FSD Beta 10.69

[aronth5]

I don’t think they are that incompetent. As you also noted, it may be an issue with latency delay. Although, I don’t understand exactly why it appears to take nearly 2 seconds of delay before they apply the brake aggressively. I have noticed a similar braking delay in other contexts such as cross traffic appearing just after my car begins moving into an intersection. That kind of latency may be tolerable in some contexts but for urgent braking due to a pending potential collision that is totally unacceptable. It takes people only a few dozen milliseconds to apply the brakes in similar circumstances. Either the car lacks sufficient cpu or they need to build in a fast path for urgent braking.

Interesting comparison to a human driver. I believe you're giving too much credit to how fast a human reacts. Of course FSD should be faster.

"Studies have shown that it takes the average driver from one-half to three-quarters of a second to perceive a need to hit the brakes, and another three-quarters of a second to move your foot from the gas to the brake pedal. "

https://www.automotive-fleet.com/driver-care/239402/driver-care-know-your-stopping-distance

 

[Mardak]

"Studies have shown that it takes the average driver from one-half to three-quarters of a second to perceive a need to hit the brakes

I was counting frames of various 10.69.1 videos of where traffic lights were visualized differently, e.g., actually red but still showing yellow in the visualization, and they seemed to range from about 15 frames to 20 frames, so at 30fps, that's ½ to ⅔ second delay. Unclear if FSD Beta driving behavior also has that latency in making a decision, and with the jitter of 5 frames, that could result in entering the intersection when it's already red. And more broadly, if there's this latency and jitter for traffic lights, is there a similar issue with other predictions?

 

[Jeff N]

I was counting frames of various 10.69.1 videos of where traffic lights were visualized differently, e.g., actually red but still showing yellow in the visualization, and they seemed to range from about 15 frames to 20 frames, so at 30fps, that's ½ to ⅔ second delay. Unclear if FSD Beta driving behavior also has that latency in making a decision, and with the jitter of 5 frames, that could result in entering the intersection when it's already red. And more broadly, if there's this latency and jitter for traffic lights, is there a similar issue with other predictions?

Just looking at the video without carefully counting frames I see a nearly 2 second delay between the stop sign being visualized on the center display and substantial braking being visualized on the display. Of, course, there could be differences in delay between sending and displaying the speed vs the NN sending and displaying the stop sign.

 

[AlanSubie4Life]

Yes, plenty of evidence that FSD is a couple seconds slower than a human. Numerous well documented and carefully analyzed recent examples.

However, it’s impossible to say where the delay arises (perception, perception confidence, planning algorithm, etc.). We can’t see what the car is seeing (we can only see the screen which of course is largely unrelated) so really hard to pinpoint anything.

My guess is, it is planning; since the human is driving there is minimal requirement for a fast response. But really have no idea.

 

[powertoold]

The delay seems to be a bug with 10.69.1

I don't think we saw it with 10.69

 

[AlanSubie4Life]

The delay seems to be a bug with 10.69.1

I don't think we saw it with 10.69

I think this is incorrect (generally speaking). Not going to check, but the head-on with TOSV was 10.69 and no reaction until extremely late there (of course, no reaction was necessary).

Not saying there aren’t delay bugs in 10.69.1; of course it could be causing extra issues.

To me it just doesn’t seem that fast reactions are being prioritized at the moment, which makes sense. It is less necessary at the moment because the car is doing driving assist so much more slowly than a human. And of course the human is driving, which is a big help! Fast reactions could also make the car much more jittery if there is a lot of uncertainty in perception at longer ranges.

 

[queue]

same ol' story, dangle the carrot, then, *crickets*.

 

[queue]

I know composing a 20 second post with a progress status is a huge burden on the team, but could someone make an effort? Maybe an administrative aide?

 

[AlanSubie4Life]

same ol' story, dangle the carrot, then, *crickets*.

It’s pretty clear from Elon’s Tweets that it could be a couple more weeks if not more. There were promises of a faster release but obviously you just take the worst case commitment.

 

[ChaoticNeutral]

How much the market will accept depends on the take rate they want. It’s a sliding scale not one fixed amount.

As an investor I don’t want Tesla to offer perpetual option. As a customer I’m happy to have bought it.

That’s a weird thing to admit, but hypocritical capitalist is a common mindset I suppose

 

[AlanSubie4Life]

Maybe we’ll get another preview this evening?

https://twitter.com/x/status/1566109350409691142

 

[willow_hiller]

In speaking of Chuck and braking, he's just put out a new video discussing braking and slowdowns:

 

[Silicon Desert]

In speaking of Chuck and braking, he's just put out a new video discussing braking and slowdowns:

I see the same pauses on an undivided road in my neighborhood when a car passes me, yet I don't see the pause with double green lights. However, the two sets of green lights I pass up here seem to be a little farther apart than Chuck's. The neighborhood slow down on a narrow street in more annoying to me. On the wider neighborhood streets, there is no pause.

 

[willow_hiller]

Chuck's confirmed 2022.20.11 is 10.69.1.1:

https://twitter.com/x/status/1566118546916122624

Unsurprisingly release notes haven't changed.

 

[Jeff N]

Interesting comparison to a human driver. I believe you're giving too much credit to how fast a human reacts. Of course FSD should be faster.

"Studies have shown that it takes the average driver from one-half to three-quarters of a second to perceive a need to hit the brakes, and another three-quarters of a second to move your foot from the gas to the brake pedal. "

https://www.automotive-fleet.com/driver-care/239402/driver-care-know-your-stopping-distance

Good article! I find in many of the cases where I notice problematic delays with FSD and braking are cases in which I am already sensing a potential need to brake and have my foot near the brake.

 

[JohnB007]

Based on the numbers reported by TeslaFi, looks like this (10.69.1.1) is yet another super careful 1k push.

So perhaps we are getting 10k push next Monday and 100k push a week after?

Can't see any videos based on this version yet. Chuck, where are you!

EDIT : I hope this is the .69.0 stability with .20 functionality...

 

[aronth5]

Based on the numbers reported by TeslaFi, looks like this (10.69.1.1) is yet another super careful 1k push.

So perhaps we are getting 10k push next Monday and 100k push a week after?

Can't see any videos based on this version yet. Chuck, where are you!

EDIT : I hope this is the .69.0 stability with .20 functionality...

Chuck has tweeted his Memorial Drive on 10.69.1.1 was good this time.

https://twitter.com/x/status/1566129390907449345

 

[DarkForest]

I’m glad we’ll be getting 2022.20 instead of 2022.16, but I also think it’s funny that 2022.28 dropped today on the main line.

Such is the FSD Beta life.

 

[AlanSubie4Life]

Chuck says he cannot say it is any different than 10.69, FWIW. Maybe subtle differences with acceleration.

 

[JulienW]

Kinda ironic but looks like Locke was one of the first to get 10.69.1.1. Guess getting on Elon's bad side isn't so bad after all.