FSD Beta 10.69

[AlanSubie4Life]

You can model subjective feelings of safety or discomfort, but how else do you validate your model other than testing it in real life?

Maybe they have run this scenario through a simulation, and their model predicted that occupants would feel safer slowly approaching the dense traffic ahead in scenarios where the traffic behind the vehicle is light. No reason that behavior couldn't have already come from a simulation, but if it did, the assumptions they programmed into the simulation have now been proven incorrect.

A bad model is certainly possible, it just seems like this would have been ruled out in a first cut.

This is the moment Chuck took over, 5:18, increasing speed from 9 to 12mph. It is about 1.5 seconds before the car arrives in his current location (at 5:19.X).

Without intervention, the ego vehicle would have still been in the #1 (fast) traffic lane when the traffic passed, and there’s no way even any first-cut of a comfort model would have deemed that ok, regardless of the truck crossing in front. Shouldn’t be in traffic lanes when traffic arrives.

I can certainly see the argument about not surging towards the truck, but a better solution is to blast across (while keeping jerk as low as possible) lanes of traffic initially (rapid increase to 15mph rather than dawdling to 8) and then start slowing down when in #2 or #3 lane, to reassure the driver that you see the passing truck.

Regardless, agreed that if they were modeling comfort, and this result is actually matched to that model (car behavior in model and reality match), then the model was wrong.

My guess is there is some sort of mismatch in actual behavior vs. modeled behavior, and not an issue with any comfort/margin model used. But no idea of course.

 

[Bladerskb]

A simulation can only respond to the inputs given for the scenario. It can not simulate the irrational, unpredictable behavior of humans.

Not true. you can train adversarial AI agents with the entire distribution of human driving like Waymo does and run it against your ego car driving policy in simulation.
31m 44s

 

[GlmnAlyAirCar]

Because you are warning anyone in the other lane that you are merging? Doesn't seem extraneous to me.

Where else would you be going? There is no ambiguity.

 

[AlanSubie4Life]

A bad model is certainly possible, it just seems like this would have been ruled out in a first cut.

This is the moment Chuck took over, 5:18, increasing speed from 9 to 12mph. It is about 1.5 seconds before the car arrives in his current location (at 5:19.X).


View attachment 844563

Without intervention, the ego vehicle would have still been in the #1 (fast) traffic lane when the traffic passed, and there’s no way even any first-cut of a comfort model would have deemed that ok, regardless of the truck crossing in front. Shouldn’t be in traffic lanes when traffic arrives.

I can certainly see the argument about not surging towards the truck, but a better solution is to blast across (while keeping jerk as low as possible) lanes of traffic initially (rapid increase to 15mph rather than dawdling to 8) and then start slowing down when in #2 or #3 lane, to reassure the driver that you see the passing truck.

Regardless, agreed that if they were modeling comfort, and this result is actually matched to that model (car behavior in model and reality match), then the model was wrong.

My guess is there is some sort of mismatch in actual behavior vs. modeled behavior, and not an issue with any comfort/margin model used. But no idea of course.

Looking at some of the heavy traffic videos, (this is the fourth and fifth attempts) also, why is the behavior inconsistent? In these videos with traffic on the far side (and arguably actually better positioned for a collision than the pickup above!), the vehicle moves at 12mph! Why did it not get to that speed in the intervention case (I am certain Chuck would not have intervened if it had!)???

Chuck expressed that it “didn’t commit…” Is this really due to traffic on the other side of the road? The traffic is definitely not in exactly the same place, but it is pretty close.

I think it is just inconsistent for reasons that are very difficult for us to assess.

And for the record 12mph is too slow for a peak velocity. That means 2.5 seconds minimum to cross all traffic lanes and probably more like 4-5 seconds, accounting for ramp up and ramp down.

 

[AlanSubie4Life]

Where else would you be going? There is no ambiguity.

I don’t know if you’ve noticed, but people often don’t notice that lanes are disappearing. Signaling is one more way to attempt to shake these people out of their reverie. It’s courteous, and fine.

I haven’t looked up the law on this specific case, but I was taught to signal whenever changing traffic lanes. Would you signal if there were temporary traffic cones forcing you out of your lane? Where else would you be going, right?
First Google link:

Know the Road with the CHP: When a lane ends and you're merging, do you have to signal? | abc30.com

If%20there%20are%20two%20lanes%20merging%20into%20one%2C%20do%20you%20legally%20have%20to%20use%20your%20turn%20signal%20when%20you're%20merging%3F

abc30.com

Getting back to the original video, didn’t watch it, but wonder if the Tesla signaling depends on presence of other vehicles. I doubt it, but curious.

 

[Silicon Desert]

Where else would you be going? There is no ambiguity.

I don't know his point, but my thought would be signaling is generally safer than not. Aside from that, where I grew up, my parents taught me to always be considerate and courteous. I'll try to remember to be rude when in front of you.

 

[aronth5]

Nice to see Tesla is still working on Chuck's turn and the next release should have further improvements.

https://twitter.com/x/status/1562443115574599680

 

[drtimhill]

Where else would you be going? There is no ambiguity.

And there is no ambiguity that you are turning left when you are in a left turn lane, but its still a good idea to use a turn signal. it's not a question of ambiguity, its making your intent clear even if its potentially redundant. YOU may know the lane is merging, but you are assuming the driver behind you ALSO knows. If he/she doesnt, that "redundant" turn signal might just prevent an accident.

 

[aronth5]

And now for a reality check with 10.69 doing poorly. This is what happens when you have really old road layouts in a small New England town/city, Newport RI. I live near a similar road layout city (Lowell, Mass) where certain sections cannot be handled by FSD. (10.12). It will be a long time until FSD can handle these edge case cities/towns in the old northeast part of the country. And once you have snow piles it gets a whole lot worse. I wonder if Elon has ever driven in areas like this?

 

[willow_hiller]

And now for a reality check with 10.69 doing poorly.

From Elon's latest tweet it sounds like they're working on a series of known issues before releasing 10.69 wider:

https://twitter.com/x/status/1562521553358581760

 

[Ramphex]

From Elon's latest tweet it sounds like they're working on a series of known issues before releasing 10.69 wider:

https://twitter.com/x/status/1562521553358581760

Why not do internal testing to discover those issues and then release to the external team? Why is it that the release goes internal, shills, external?

 

[Mardak]

Wow. I'm quite surprised not only that this intersection layout exists but FSD Beta 10.69 seemed to handle it mostly fine:

It's 2 opposing adjacent T intersections where I would guess traffic signaling allows one T to go then the other T then the through/cross traffic. But the lanes on the ground are more for the cross traffic and I guess one is indeed supposed to basically cut straight accross the intersection from this right-turn lane.

Shannon Rd & Los Gatos Blvd, Los Gatos, CA 95032 - Google Maps

 

[KArnold]

Why not do internal testing to discover those issues and then release to the external team?

My initial thought was "because they didn't know?". But that seems unlikely - how many super users are running .69? One hundred? Two? So the remaining 80% or 90% are employees - surely they would have seen the same issues as the super users. No?

Good question.

 

[legendsk]

And 10.69.2 almost unnoticeably slips from “the week after 10.69.1” to “a few weeks”. Anyone want to make a bet on whether we’ll see it before the end of September?

Possibly. Which September were you thinking of?

 

[MP3Mike]

Why is it that the release goes internal, shills, external?

That isn't the release schedule; James, asked, and got added to the internal release group. (A select few non-employees are part of the internal release group.)

 

[Ramphex]

That isn't the release schedule; James, asked, and got added to the internal release group. (A select few non-employees are part of the internal release group.)

soooo why doesn’t it go internal then external? Why is it internal, shills, external?

 

[Dan D.]

That isn't the release schedule; James, asked, and got added to the internal release group. (A select few non-employees are part of the internal release group.)

I wonder why should it matter which group he's in? He's not an employee and not bound by an agreement to say only nice things (I think). It's not like Whole Mars doesn't sometimes say things about FSD Beta suck.

I suppose if James had said "Hey this 10.69 sucks, why isn't that thing that's going to be fixed in 10.69.1 done yet?", well yes it may sound like he's whining about having something earlier than everyone else.

But he's saying "Hey this thing in 10.69 is great, but there's lots of issues from way back that still haven't been fixed". Which is irrelevant to whichever release group he's in. I guess Elon is saying "You are in the VIP section so don't complain about garbage under your table, you're special and got in before everyone else - you be quiet now".

 

[MP3Mike]

But he's saying "Hey this thing in 10.69 is great, but there's lots of issues from way back that still haven't been fixed". Which is irrelevant to whichever release group he's in. I guess Elon is saying "You are in the VIP section so don't complain about garbage under your table, you're special and got in before everyone else - you be quiet now".

But that isn't at all what he said. If he said that Elon probably would have been fine with it.

His exact quote is: "Unfortunately I have to say I'm still having to intervene to correct #FSDBeta 10.69 in my area. Still lots of work to go.

I know this is probably not a popular opinion but the focusing on the "Chuck" complex left is getting ahead of the needs of some more basic control issues."

Nowhere did he say anything was great. He essentially said you are misallocating resources and working on the wrong problem. In other words, he said that Tesla/Elon/Ashok don't know what they are doing, and they should let him prioritize their work queue.

 

[aronth5]

soooo why doesn’t it go internal then external? Why is it internal, shills, external?

I don't understand your question, would you mind clarifying.

I'm assuming a much smaller internal group was testing before the most recent release that included approximately 1,000 users consisting of mostly employees with a few external owners. (i.e. Chuck Cook, Dirty Tesla (Chris), Omar). And be careful lumping the external testers as all shills. Check out Kim Paquette's videos as an example. It's always risky to jump to conclusions and make assumptions when we have limited information of what is actually going on at Tesla.

 

[sleepydoc]

But that isn't at all what he said. If he said that Elon probably would have been fine with it.

His exact quote is: "Unfortunately I have to say I'm still having to intervene to correct #FSDBeta 10.69 in my area. Still lots of work to go.

I know this is probably not a popular opinion but the focusing on the "Chuck" complex left is getting ahead of the needs of some more basic control issues."

Nowhere did he say anything was great. He essentially said you are misallocating resources and working on the wrong problem. In other words, he said that Tesla/Elon/Ashok don't know what they are doing, and they should let him prioritize their work queue.

This. He wasn’t providing constructive criticism, he was simply complaining. There’s a difference.

I don't understand your question, would you mind clarifying. I'm assuming a much smaller internal group was testing before the most recent release that included approximately 1,000 users consisting of mostly employees with a few external owners. (i.e. Chuck Cook, Dirty Tesla (Chris), Omar).

In past releases FSDb x.0 has gone out to a few internal testers, then x.1 is released to ‘influencers’ and finally x.2 is released widely. For some reason the first two groups were combined for the 69.0 release this time. It’s quite likely that there are relatively more bugs vs other x.1 releases.

Regardless, you can be sure they’re evaluating the feedback and looking for any significant bugs to fix before the wide release. It might be next week, it might be later - it all depends on what they need to fix. Anybody who’s getting worked up because Elons said ‘next week’ and it’s turning out to be later is either naive, clueless, or both.