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

FSD Beta 10.69

This site may earn commission on affiliate links.
Regarding the release notes and how technical they are, I feel like an important point is being missed here about about why they do this.

Tesla engineers write the release notes before the release goes out and gets widely tested. Obviously, they are of limited use to end users that want to know in their actual driving what will be better, is the car more reliable now, where are there changes, etc. I get that and it's a valid complaint.

The problem is you can't answer those higher level questions before collecting data about the actual performance of the system. Part of the point of having such a large group of testers is Tesla can make very specific change (say, increasing the training data set for lane topology perception in snowy conditions), do some initial testing to look for obvious safety issues, ship it to 100k+ cars, and look for improvements at scale in the telemetry they get back.

Maybe the change makes 1% of left turns less jerky; maybe it makes 50% of intersections smoother. I would be really hard to predict precisely how the average user would actually experience these changes - plus it can depend so much on your local road conditions. If this were easy, they'd just tell us and they wouldn't even need a big real world test group.

That's why Tesla gives us a change log of what they changed and their intent with it. It's either that, or vague "Bug fixes and improvements". They don't tell us the impact of those changes, but plenty of Youtubers do if that's what you're interested in.
Back to joining the party: It's been end-of-the-year stuff all day, up to and including cleaning the lint out of the clothes drier.

In a very weird way, I don't mind reading the obtuse, hard-to-decipher release notes. It's clear that whoever it is or they are that are doing it, they know what's going on. They're using the lingo they've invented or manipulated to describe what they're doing. Given the research level of FSDb development, it's likely that the lingo is changing over time, as breakthroughs in method and concepts kick in.

It's kind of like a U.S. person traveling to Australia, going into a bar, and trying to understand what the locals are talking about. The slang and idiom will be obtuse, hard to understand.. but it's got grammar, fits in with words that the listener does know, and, with a little experimentation (and getting laughed at), the listener learns the language. And concepts. And how they fit together.

None of this is a surprise. A little bit of thought reveals that this is how kids learn. Admittedly, we'd all like to have a reference manual to speed up the uptake; but given the flex and weff of what's going on in the software coders' heads, any ref doc so written would probably get obsolete pretty quickly.

So, the release notes currently resemble a firehose: Open wide, and hang on. It's the experts speaking; they're not slowing down (much) for us; but one can glean a fair amount of information each time a collection of notes gets issued.

I've been messing with new technology since I was a high-school student back in the 1960's and was around for the beginnings of the computer age: Programming on an Altair, keypunch and tape punch on local and remote IBM mainframes, soldering on studs and terminal strips to wire wrap to through-hole circuit boards to surface mount and fine-pitch to BGAs and flip-chips. All of the new technology started off in research labs with obtusely written papers; language got changed and pushed down into undergraduate, then high school, and right into elementary school, madly receiving pedagogical improvements as the tech ideas and concepts matured. Frankly, FSDb is just another one of these major hunks of technology coming into common use; complaining about the lingo is exactly like old farts complaining about terms like RAM, ROM, and "double-U-double-U-double-U dot" when the technology got pushed to the masses.

It's the new stuff. Enjoy it and hang on by your fingertips. Your grandkids won't believe you when you tell them about the beginnings 😀.
 
This was the longest post, (by far), that I’ve read and went all the way to the last word. You got my attention and kept it. (not a small feat)

SQUIRRL!

Well done.
All right. Ya got me. "SQUIRRL"?

There is an oddball usage of the word in debate.. which, given that the conversation is obtuse usages of the English Language, this is a good dig. But did you have something else in mind?
 
Wrong. He's got (at least) a BS in Mechanical Engineering. That's STEM, sir.
Wow, he does indeed. I stand corrected. I read a while back (years ago), that he didn’t.

9776D07D-A1C9-4C2C-A6D6-CA096535ED8A.png
 
All right. Ya got me. "SQUIRRL"?

There is an oddball usage of the word in debate.. which, given that the conversation is obtuse usages of the English Language, this is a good dig. But did you have something else in mind?
No Sheldon lol. The squirrel word was in reference to my propensity to distractions.
 
Drove 10.69.25.1 on the same rural route that I did on the previous version. The car made all of the exact same mistakes as the previous version. There was no discernible difference. I had one disconnect each way on a 100 mile round trip that included FSDb and NOA. The disconnects were for moving into a turn lane when the car should have continued straight. A couple other interventions of accelerator when the car hesitated or wanted to stop at an overhead flashing yellow.

Whatever changes are in this version seem to be very minor. Perhaps it's more bug fixes from the production branch software carried over. So, nothing to get excited about one way or the other.
 
Drove 10.69.25.1 and saw some sections of my routes flowing well and a few “oh crap” moments where I had to intervene because we were nervous about the outcome. Lots of traffic and potential for issues based on the car wanting to pull out into traffic. It still accelerates on a UPL like we are coming from the NASCAR pits onto the track.

I was concerned about it at a two-way stop and the car tried to pull out slowly into oncoming Left/Right traffic. I had to intervene again but that was a definitely failure. Other turns were smoother though so I see some real improvements but need to continue testing to find the gaps.
 
In my fairly long drives today … all 3 disengagements were because of wrong lane choices.

BTW, one thing that has improved in later 69.x releases is turning with other vehicles turning in multi-lane turning at signals. Earlier it was bad enough that I’d normally disengage… it’s been good enough to let it turn on it’s own in the last couple of releases.
 
In my fairly long drives today … all 3 disengagements were because of wrong lane choices.
(Happy New Year everyone!)
Yes, lane selection is pretty clearly the most common (though not the most dangerous) misbehavior of FSDb currently. I find myself repeatedly pouncing on the left stalk to cancel the undesirable lane move, while admonishing the car just like I'm training our over-excited puppy.
"Stay - staaay- no no no - You Stay! Good boy!"

One very helpful thing I found since getting it, of which I was prviousoy unaware, is that you can request and execute a user-commaned a lane change the same way you do in standard FSD Autosteer. This allows me to stay in FSDb and mostly avoid the issue of the car waiting too long to select the correct lane for an upcoming turn.

If this user-commanded lane selection could then be combined with a way to suppress otherwise optional/unwelcome lane changes by the system (meaning those that are not needed to follow the navigation route), most of my drives would be free of disengagements or turn-signal-cancel type interventions.

The car would still be FSDing and handling the maneuvers, but the driver would be guiding its planning instead of jumping in to disagree with it. It would also avoid the amateurish "I'm confused" signal flashing that happens when you react to cancel an unwanted FSDb lane change.
 
Does anyone feel like it's getting worse rather than better in regards to drifting too far right in very wide lanes like two lane streets with unmarked curb parking? Even worse when it comes to a red light and blocks the ability for people to turn right while basically sitting 5 feet over from the center line. Driving down these streets can also be terrifying since it'll hug parked cars within a couple feet while still being a stone's throw from the center line.
 
  • Like
Reactions: jebinc
Drove 10.69.25.1 on the same rural route that I did on the previous version. The car made all of the exact same mistakes as the previous version. There was no discernible difference. I had one disconnect each way on a 100 mile round trip that included FSDb and NOA. The disconnects were for moving into a turn lane when the car should have continued straight. A couple other interventions of accelerator when the car hesitated or wanted to stop at an overhead flashing yellow.

Whatever changes are in this version seem to be very minor. Perhaps it's more bug fixes from the production branch software carried over. So, nothing to get excited about one way or the other.
seems so simple doesn't it? Tell the difference between a single yellow light and an actual traffic light.
One of my regular trips has three of those flashing yellow warning lights, one is on a 65mph road, the other two are on a 70mph road.
Its so much fun when it tries to stop for them.
 
  • Like
Reactions: FSDtester#1
Fairly sure Karpathy had no involvement in actual planning / execution. He was basically the NN / perception lead.

Sorry, it was tongue n cheek. The 2nd and 3rd characters 'ar' was a compelling similarity to Parkinson's. :) Sorry for mentioning the P word. I've had in-laws pass from the disease as well worked professionally with P patients. TMI.

But more to the point, there have been steering assists for at least 6 years now and I'm not aware of any other AV design that so directly connects the inner path insanity/uncertainty to the steering feedback loop. The two subsystems have much different bandwidth needs. Steering never needed those crazy herky-jerky motions or humans would drive like that. Again, I suspect the reason is a desperate need to hide system latency issues. Normally a good systems engineer and/or vehicle integration engineer would have solved this well before the first customer touched the steering wheel (i.e. years ago).
 
Last edited:
  • Like
Reactions: FSDtester#1
Does anyone feel like it's getting worse rather than better in regards to drifting too far right in very wide lanes like two lane streets with unmarked curb parking?

Yeah degraded in 10.69.3.1. Not sure current status.

Even worse when it comes to a red light and blocks the ability for people to turn right while basically sitting 5 feet over from the center line.

Lane positioning has been bad from the beginning and remains so. They haven’t added the contextual awareness required to get this right.