FSD Beta 10.69

[EVNow]

I've been trying to use NOA (instead of just AP) on freeways to get an idea how it is now so I can compare to FSDb when V11 finally drops.

The result has been mixed. Once on freeway, it seems to be ok most of the time - but taking the entry and exits have been problematic.

Today FSDb tried to change lane to take the entry ramp too late and had to reroute. While exiting, again tried to change lanes too late and had to reroute. Its not like the entry/exit lanes were busy. They were fairly sparse, but the lane change was just too late - and at that time there were vehicles.

I wonder if V11 would be any different.

 

[aronth5]

I've been trying to use NOA (instead of just AP) on freeways to get an idea how it is now so I can compare to FSDb when V11 finally drops.

The result has been mixed. Once on freeway, it seems to be ok most of the time - but taking the entry and exits have been problematic.

Today FSDb tried to change lane to take the entry ramp too late and had to reroute. While exiting, again tried to change lanes too late and had to reroute. Its not like the entry/exit lanes were busy. They were fairly sparse, but the lane change was just too late - and at that time there were vehicles.

I wonder if V11 would be any different.

I never had trouble with the decision to exit except that NoA/FSD exits much too fast on some. My biggest problem is moving to a faster lane when a car is approaching to quickly and I cancel the lane change. Otherwise it works very well as long as it's not raining when I often lose FSD.

 

[sleepydoc]

I had bad phantom braking around 7-8 updates ago. I honestly can't remember the last one.

At one time I was experiencing PB every 10 minutes or so. It’s been much better of late but I’m also using FSD very heavily and I don’t consider phantom braking on FSD to be the same as PB on TACC.

 

[willow_hiller]

At one time I was experiencing PB every 10 minutes or so. It’s been much better of late but I’m also using FSD very heavily and I don’t consider phantom braking on FSD to be the same as PB on TACC.

Also worth considering what we define as phantom braking. FSDb brakes in a lot of situations that I probably wouldn't brake for, but I wouldn't call it "phantom" in the sense that there is no apparent reason. I can see and understand the reason for the braking, I just don't think it's absolutely necessary.

 

[EVNow]

Also worth considering what we define as phantom braking. FSDb brakes in a lot of situations that I probably wouldn't brake for, but I wouldn't call it "phantom" in the sense that there is no apparent reason. I can see and understand the reason for the braking, I just don't think it's absolutely necessary.

FSDb behavior at one of the entrances was really weird. There is a carpool only ramp and that is the lane FSDb tried to change to. noncarpool entrance is one more lane away.

What might have tricked NOA/FSD was there were close merges near the exit. This is where experience at that particular exit/entrance helps. We know to change lanes, even though a merger be coming soon.

 

[uscbucsfan]

I’d want to see that data before making assumptions about safety. I expect that TACC is safer than Autopilot or dumb cruise but I don’t know for sure. Who knows though, maybe dumb cruise with automatic emergency braking and forward collision warning is safer because it keeps the driver more engaged.

I think there's some common sense that TACC is safer than dumb cruise control. Often, on lower priced cars, cruise control doesn't include adaptive components...you manually have to reduce your speed or hit the brakes or you will run into someone.

Regardless, there's a reason why the industry is moving away from dumb cruise on luxury cars.

 

[AlanSubie4Life]

Regardless, there's a reason why the industry is moving away from dumb cruise on luxury cars.

Presumably because it is nicer to use and there is demand for it.

Safety is unclear though I would guess TACC would be safer, even with FCW and AEB.

No idea though!

 

[JulienW]

Presumably because it is nicer to use and there is demand for it.

Safety is unclear though I would guess TACC would be safer, even with FCW and AEB.

No idea though!

Yea, unless you live and drive in the "country" dumb cruse control can be more of a PIA to use than just driving. I bet many rear ends happen when people are on dumb cruse control while checking their cell phone. It is just not "blamed" on cruse control even if it contributes.

 

[sleepydoc]

Yea, unless you live and drive in the "country" dumb cruse control can be more of a PIA to use than just driving. I bet many rear ends happen when people are on dumb cruse control while checking their cell phone. It is just not "blamed" on cruse control even if it contributes.

umm...no - distracted driving was at fault, not cruise control.

While I agree that Adaptive cruise seems like it should be safer, there are many things that are 'intuitively obvious' but turn out not to be true.

I haven't owned a car with adaptive cruise before my Tesla and I used cruise control extensively, even in town because it was an easy way to keep myself from speeding. I never had an issue forgetting to brake and I have a hard time understanding the argument that it would somehow make people forget to brake.

I will say that when I switch from our Tesla to our Odyssey with dumb cruise I have to remind myself that it doesn't stop automatically.

 

[PACEMD]

It wouldn't phantom brake every 5 minutes.

Dang, you have phantom braking every 5 minutes? That's heart braking..........

 

[AlanSubie4Life]

Really not impressed with this performance. I mean, it was obviously going to happen…

No phantom braking here! No braking at all!

Not superhuman yet…

 

[swedge]

Has there been evidence yet of further rollout of 2022.40.5 ( FSDβ v11 ) since the single case spotted at 11/11 at 11:11?

I wonder about the implications of v 11 for the beta program. For example, TeslaFi shows only 6 copies of 10.69.3, all on 11/2. If v 11 is being rolled out (albeit slowly) wouldn't make any feedback on 10.69.3 moot, and so obviate any further release of .3 ? And, would further feedback from us on 10.69.2.4 be of any use to Tesla, as bug reports on obsoleted code?

My update history with 10.69 was:

10.69.2 on 9/12/2022,​

.2.3 on 9/22,​

x.2.3 on 10/8, and​

10.69.2.4 on 11/1 .​

​

Quicker than one per month. On the other hand, from 10.12.2 (6/9) to 10.69.2 was over 3 months.

I am ready for v 11, a bit tired of the seeing the same old problems... So I'm hoping to see some improvements and probably new (and old) misbehaviors to report. Mostly, however, the new non-FSD features would be nice. Hopefully the 3 month delay of the last "major upgrade" won't be repeated....

SW

 

[swedge]

Chuck Cook and his ULT made the Big Times, i.e. a story in New York Times yesterday titled:

What Riding in a Self-Driving Tesla Tells Us About the Future of Autonomy​

Sadly, it may be paywalled:

What Riding in a Self-Driving Tesla Tells Us About the Future of Autonomy

We took a ride with Tesla’s experimental self-driving system. It was by turns intriguing, impressive and scary.

www.nytimes.com

Interestingly, the url calls it ".../tesla-self-driving-flaws.html". As a Tesla fan, I'd have preferred ".../tesla-self-driving-progress.html". Thankfully the editor used a neutral title.

SW

 

[Mardak]

would further feedback from us on 10.69.2.4 be of any use to Tesla, as bug reports on obsoleted code?

Since mid-October, Tesla has turned off collecting manually pressed video snapshot button trigger, but other triggers are still active for situations Tesla specifically cares about. This could point to Tesla being in crunch mode potentially focusing on FSD Beta 11 wide release, so triggers might focus on highway driving or low visibility scenarios.

Theoretically, those data collection triggers could apply to the rest of the fleet and not just those in FSD Beta, so it does seem like "extra" testing of FSD Beta 10.69.x isn't as valuable right now. I still push the video snapshot button just to check if the car ends up sending data back, but nothing so far, so I've also not gone out of my way to test things either.

 

[Mardak]

Chuck Cook and his ULT made the Big Times, i.e. a story in New York Times yesterday titled: What Riding in a Self-Driving Tesla Tells Us About the Future of Autonomy

They originally rode with Chuck back in August, so not entirely clear if it was with early versions of 10.69 or 10.12.x. Although the end of the article mentions "Just last week, he and his car revisited a few of the scenarios we encountered in August," so that might have been with 10.69.3 still failing at the parking lot / road confusion.

 

[swedge]

No braking at all!

Really? Am I missing something?

Regenerative braking (shown by the green bar) is braking, isn't it.

It it slowed from 45 to 28. Slowing started while the car ahead was crossing the far-side crosswalk and the painted lane divider ahead was obscured. And they didn't signal the lane change. Not superhuman, agreed, but it looks pretty good to me.

SW

 

[AlanSubie4Life]

Since mid-October, Tesla has turned off collecting manually pressed video snapshot button trigger, but other triggers are still active for situations Tesla specifically cares about. This could point to Tesla being in crunch mode potentially focusing on FSD Beta 11 wide release, so triggers might focus on highway driving or low visibility scenarios.

Theoretically, those data collection triggers could apply to the rest of the fleet and not just those in FSD Beta, so it does seem like "extra" testing of FSD Beta 10.69.x isn't as valuable right now. I still push the video snapshot button just to check if the car ends up sending data back, but nothing so far, so I've also not gone out of my way to test things either.

Overall this was a decent article.

Seemed to cover the subject reasonably well for someone not familiar with it. And provided the reasonable estimate that car companies are years from success in this endeavor.

Seemed to stick to the facts too. Good to see them have Chuck at the wheel (even though I think he’s a bit too lax with it, at least he is calm and collected with minimal drama). Way better than some prior pieces.

 

[AlanSubie4Life]

Regenerative braking (shown by the green bar) is braking, isn't it.

Note when I disengaged (using the brake pedal). Hard to hear with the music but just keep an eye on the blue steering wheel. You can also see me move my hand from my thigh (old, bad habits die hard) to the wheel, since I knew I would need to make a steering intervention (my other hand was always at the 9 'o clock position of course).

You can also tell it is not the car braking, because the dark bar does not show up as I slow from 45mph to ~28mph.

This was a combination of light brake pedal use and regen. I saw it coming as soon as the Nissan started moving into the intersection (you can tell at 0:05 that something is not right), so it was not dramatic. I just let it go until the latest moment I could, without having to jam on the brakes. There was no response from the vehicle. And in the visualization it doesn't look like it was flagged as a vehicle of note (though I don't pay any attention to how these visualizations work).

It'll be interesting to see when AI can detect that things are not right at 0:05 in this video. It's kind of amazing how well the human brain works. I can't really describe how I know, though I guess for the most part it was the car trajectory and its precise positioning as it moved through the intersection - it was just not right.

If I had had the car under manual control I would have eased off as soon as I detected the weirdness, and probably would not have exceeded 40mph or so. And then just made a lane change (or just canceled the one I was making) without any braking. I normally would not have passed on the right here at this speed, but the light changed earlier than I expected and at that point I had decided to get in that lane (I made the lane change request, I don't think it was FSD). (I needed to make a right a ways up the road, but there was no rush...it's just easier to be in the right lane early, especially with FSD, as it handles that particular right turn lane up the road a mile particularly poorly.)

 

[willow_hiller]

They originally rode with Chuck back in August, so not entirely clear if it was with early versions of 10.69 or 10.12.x. Although the end of the article mentions "Just last week, he and his car revisited a few of the scenarios we encountered in August," so that might have been with 10.69.3 still failing at the parking lot / road confusion.

He tweeted some follow-ups. The failures they highlighted were fixed in 10.69.3, but they chose not to publish the clips he sent them:

https://twitter.com/x/status/1592498405993246721

That, and the author was screaming "for effect" and Chuck called him out about it:

https://twitter.com/x/status/1592524826719453187

 

[swedge]

Since mid-October, Tesla has turned off collecting manually pressed video snapshot button trigger, but other triggers are still active for situations Tesla specifically cares about. This could point to Tesla being in crunch mode potentially focusing on FSD Beta 11 wide release, so triggers might focus on highway driving or low visibility scenarios.

Theoretically, those data collection triggers could apply to the rest of the fleet and not just those in FSD Beta, so it does seem like "extra" testing of FSD Beta 10.69.x isn't as valuable right now. I still push the video snapshot button just to check if the car ends up sending data back, but nothing so far, so I've also not gone out of my way to test things either.

You remind me that besides bug reports, Tesla does want us to log miles on FSD to support the measurement of the accident rate. As far as I have heard, Calif DMV is still considering their "deceptive advertising" concern. I expect DMV will drop the concern if Tesla can prove that the marketing of "FSD" has actually resulted in fewer accidents, even if the name does some confuse some journalists. Such proof requires a whole lot of miles on FSD to be statistically significant. So I'll continue to engage it, thanks!

Perhaps the names "Full Self Driving" and "Auto Pilot" are actually bit of clever trolling on Tesla part. It tricks journalists into complaining, but in effect educating the public at large about EV's and auto-automation. Any PR is good PR, especially if it is free.

SW