April Fools FSD Download button.

[AlanSubie4Life]

I think that's almost certain. The old AP stack will be gradually retired in favor a new stack that uses the same AI/NN basis as FSD. However, Tesla will probably make this an FSD only feature (maybe "AP+" ?) because it needs HW3 and they won't want to take on the burden of more HW3 upgrades.

Yeah, inevitably it will happen, but the timeframe seems really uncertain. I could imagine them not touching it for a long time since it is "good enough," and they have other things to focus on (they don't want to have to spend a bunch of effort regression testing freeway driving). But we'll see. I hope it doesn't take years to get upgrades to freeway AP ("AP+" or whatever they call it), but it might!

 

[momo3605]

Yeah, inevitably it will happen, but the timeframe seems really uncertain. I could imagine them not touching it for a long time since it is "good enough," and they have other things to focus on (they don't want to have to spend a bunch of effort regression testing freeway driving). But we'll see. I hope it doesn't take years to get upgrades to freeway AP ("AP+" or whatever they call it), but it might!

I think Tesla will eventually move autopark, basic summon, and maybe lane change into standard autopilot, but i think a lot of it has to do with the competition catching up. Elon usually makes moves when the competition announces something. i.e. Lucid and Ford's price drops pushed Elon to drop the price of the model S, and then release the SR Model Y. And now possibly stock price drop/Honda's L3 in Japan announcement has pressured Elon into expanding FSD Beta. But most of the standard lane keep assists you get with like Honda, Kia, etc... are quite basic. The only system i know of that can even change lanes is Super Cruise, but they charge quite a bit of money for that as an option, and eventually becomes a subscription. So I don't think Tesla has any incentive to move features down to standard AP for at least a while.

 

[goRt]

Yeah, inevitably it will happen, but the timeframe seems really uncertain. I could imagine them not touching it for a long time since it is "good enough," and they have other things to focus on (they don't want to have to spend a bunch of effort regression testing freeway driving). But we'll see. I hope it doesn't take years to get upgrades to freeway AP ("AP+" or whatever they call it), but it might!

I think that freeway driving is just a small rule set add on once city street driving and video feed are done - that's the way things appear to be architected to me.

 

[drtimhill]

I think that freeway driving is just a small rule set add on once city street driving and video feed are done - that's the way things appear to be architected to me.

Agreed, and the FSD stack is far smarter at understanding stuff around it. Though, from what I understand, there will have to be a whole bunch of new AI/NN learning done for freeways, since much of the freeway/AP code is not based around NN processing (beyond the basic recognition).

 

[israndy]

Perhaps the April fools comment is coming true

https://twitter.com/x/status/1370522745709158407

Man, there are a LOT of threads on this topic

 

[powertoold]

Going with pure vision is a big deal. It means they've "solved" vision.

 

[AlanSubie4Life]

Going with pure vision is a big deal. It means they've "solved" vision.

I guess that depends on what you mean by "solved." If by "solved," you mean they are going to using vision-only (assuming that is true for a moment), then I guess that is true, if a bit tautological.

 

[powertoold]

I guess that depends on what you mean by "solved." If by "solved," you mean they are going to using vision-only (assuming that is true for a moment), then I guess that is true, if a bit tautological.

Yup, solved is opinion here. I’d say it’s solved if the vision can predict the distance and size of an object as well as radar for the purposes of achieving human level or better fsd.

The reasoning is that if you have vision-based NNs capable of above, then it’s also capable of visually recognizing / predicting many other real world events / objects.

For example, AlphaGo evolving into AlphaZero.

 

[qdeathstar]

If they are using vision why is Tesla investing in not lidAr high definition radar?

 

[AlanSubie4Life]

Yup, solved is opinion here. I’d say it’s solved if the vision can predict the distance and size of an object as well as radar for the purposes of achieving human level or better fsd.

The reasoning is that if you have vision-based NNs capable of above, then it’s also capable of visually recognizing / predicting many other real world events / objects.

For example, AlphaGo evolving into AlphaZero.

That seems like a very low bar; radar is highly imperfect and limited (some would call it a crutch).

To me it seems like vision-only is essential to get working. But it's not clear to me that rolling out vision-only means that vision has been solved. I'm not saying it's an unsolvable problem, either. I would bet that it hasn't been solved though.

 

[powertoold]

If they are using vision why is Tesla investing in not lidAr high definition radar?

All rumors, except for the mmWave radar filing, which was meant for interior cabin use.

 

[BobDolesGhost]

Perhaps the April fools comment is coming true

https://twitter.com/x/status/1370522745709158407

Man, there are a LOT of threads on this topic

W T F... /sigh

 

[Microterf]

April fools!!! No button yet

 

[pilotSteve]

Sigh. Elon, just give us an honest update for once about FSD availability. When is this jerking around going to end?

 

[SunCharged]

Sigh. Elon, just give us an honest update for once about FSD availability. When is this jerking around going to end?

Seriously, it’s getting old.

 

[run-the-joules]

Seriously, it’s getting old.

Just like all of our cars, while we wait.