How smooth is 10.69.2? My wife fell asleep in the passenger seat today.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.69
- Thread starter Buckminster
- Start date
AlanSubie4Life
Efficiency Obsessed Member
No, because it happens with no lead traffic. Ignore the first unnecessary slowdown, which was due to lead traffic. I’ll post better examples once I capture them.
Last edited:
MrTemple
Member
You are completely missing the point I’ve tried several times to make. So trying to make it again seems like a waste of both our time.
(Hint: Imagine what the graph would look like if Tesla only ever released the first beta version to all those different people at those same expansion times AND as at today there were only one bollard boop.)
MrTemple
Member
Technically, Musk didn’t say that 10.69.2.1 was the version that would go out to the 80+…
It’s a fair assumption. But given Musk’s tendency for highly qualified statements (that people often mis-parse as promises without qualification), I’d lay even odds as to whether the ambiguity was a careful choice of words or just the result of a brief tweet.
Sorry, I have to say I failed to feel the problem without the in car motion feedback. The stop seems quite smooth visually.
AlanSubie4Life
Efficiency Obsessed Member
Just watch the regen bar. It tells you everything. When it is long, you do not want to see it jump to short, then to long again. When it is very short, fluctuations matter less.
In this video, you’d see the regen bar get long three times:
1) Response to lead vehicle slowing, and maybe the light (probably not).
2) Initial slowdown transient. But then you see it jump to get short again (this is bad and is lost opportunity to use regen!)
3) Final slowdown bar gets long again. Then gradually shortens to nothing.
What should have happened is one elongation of the bar to a fairly long but perhaps not max length, and then stay that way for a bit, then gradually shorten to nothing.
The rapid changes in the length will directly tell you how lacking in smoothness the stop is. You can do this in any FSD Beta video and “feel” the drive. An initial large jump is not necessarily bad, but the more gradually it goes to max length, the better. You just don’t want to see the long-short-long pattern. We’re not trying to signal SOS here.
Deceleration is mandatory, but high jerk is not.
Those HDMI hacks are definitely helpful for this!
Last edited:
Daniel in SD
(supervised)
Since I’m apparently too dense can someone explain to me MrTemple’s point regarding cumulative FSD beta miles?
It's hard to see your visualizations, but what I think is happening is that the lead car (which is highlighted in black) started to slow down and make a lane change. FSD wants to slow down for it until the car is completely in the new lane (like leaving your hand on a chess piece after making the move so you can take it back), so the car stays black until it's fully moved and then switches to a lighter grey. At that point the car feels it's okay to speed up, but then almost immediately hits the threshold for slowing for the light - so you had a brief moment where the regen dropped as if it was going to accelerate, and then decline again for the light.
AlanSubie4Life
Efficiency Obsessed Member
Remember I said to ignore that first pulse - that is why it happened, as you say (of course it was WAY overdone, but a separate issue). Just look at the TWO subsequent pulses. There are two - that’s the problem. That is a huge amount of lost stopping power over the course of less than a second. And clearly could have used less brakes instead (obviously zero brakes was super trivial here too - why did it use them?).
Again, there were TWO drops in regen (three major regen peaks). If there had just been one, this wouldn’t have even met the threshold for a stopping issue.
There are obviously better video examples, which I will capture (happens quite a lot so not difficult!). I’ll move the camera slightly closer at the expense of field of view. But you can’t really watch these easily on a cell phone (you can, just not ideal). Need at least an iPad or something.
It’s frustrating because occasionally it slows down just fine.
Last edited:
AlanSubie4Life
Efficiency Obsessed Member
No, I can’t figure it out. In response to his hint: I think the miles plot would look the same if Tesla had just progressively rolled out 10.2 to progressively larger groups. People would be diligently using it to provide Tesla with training data, and accepting the somewhat rough turning and other limitations. Those limitations are not an impediment to racking up the miles; I think that is clear. 10.2 was likely good enough for wide release, based on what I am hearing. It’s just exciting to have a piece of software that drives you around, so people are going to use it!
MrTemple
Member
So all these people in this thread, on Twitter, on the Tesla Facebook groups saying that they are using it more…
They are lying and you know the real score that they would be using a way less capable version just as much?
AND that many miles driven on that way less capable version would have resulted in exactly zero more collisions.
This is a very curious standpoint.
But let’s assume it’s true.
Why then would Tesla not have released 10.2 to everybody who bought FSD. Since it’s equally capable and equally safe, why wouldn’t they HUGELY increase the number of miles driven, and accelerate the harvesting of edge-cases?
They just didn’t feel like it?
I mean. Come on.
Daniel in SD
(supervised)
I think he’s saying that even if the miles per user is constant the expansions bring in users who would have driven less on previous versions. The evidence against this hypothesis is 10.11 was released with no expansion of users and while that release was
it did not result in an increase of usage. It will be interesting to see how much people use it once it is available to everyone (assuming Tesla keeps publishing this stat).
powertoold
Active Member
MrTemple
Member
One downloading 10.69.2.1 on TeslaFi TeslaFi.com Firmware Tracker
2 downloaded on Teslascope 2022.20.16 Official Tesla Software Update | Teslascope
2 downloaded on Teslascope 2022.20.16 Official Tesla Software Update | Teslascope
AlanSubie4Life
Efficiency Obsessed Member
But it looks like Tesla’s data may not show any increase, and I think that seems more credible than what people say on Facebook (who uses Facebook anyway?).
But you were supposed to gather the user counts so we could create a definitive plot. Then we can see.
AlanSubie4Life
Efficiency Obsessed Member
Sure hope Chuck gets out there to do his busy turns before he gets this! This could solve the CULT! I was feeling good but now worried.
MrTemple
Member
From what are you drawing this conclusion?
User counts have literally nothing to do with my point.
—
You didn’t answer my question. If 10.2 was equally capable and safe as the latest, why wouldn’t Tesla have rolled that as wide as it has gone now?
AlanSubie4Life
Efficiency Obsessed Member
Aren’t user counts required to determine average usage per user?User counts have literally nothing to do with my point.
Daniel’s plots looked pretty flat when user counts were stable. But it’s a little hard to be definitive without dividing the plot by the number of users at each point.
I think they probably wanted to improve the driver monitoring (which, yes, indirectly impacts safety but I don’t think that’s what you mean by “safe”). That’s clearly been the biggest shift since 10.2 - huge changes in how monitoring is done (accompanied by strike resets as they iterate). Obviously there have been other improvements too, but I’m not sure they would greatly increase the usage, at least in the current beta group.
They also probably wanted to get a sense for how much people would use it and in what scenarios, and their degree of attention, before taking a risk on a larger group.
It might be a bit more safe now, but it’s a bit hard to tell. If an accident occurs on a future version, that will potentially suggest it is less safe than prior versions (it would technically be infinitely less safe)!
Last edited:
That's quite literally what people do when waiting to make a left turn from the main road. I believe that's basically what FSD Beta is looking for these median crossover regions -- presumably, if the main road has space for a car to wait to turn to the side street, someone coming from the side street could theoretically pose almost parallel to the main road.
I've seen 10.69 think about doing this 2-step maneuver even on residential 25mph streets with 1 through lane in each direction and a temporary left-turn lane into the side street. I would never have thought to wait in the middle of the road in these cases, and practically FSD Beta will just roll on through as there isn't that much traffic (and probably would confuse others if it did it).
Similar threads
- Article
