you attempt maintain the speed limit of Carlos Perkins (65mph) while on the off ramp or follow the 20mph suggestion that is CLEARLY POSTED TWICE?

We find FSDb doesn't read yellow signs so won't take ramps at the suggested speed. In this case, that means the car doesn't know it isn't supposed to be driving at 65mph for the hard right curve and immediate left turn. Obviously an FSD fail. (And I'm betting FSD (not b, but EAP) would pull the same bad exit strategy and hit those curves going way to fast.)

I’ve always noticed marked or not if my car enters an off ramp or aggressive turn it automatically reduces speed to maintain center lane.
Technically, the yellow signs are not speed limits, just advisories, but if you're traveling 60 MPH/100 kmh and the yellow sign says 25/40 you'd better slow down if you want to stay on the road.

As @swedge said above, FSD has gotten better with more recent releases at anticipating turns and slowing down appropriately. It still is a bit aggressive for my tastes but it's far better than it was. I don't know about the roads in CA or TX, but in MN we have lot of cloverleaf intersections that require you to slow down significantly. FSD has gotten markedly better and managing these but still fails somewhat regularly.

Cloverleafs are actually poor road design and one of the more difficult maneuvers a driver has to manage. To exit you need to watch the traffic coming off the cloverleaf, pick a spot, match speeds and merge in a very short period, all while other drivers are doing the reverse, trying to get on the freeway. To exit the cloverleaf you have to maintain your turn, gauge the traffic on the freeway, pick a spot and accelerate out of the turn appropriately while watching for other traffic merging into you. My daughter is just learning to drive and I still won't let her try them if there's any traffic at all.
 
Technically, the yellow signs are not speed limits, just advisories, but if you're traveling 60 MPH/100 kmh and the yellow sign says 25/40 you'd better slow down if you want to stay on the road.

As @swedge said above, FSD has gotten better with more recent releases at anticipating turns and slowing down appropriately. It still is a bit aggressive for my tastes but it's far better than it was.
Interesting, i find the opposite in that i feel FSD takes curves way too slowly.

But then again, i usually take curves at 50-100% more speed than the suggested yellow signs. For example, if the sign recommended a 30mph speed, i would normally take it somewhere between 45 and 60mph 😁