FSD Beta Videos (and questions for FSD Beta drivers)

[qdeathstar]

So does every single AP feature, auto windshield wipers, and the Dashcam.
It's not like V10 won't get exposed to traffic 5 hours after release. Are they going to watch the videos and possibly pull it if it does bad at night?

yes, it’s what what Elon does during his refractory period.

 

[Dan D.]

At a very high level-

Previous system- Each camera ran its own single-frame NN to act as visual perception for the car. Radar was a primary sensor for forward speed and distance of objects. Nothing persisted over time.

Current beta FSD system- All camera inputs are fed into a series of cascading NNs that essentially process a BEV 4D perception of the world (think of it as 360 surround video that persists over time). Radar is not used at all. Speed and distance are determined from the video inputs creating a point cloud.

There were steps in between, where for example some BEV views existed by still frame by frame, and still using some radar inputs as well. But that's the general "How it started and how's it going" of the design philosophy.

This also required rewriting of a lot of the training code too--- the upside being if it's rewritten to understand 360 view and time, you only have to manually label something in frame 1, and the new code can self-label that object for the rest of the video as long as it remains in view of any camera (and even to make predictions about it reappearing if it moves behind something briefly- they gave examples of this on AI day).

Here's Karpathy discussing the overall transition with examples in practice from mid 2020:

Do you happen to know if the Tesla approach is fundamentally different from say the method Waymo and Mobileye are using? I presume they all use neural net processing. I would guess that the Mobileye Lidar/Radar version works from a whole-scan sensor fusion and works down to individual elements, whereas Tesla's (and maybe Mobileye's) vision systems work on camera inputs and create the whole-world later. Possibly Waymo combines all sensor inputs at the beginning?

What's the simplest way of explaining the different approaches? (Without needing to qualify them as better/worse, just how are they doing it in a few sentences, as I'd just like the overview).

 

[n.one.one]

A little late but it looks like it will happen. Tweets from various people.

 

[n.one.one]

These live downloads are typically pretty boring and I don't recommend watching them...but here is one for the hardcore among us.

 

[mhan00]

I wonder if Brandon is going to do a quick drive after the download? I seem to remember Tesla not allowing people to do live streams of drives, but I don’t know if that’s still an issue or not.

 

[heltok]

Let’s not get too high expectations. Would be great if it nails left turns and monorail, but any incremental improvement would be great.

 

[n.one.one]

I wonder if Brandon is going to do a quick drive after the download? I seem to remember Tesla not allowing people to do live streams of drives, but I don’t know if that’s still an issue or not.

He confirmed that live streams of driving are not allowed but he is going to go through all of the release notes and UI while he is still live.

 

[mhan00]

Brandon just tweeted that it’s a lot smoother and he can’t wait to try it in traffic.

https://twitter.com/x/status/1436622376695844866

 

[diplomat33]

V10 has a nice zoom out feature for intersections:

https://twitter.com/x/status/1436620672109744129

 

[aronth5]

Lombard street- Zero interventions. Having seen how much FSD has struggled with Lombard street in previous attempts the improvement is very encouraging.

 

[powertoold]

Lombard street- Zero interventions. Having seen how much FSD has struggled with Lombard street in previous attempts the improvement is very encouraging.

Holy cow!!!! No interventions, no hesitations, 10x better than any other video I've seen of Lombard St.

This bodes really well for parking lots. I think.

 

[powertoold]

 

[Terminator857]

Lots of disengagements.

 

[powertoold]

Odd phantom brake here for yellow light:

https://twitter.com/x/status/1436669239121879046

 

[powertoold]

Lots of disengagements.

Mostly for the monorail test (still fails). It's much much improved over 9.2.

The streets of Seattle seem pretty challenging: lots of hill crests, rough roads, islands, road / bike markings, monorail pillars, etc.

 

[dckiwi]

I may have missed it, but has anyone confirmed whether v10 is single stack for city streets and highway? This is the biggest change we were anticipating for v10, so I’m surprised I haven’t seen any confirmation one way or the other.

My bet is no single stack yet, but would love to be wrong.

 

[powertoold]

I may have missed it, but has anyone confirmed whether v10 is single stack for city streets and highway? This is the biggest change we were anticipating for v10, so I’m surprised I haven’t seen any confirmation one way or the other.

Haven't seen any highway videos yet. I'm sure we'll get them within next few hours.

 

[diplomat33]

I may have missed it, but has anyone confirmed whether v10 is single stack for city streets and highway? This is the biggest change we were anticipating for v10, so I’m surprised I haven’t seen any confirmation one way or the other.

My bet is no single stack yet, but would love to be wrong.

According to Whole Mars who has V10, it does not have the single stack:

 

[diplomat33]

https://twitter.com/x/status/1436672190821306370

 

[dckiwi]

Haven't seen any highway videos yet. I'm sure we'll get them within next few hours.

Ok cool. If I was a beta tester I would have ma

According to Whole Mars who has V10, it does not have the single stack:

Thanks for finding this.

Not surprising but does make me wonder whether they’ll expand the beta without it (ie the button in a couple of weeks). I suppose it’s not a gating factor to widening the beta, as NoA is already out there as-is.