FSD Beta 10.69

[Knightshade]

If Tesla does not have some way to catch and log these failures, they will not be able to prioritize and make changes to prevent them in future updates. It is a flood of data, but some system to capture and catalog and evaluate these is essential to the success of FSD.

FWIW the way Tesla has historically done data capture for improvement was pushing campaigns to cars.... for example if the small test fleet that DOES have a report button (and tiny #s have reported they still do) found the system had trouble recognizing say a bicycle--- they'd send a campaign to the fleet "Send in clips of anything you think might be a bicycle with a chance of being a bike above X percent" and get back a ton of examples from different angles, in different conditions, etc... use labeling to sort bike and not-bike, and use that to train the system to better recognize bicycles.
(the other upside to this was they didn't care you had paid for FSD- or EAP or AP- all cars with the same sensors could collect the data)

It's certainly possible they're using geo-coordination of disengagement data to also look for trouble spots, that'd certainly make sense since all that data IS reported back every time (that you disengaged, where you did it, and how you did it)

I do agree that means there will be SOME really obscure cases they don't get because so few folks drive them... but offering the report button to a massive fleet was never a tenable option long term.

 

[jebinc]

Usually I report of issues I have with beta, but this time i would like to report a positive one. Yesterday, i plotted a course to HD that included roundabouts, construction zones, double stop intersections, a slow moving garbage truck with flashing lights, and more. The drive is about 15 minutes.

I am happy to report zero disengagements. Handled all obstacles well, and even when around the garbage truck confidently. Of course, you still have white knuckles with the wide turns and wheel curbing potential.

What’s most concerning is the lack of repeatability of results on the same route in similar conditions. Not ready for wide release in my opinion until their is some repeatability; good, bad, and ugly.

@WilliamG

 

[aronth5]

If her hands were in her lap, she had no chance of not slamming that cart.
She reacted very quickly, good example of how to use fsd.

Also, a good example of where a traditional steering wheel provides for better control in a situation like this. Her hands had to be removed from the yoke as it spun before she could grab the yoke to manually steer away from the shopping cart.

 

[nvx1977]

I drove a lot yesterday and had a handful of interventions (accelerator pedal) and disengagements, a couple that were forceful. Today I checked my router stats, and the car has sent absolutely nothing back to the mothership. Will continue to check, but at this point, I don't think anyone who got FSDb via safety score is considered a "tester" anymore.

 

[kabin]

All you are saying is that you do not know how to solve the problem.

Ad hominen aside, a summary might be the impossible versus possible, or science versus faith.

 

[thewishmaster]

Took a few more longer trips:
* Had another instance of false “creeping for visibility” and braking while traveling through an intersection on a fast-moving road
* Tried to stop for a stop sign that was awkwardly angled but did not apply to my road
* Well-marked speed humps in a little bit of shade are also invisible (same ones 69.2.4 was ignoring when the road was wet from the rain). It slowed down slightly for the one it saw way too late (24mph down to 21mph or so) and the second one it showed no signs of slowing for

Bonus: not quite FSDb, but on the highway the shadows lined up such that it hallucinated an entire semi on top of my car for 10-15 seconds. I checked the rear view cam after the semi disappeared and I think it was because the car’s shadow was fully visible behind me

 

[bhakan]

So having finally gotten FSDb and trying it out, I'm disappointed that I won't be able to use it much since my insurance rates it poorly and I don't want my rates to go up.

View attachment 878585
The hard braking is the biggest problem. Though even in chill mode it accelerates too aggressively.

I have seen that too and posted a while back. Accelerating on the crowded narrow neighborhood streets with cars parked and a with an incoming car my heart races too much. I have State Farm's Drive Safe app. I uninstalled their app as it was giving me a poorer score even tho I have no accidents or anything in looooong yrs of driving. The sudden acceleration in my Tesla is not well conceived in their insurance algorithm. Still my insurance rate for my next 6 Mon went up by 25% ($900 for 6 Mon for just for MSLR and I have insured 2 more cars, a home, and have umbrella).

 

[powertoold]

Today I checked my router stats, and the car has sent absolutely nothing back to the mothership. Will continue to check, but at this point, I don't think anyone who got FSDb via safety score is considered a "tester" anymore.

Or maybe the AP team is now working at Twitter, and the interns are running the fsd show.

 

[sleepydoc]

Also, a good example of where a traditional steering wheel provides for better control in a situation like this. Her hands had to be removed from the yoke as it spun before she could grab the yoke to manually steer away from the shopping cart.

Yeah - every review I’ve read by an unbiased automotive journalist has said the same thing about the yoke. It looks neat but that’s about it. As a driving control it is clearly inferior to a wheel.

 

[FSDtester#1]

Or maybe the AP team is now working at Twitter, and the interns are running the fsd show.

I already said that you post thief! Lol

 

[JulienW]

Or maybe the AP team is now working at Twitter, and the interns are running the fsd show.

Wonder how this is working. While Tweeter employees are Musk employees that is NOT the case with Tesla employees. They are employed by TSLA, a public company. Are they taking leaves of absents from Tesla? If Tesla is still paying them that is a BIG problem.

 

[Ramphex]

@Ryan27 dude you missed it today in Wesley Chapel…. People from here would’ve been maaaaaaaaaad

 

[FSDtester#1]

@Ryan27 dude you missed it today in Wesley Chapel…. People from here would’ve been maaaaaaaaaad

View attachment 878822

That looks awesome ... I couldn't make it. Had friends and family over last night.
They didn't leave until 3 am... so I slept in.
Gonna take it easy today and just want for V11 to pop up on the app being ready to download...
Did I just say that, I must still be sleeping

 

[Ramphex]

That looks awesome ... I couldn't make it. Had friends and family over last night.
They didn't leave until 3 am... so I slept in.
Gonna take it easy today and just want for V11 to pop up on the app being ready to download...
Did I just say that, I must still be sleeping

Dang, another gathering I wasn’t invited to. Guess no pets allowed type deal

 

[FSDtester#1]

Dang, another gathering I wasn’t invited to. Guess no pets allowed type deal

If you would of come, the 3 am peeps and you would still be here

 

[EVNow]

What’s most concerning is the lack of repeatability of results on the same route in similar conditions. Not ready for wide release in my opinion until their is some repeatability; good, bad, and ugly.

I think 69.3 was sent "wide" to prepare for "real wide" V11. The marginal risk is actually very small. Atleast on Teslafi 84% of FSD cars coming from 69.2. Just 16% are new. Outside of Teslafi the ratio may be somewhat different - but I guess no more than 25% are new.

Anyway - "wide" release was always a question of "how much of risk is Tesla willing to take". I'm 100% sure, with this kind of quality FSDb would not have been released to hundreds of thousands of cars by any legacy OEM.

 

[jebinc]

I'm 100% sure, with this kind of quality FSDb would not have been released to hundreds of thousands of cars by any legacy OEM.

100%

 

[Dewg]

I just got 69.3.1 and recalibrated the cameras. I can definitely see the lane selection NN is processing things differently. The car is highlighting the lane I'm currently in as if it was going to change lanes, but doesn't need to. It's just thinking about lanes a bit more. I'm also experiencing the slight wide movements, where it hits the lane marker on the left when making a right turn. I'm sure this will get smoothed out in the next update. The wide turn being said, lefts and rights are smoother for me than 2.4. A friend that refuses to let me use FSD is now comfortable with it, commenting it's "acceptable" now for him.

 

[FSDtester#1]

100%

This is what it takes to do a wide release, which only Tesla has:

 

[Pdubs]

...

What’s most concerning is the lack of repeatability of results on the same route in similar conditions. Not ready for wide release in my opinion until their is some repeatability; good, bad, and ugly.

@WilliamG

Like 99% of golf swings. I get it.