FSD Beta 10.69

[Mardak]

Dirty Tesla skipped commenting on some 10.69 behaviors in the most recent "10.69 Dominates Downtown Driving" video where early on, FSD Beta 10.13's new behavior for "Improved in-lane positioning on wide residential roads" seems to be causing a regression for parking lots almost driving into a parking sign pole:

Although parking lot behavior has always been pretty wonky with 10.x, and that might not be addressed until later in 11.x with single stack (initial release for highway and later improvements for parking lots?). Also interesting that the occupancy network didn't show the gray "little bumpies" in the visualization for that concrete sign base.

On the flip side, 10.69 is doing some maneuvers so naturally that maybe they aren't even worth pointing out anymore and just fast-forwarded such as being able to switch 2 lanes to the left within 200 feet after turning into the right-most lane and needing to get to the left-most:

 

[jabloomf1230]

I wonder if there aren't as many of these style intersections here because there's plenty of center turn lanes allowing merging to the main street?

Before we have more discussion regarding the legality of doing a ULT and then merging using a TWTL, has anyone here either been ticketed or warned for doing so in a state where such behavior is deemed illegal?

 

[GlmnAlyAirCar]

I have to disagree. The point of the creep barrier - according to how people are talking about it - is to show the maximum distance the car is going to creep. That is what the car has determined is the maximum safe distance it will go. So people can trust it's not going to end up too far into the road.

However we saw in Chuck's video that the car did creep past the barrier and ended up too far into traffic. If it cannot see well enough at the creep barrier mark then it needs to disengage and tell the driver to complete the turn.

It's an example of a poor safety system.

We need more info. Like you said, it can creep up to the barrier. When the car decides to go, it is obviously going to pass the barrier. The question of when the barrier comes down is a separate question. Must it disappear before the car goes? I don’t see why it does. However, that might be the normal behavoir.

 

[JulienW]

Dirty Tesla skipped commenting on some 10.69 behaviors in the most recent "10.69 Dominates Downtown Driving" video where early on, FSD Beta 10.13's new behavior for "Improved in-lane positioning on wide residential roads" seems to be causing a regression for parking lots almost driving into a parking sign pole:
View attachment 844185

Why are you saying "...almost driving into a parking pole"? He did NOT have to disengage and didn't hit it. While it shouldn't be driving in the parking spaces and it didn't visualize the pole there is no way to conclude conclusively that it was problem UNLESS he had to disengage to avoid hitting it.

EDIT: Just to add using it in a parking lot is out of the scope of the Beta anyway and therefore meaningless to use as a reference for how Beta is improving or regressing.

 

[jabloomf1230]

We need more info. Like you said, it can creep up to the barrier. When the car decides to go, it is obviously going to pass the barrier. The question of when the barrier comes down is a separate question. Must it disappear before the car goes? I don’t see why it does. However, that might be the normal behavoir.

If the "barrier" was always there but invisible, you could have fooled me. My car has always been erratic in its creeping behavior. Sometimes it creeps bit, stops then creeps some more. Other times, it starts creeping only to immediately rush into the intersection.

It's possible that the creep line is shown only to give the driver comfort in the car's creeping behavior, but I tend to think that it actually represents a change in maneuvering.

 

[Dan D.]

We need more info. Like you said, it can creep up to the barrier. When the car decides to go, it is obviously going to pass the barrier. The question of when the barrier comes down is a separate question. Must it disappear before the car goes? I don’t see why it does. However, that might be the normal behavoir.

In Chuck's video (link a few dozen posts ago) the car went through the "barrier" and stopped beyond into the road, speed zero. So, yes it needs to go through the barrier at some point to cross the road, but if it's still creeping and stopping while it passes the barrier then it's not a barrier.

I argue that if it gets to the barrier (that it has established itself, and pre-advised you that's the limit) and still cannot see, it cannot be allowed to creep any further. It's done. Disengage and tell the driver it can't do it.

 

[sleepydoc]

Dirty Tesla skipped commenting on some 10.69 behaviors in the most recent "10.69 Dominates Downtown Driving" video where early on, FSD Beta 10.13's new behavior for "Improved in-lane positioning on wide residential roads" seems to be causing a regression for parking lots almost driving into a parking sign pole:
View attachment 844185
Although parking lot behavior has always been pretty wonky with 10.x, and that might not be addressed until later in 11.x with single stack (initial release for highway and later improvements for parking lots?). Also interesting that the occupancy network didn't show the gray "little bumpies" in the visualization for that concrete sign base.

On the flip side, 10.69 is doing some maneuvers so naturally that maybe they aren't even worth pointing out anymore and just fast-forwarded such as being able to switch 2 lanes to the left within 200 feet after turning into the right-most lane and needing to get to the left-most:
View attachment 844186

Agree with @JulienW - I’ve never considered parking lots to be someplace that FSD could/should drive in. Sometimes it will do OK, but it’s one of the many examples that clearly show how cars without steering wheels are far from reality. The truth is, parking lots are one of the most difficult places to drive. Tight spaces, tons of pedestrians including kids that behave erratically and cluelessly, lots of cars backing, lots of relatively blind intersections, etc.

The ability to cross multiple lanes quickly is awesome. I have a couple of areas where I have to make a right turn, then a left turn 2 lanes over in short order and FSD 12 routinely fails this maneuver.

 

[willow_hiller]

Got this Tweet as a promotion from Mobileye today, I thought it was a really interesting point of comparsion:

https://twitter.com/x/status/1559139372196323328

You can really see the limitations of traditional image-space segmentation, as highlighted by Ashok's CVPR presentation. They seem to be doing the vehicle and driveable space segmentation in individual 2D views.

You can clearly see:
1. The density at the horizon causing a lot of noise in the distance of the VIDAR representation
2. The VIDAR representation exhibiting some fisheye warping, presumably due to the stitching together of of individual 2D viewpoints
3. Driveable space not persisting behind vehicles and other obstacles

 

[n.one.one]

Elon laying down the law (or at minimum his strong opinion).

 

[Mardak]

Why are you saying "...almost driving into a parking pole"?

I was just trying to highlight FSD Beta got awfully close to it although maybe @DirtyT3sla didn't even comment on it as he didn't feel it was actually that close (or maybe he didn't even realize it was there as it's an odd spot to have the concrete base to begin with??). It seemed like FSD Beta happened to avoid it because it realized the actual driving lane for the parking aisle was narrower because the parking spots ended just ahead with the trees and curb so started shifting back towards the left. But that coincidence of shifting left at the right time seems to also show some limitation of the initial occupancy network behavior in 10.69.

It very well could just be distortion from the in-vehicle camera making it seem like this parking pole was just inches away to the right than it actually was:

I know quite well FSD Beta doesn't work great in parking lots, and Chris does too, so that could also be a reason for a lack of commentary on this odd behavior.

 

[TS.42]

Elon laying down the law (or at minimum his strong opinion).

View attachment 844214

Both of them have excellent points though. Come on 10.69, send it my way! I will pay the $200 for another month of testing to see if you can permanently hook me.

 

[BnGrm]

Elon laying down the law (or at minimum his strong opinion).

View attachment 844214

Didn’t seem like complaining to me. Seems like honest feedback. Thin skinned Elon response. I guess critical feedback doesn’t help with the imminent price increase.

 

[MP3Mike]

Didn’t seem like complaining to me. Seems like honest feedback. Thin skinned Elon response. I guess critical feedback doesn’t help with the imminent price increase.

It wasn't constructive feedback. It was telling Elon that they are allocating their resources incorrectly. (As if James knows what they are working on internally and he knows better than Elon and Ashok how the resources should be allocated.)

Constructive feedback would have been something more like "It is great that you have made so much progress on Chuck's ULT, but what would really make it better in my area would be to add support for U-turns", or whatever it is that he thinks is a "basic control issue" that he didn't provide any details on.

edit: I see in a follow-up tweet he mentioned "problems getting into a right turn lane". I think I have watched most of his videos, and yes, he complains that it gets into the right turn later than he would like it to. But I don't think he has shown any examples of it actually having a problem getting into the lane.

 

[Supcom]

Didn’t seem like complaining to me. Seems like honest feedback. Thin skinned Elon response. I guess critical feedback doesn’t help with the imminent price increase.

He was complaining that Tesla worked on ULTs instead of his priority problem areas.

Pointing out what does and does not work is fair. But when you complain about them making great progress on a fundamental part of self-driving at the expense of not fixing some other issue, you are not being fair. There are limited development resources so everything is worked based on an internal priority ranking.

In the case of 10.69, Tesla didn't just work on Chuck's ULT. They implemented some fundamental architectural changes to set themselves up for solving a ton of other issues. The guy complaining obviously did not understand that.

 

[BnGrm]

He was complaining that Tesla worked on ULTs instead of his priority problem areas.

Pointing out what does and does not work is fair. But when you complain about them making great progress on a fundamental part of self-driving at the expense of not fixing some other issue, you are not being fair. There are limited development resources so everything is worked based on an internal priority ranking.

In the case of 10.69, Tesla didn't just work on Chuck's ULT. They implemented some fundamental architectural changes to set themselves up for solving a ton of other issues. The guy complaining obviously did not understand that.

Agreed and all fair points but there’s a lot of low hanging fruit for improvement. I can’t wait to receive .69 and see for myself.

 

[Dan D.]

I was just trying to highlight FSD Beta got awfully close to it although maybe @DirtyT3sla didn't even comment on it as he didn't feel it was actually that close (or maybe he didn't even realize it was there as it's an odd spot to have the concrete base to begin with??). It seemed like FSD Beta happened to avoid it because it realized the actual driving lane for the parking aisle was narrower because the parking spots ended just ahead with the trees and curb so started shifting back towards the left. But that coincidence of shifting left at the right time seems to also show some limitation of the initial occupancy network behavior in 10.69.

It very well could just be distortion from the in-vehicle camera making it seem like this parking pole was just inches away to the right than it actually was:
View attachment 844217

I know quite well FSD Beta doesn't work great in parking lots, and Chris does too, so that could also be a reason for a lack of commentary on this odd behavior.

Do we feel that FSD Beta saw the pole though? Or to put it another way CAN it see the pole? Can it see the pole at 25mph, 40mph? I barely saw it the first time since the video is sped up and the base is grey & pole is silver, wasn't very easy to see.

What are the limits to what FSD can detect and avoid?: The thin vertical, horizontal, angled objects, protruding 2x4's from truck beds, wires hanging low across the road. I wonder what will happen in the situations that exceed the limits... (well I suppose that's a rhetorical question, it will not detect & avoid them...)

 

[JHCCAZ]

He was complaining that Tesla worked on ULTs instead of his priority problem areas.

Pointing out what does and does not work is fair. But when you complain about them making great progress on a fundamental part of self-driving at the expense of not fixing some other issue, you are not being fair. There are limited development resources so everything is worked based on an internal priority ranking.

In the case of 10.69, Tesla didn't just work on Chuck's ULT. They implemented some fundamental architectural changes to set themselves up for solving a ton of other issues. The guy complaining obviously did not understand that.

The guy complaining is one of the established YouTuber FSD testers, and I don't think he has a general record of complaining. P.erhaps his comment could have been worded in a different way, expressing encouragement and support and then asking for future attention to be turned to his issues. But it also seems that Elon gave a response that was overly irritated. Remember however that he's on the receiving end of an unbelievably high load of traditional media and social media criticism every day. I just picked up in surfing, a couple of articles about a massive ridicule campaign over his jet flying a short hop from San Jose to San Francisco. There are a lot of possibilities behind that and he's hardly a leading climate hypocrite compared to others.

Elon's response seemed harsh to me, but probably not as bad as the defensive "You're holding it wrong" comment from Steve Jobs.

 

[Silicon Desert]

https://twitter.com/x/status/1561854942402424834

Some moron turns into the wrong lane. Beta brakes immediately, but this guy should be holding his steering wheel and should have been ready to swerve to the right if the idiot kept going.

This is one of the things that bothers me here. Calling people idiots and morons for making a mistake as if you and others on here never made a driving mistake. Just one more person to put on ignore.

 

[Mardak]

He was complaining that Tesla worked on ULTs instead of his priority problem areas

I wonder if Elon Musk knows that James Locke has been sending back video snapshots of FSD Beta not signaling for 2 lanes merging into 1 as recently as March 2022 but also going back now literally years to December 2020 (and probably October 2020). I would not be surprised if Tesla has over 1TB of video snapshots (with actual ongoing costs to Tesla) from his vehicle just for this one priority problem of his to use the turn signal.

Hopefully this doesn't mean 10.69.1/.2 will be delayed for the rest of us.

 

[MP3Mike]

I wonder if Elon Musk knows that James Locke has been sending back video snapshots of FSD Beta not signaling for 2 lanes merging into 1 as recently as March 2022 but also going back now literally years to December 2020 (and probably October 2020). I would not be surprised if Tesla has over 1TB of video snapshots (with actual ongoing costs to Tesla) from his vehicle just for this one priority problem of his to use the turn signal.

If that is his priority problem, then he really doesn't have any issues. (Almost nobody in my area signals at lane merges, so FSD Beta would fit right in.)