FSD Beta 10.69

[AlanSubie4Life]

No and this will be somewhat controversial anyway. Many people turn them on and then ride with them on. Then of course BMW's don't have signal lights. So Beta would have hard time determining if the signal was correct, left on or not used at all.

Yeah, for FSD they’re really kind of optional, like brake lights. Shouldn’t really matter. After all: people don’t drive around looking at lights; they’re a safety feature that help tremendously with human perception, and reduce collision risk, but neither brake lights nor signal lights are needed. No one assumes people’s brake lights work or that other drivers will signal!

And with the computer’s superhuman perception, they’re even less necessary. In theory.

Certainly currently perceiving them for FSD as a signal to yield when appropriate would never work, since the false positive rate is so high. Might help at the margins as a confidence enhancer (what @momo3605 may be describing), but nowhere near that being super useful yet.

I guess I have to think through the statistics on this one though. To me my intuition says that it’s only a meaningful safety improvement if you’re already catching 99.999% (or whatever) of cut ins even without signaling. But I could be wrong.

 

[Dewg]

What Elon thinks Tesla cars can do:

What they actually do:

 

[LowlyOilBurner]

I’d gladly trade 10.69.2.4 for 10.69.2.3. Sayin.

 

[PACEMD]

And critical thinking skills, which SOOO many people lack........

Wha'choo talkin bout, Willis? This here's Merica and the Merican internet, we don't need no stinkin critical reasoning!

Apologies to folks from the free world........

 

[D Good]

Am I the only one who thinks that 69.2.4 is WAY worse that 69.2.3? Today the car literally tried to change lanes into a car in the lane next to me multiple times. I have absolutely never experienced that before. Very weird and concerning. It happens even when the cars in the goal lane are blue and everything. I will save a dashcam next time it happens. Seems like they messed with the lane change logic or something.

 

[jeremymc7]

Am I the only one who thinks that 69.2.4 is WAY worse that 69.2.3? Today the car literally tried to change lanes into a car in the lane next to me multiple times. I have absolutely never experienced that before. Very weird and concerning. It happens even when the cars in the goal lane are blue and everything. I will save a dashcam next time it happens. Seems like they messed with the lane change logic or something.

I had this happen on my last night. Totally scared as it as a huge pickup with lift kit and knob tires.

Tesla didn’t try and avoid in any way and no warnings went off either.

 

[LowlyOilBurner]

Yeah, it flips out (swings waaaaay out of the way, as if the truck is in the way, when it clearly isn’t) if a large vehicle is at a stop sign or light when FSDb is trying complete a turn. Awful last SECOND lane choices to try to get into the correct lane. Lots of regressions on 10.69.2.4. Like I said, I’d trade this for 10.69.2.3.

 

[wknickless]

Then of course BMW's don't have signal lights.

Although widely believed, this is in fact not true. The real issue is that BMW blinker fluid reservoirs are relatively small, and can only be refilled at dealership service departments.

 

[GregalseT]

I wonder if FSD is doing A/B testing? I ask because I can do a route and it performs wonderfully, then I can do the very same route and it is completely useless, dangerous, frustrating and worthless. Why?

I understand that the end result of neural net calculations is a choice made based on statistics. So could it be that the decision to perform an action is sometimes so close to being 50/50 that the "correct" decision oscillates? Or is my car A/B testing a different set of neural net results to see which performs best? Or are there variables that the car sees that I don't, that causes it to make different decisions on routes that appear to me to be identical?

It's like our car has good and bad days.

 

[Gwgan]

Seems like they messed with the lane change logic or something.

Probably exactly this if they moved lane change decision making into the neural network stack, basically removing programming logic.

 

[sleepydoc]

Although widely believed, this is in fact not true. The real issue is that BMW blinker fluid reservoirs are relatively small, and can only be refilled at dealership service departments.

I can confirm this. A friend had to bring her BMW in to have if filled on a regular basis. She tried going to AutoZone but they said they don’t carry BMW-approved blinker fluid.

 

[Supcom]

I had this happen on my last night. Totally scared as it as a huge pickup with lift kit and knob tires.

Tesla didn’t try and avoid in any way and no warnings went off either.

I'm seeing the opposite problem. the car aborts lane changes despite there being no other vehicles of obstructions in the other lane.

 

[Krash]

I spend a day watching election results and when I come back I have to snip a bunch if posts. Here is a little reminder. If your post has any of the following when you draft it, please take a minute to reread and think about it…

1. The f$&@ word. Or a GIF with that word. Or any similarly offensive words.
2. Goading about the Tesla stock price.
3. The word troll or shill
4. inferences to others’ medical conditions

Unfortunately some of the posts contained value, just not enough to easily extract from content needing to be snipped.

 

[mike01]

The lane change logic can be awful on occasion. I’m on 10.69.2.4 and I’ve had multiple instances where I’m turning left soon and about to get into the left turn lane and the car will randomly decide it wants to change to the right lane, causing me to intervene to stop it. It seems like of all the problems to solve self driving, deciding the proper lane based on the planned route should be one of the simpler to solve, but what do I know?

 

[AlanSubie4Life]

No problems like this worse with 10.69.2.4. Lane changes occur sometimes when not desired; I just keep cancelling it as needed when I am in the mood to tolerate FSDb enough to use it. No need to allow it to do its thing!

Nothing like the above though. Just changes lanes sometimes when I want to stay where I am to avoid manhole covers.

No ramming into other vehicles.

 

[willow_hiller]

No ramming into other vehicles.

10.69.3 release notes mentioned moving multiple systems over to two-stage neural networks. I wonder if they've accidentally introduced some substantial delays in vehicle controls, like they did for one of the early 10.69 releases.

 

[kabin]

Admittedly there's much randomness to FSDb but I thought 10.69.2.4 was a subtle improvement. Certainly not perfect but smoother accel/decel and may have lost some of the crazy herky jerky steering so unique to FSDb. Sometimes lane changes were smoother but many other issues remain the same.

FSDb doesn't work well enough for me to keep it engaged for an entire drive. Instead I use it where it's more reliable until another version is available for test. I dislike indecisiveness at intersections/lane changes lane changes as well as sliding forward in my seat during braking during normal city driving. Gawd knows there's oodles of brake dust on the wheels with FSDb's name on it.

 

[Dewg]

Many times people use the word "regression" to categorize a feature that used to be hard-coded and was moved to a new/existing neural net. Whenever something is transitioned to a neural net it has to be trained appropriately, which can take a few releases to lock down as they get lots of telemetry from the FSD Beta fleet.

It would be like teaching someone to walk, and when they get really good at walking, you ask them to toss a ball in the air and catch it while walking. Suddenly they're not walking as well while they learn to catch the ball at the same time. Soon after, they learn to walk well and catch the ball. Then you toss in another ball and start the process over again until they can eventually walk well and juggle 3 balls while doing so.

 

[cbdream99]

I see definitely improvement in 10.69.2.4, overall smoother, much more confident, at times I know what it is trying to do, but I think it is over zealous and if it slow down more or wait half a second then it should be fine. For example, I was approaching the red lights, there was a car waiting but the light just turn green, just at the moment my car almost stopped, as soon as it sees the other car was slow moving, then it tries to swing to the right lane but then it decided to swing back as the other car speed up, I literally saw the steering wheel spinning. I am sure if there is another car in the left lane, it wouldn’t do that, but it is enough for me to disengage.

Today, while it is trying to turn into the parking lot of the supermarket (yea!! It knows where to turn) … but the car is traveling down slope and the parking lot entrance is up hill, since it is a little tight, it swing to the left lane a bit and trying to get enough room to turn, again enough to scare me to disengage and take over, it probably can make it but I wouldn’t drive in that way even if there is no traffic next to me. Overall, I am quite impressed with this release, not problem free but progress.

 

[sleepydoc]

Drove in to Minneapolis this evening and had FSD give a red wheel panic then say AEB unavailable. Then Cruise control unavailable. Driving home 2 hours later it did the same thing. I dont’ know if this is 10.69.2.4/2022.20.19 but I haven’t had any issues before.