What will FSD look like when fully released?

[KArnold]

For the sake of this poll, I'm asking "What will FSD look like when released to anyone with MCU2/AP3/RCCB hardware who has bought/subscribed and obtainable by the same by just requesting it?" There's no time-frame specified here - just what will it look like when I can get it simply if I paid for it? We know Elon announced "wide release" by end of year (again), but exactly what he has in mind and what we have in mind could be very different things.

I dare say this will be in "beta" (like windshield wipers, automatic high beams) longer than I'm alive but that's not part of the question either. I'm more curious as to what our collective might think is in Elon's mind and what the end product (given the above constraints) might be. Whether that is "end of year" or some other time is obviously TBD. And based upon their progress to date, I doubt anyone would say it's "Level 5" before anyone can buy/get it.

So what is the criteria ("safe enough"? 99.99% better than human? 99.999%?) and what functionality would you expect?

 

[diplomat33]

I voted "L2 but 90% better than average human, nags". That is the option closest to what I think it will be. I think the disengagements will continue to go down as Tesla improves the software and it can handle more situations. So I think it will be better than "L2 with slight improvement". But I am not sure it will be "L2 but with >99.99% better than human". I definitely don't think it will be L4 or L5. Conceivably, Tesla could do L3 highway, I think.

I am curious about the complacency problem because if FSD Beta gets to the point where it can handle city driving with very few disengagements, drivers will likely get complacency which will make it harder for them to intervene in time for that (rare) but potentially serious issue.

 

[redmunds]

What exactly do you mean by "released" ?
Do you mean when the current "FSD beta" is rolled into FSD, which is still in "beta"?
Or, do you mean further down the line when all "beta" labels are removed from FSD?

 

[Tam]

...I dare say this will be in "beta" (like windshield wipers, automatic high beams) longer than I'm alive but that's not part of the question either...

Tesla's mind is clear that it will drive on its own to make money while the driver sleeps at home, and it's worth at least $100,000.

However, seeing what the Tesla software team is doing with its map, speed sign reader, music player, disastrous user interface transition from 10 to 11... I have very little confidence that Tesla's mind will be a reality in our lifetime.

They can't even handle sensor fusion, and they had to take the radar away!

In contrast, Xpeng was founded in 2014 when Tesla had already started its Autopilot team and already sold AP1.

Now, Xpeng is announcing its L2 door-to-door version just within 8 years.

XPeng challenges Tesla FSD with City Navigation Pilot, ‘China’s most advanced’ Advanced Driver Assistance System

Chinese EV automaker XPeng Motors announced it has launched a pilot program for its City Navigation Pilot (NGP) Advanced Driver...

electrek.co

My take: It's more hopeful if it's a non-Tesla software team.

 

[Tam]

What exactly do you mean by "released" ?
Do you mean when the current "FSD beta" is rolled into FSD, which is still in "beta"?
Or, do you mean further down the line when all "beta" labels are removed from FSD?

Currently, if you paid for your FSD and want your car to have the function to navigate from door to door, you must pass Safety Score and be selected.

Fully release means once you have paid for FSD, you don't need to pass Safety Score, and you don't need to be selected; your money did all the work for you, and you can command your car to perform door-to-door service on day one. It doesn't matter what it's called, such as FSD beta; it's the function to navigate from door to door that counts.

On the other hand, whether that function is perfected or still in beta is the question for this thread.

 

[Dewg]

Tesla's mind is clear that it will drive on its own to make money while the driver sleeps at home, and it's worth at least $100,000.

However, seeing what the Tesla software team is doing with its map, music player, disastrous user interface from 10 to 11... I have very little confidence that Tesla's mind will be a reality in our lifetime.

They can't even handle sensor fusion, and they had to take the radar away!

In contrast, Xpeng was founded in 2014 when Tesla had already started its Autopilot team and already sold AP1.

Now, Xpeng is announcing its L2 door-to-door version just within 8 years.

XPeng challenges Tesla FSD with City Navigation Pilot, ‘China’s most advanced’ Advanced Driver Assistance System

Chinese EV automaker XPeng Motors announced it has launched a pilot program for its City Navigation Pilot (NGP) Advanced Driver...

electrek.co

My take: It's more hopeful if it's a non-Tesla software team.

I think this is very much a talented group of engineers being constrained by Elon's vision. If Elon simply threw a crap-ton of money at the engineers and said "I want L4 autonomous driving on Telsa cars", we'd have a very different setup today. Instead, he told them to do the job the way he wanted, with the hardware he demanded. I'm still impressed with what my car can do today on vision alone. But seeing Xpeng and Waymo, Tesla has some catching up to do.

 

[Mardak]

https://twitter.com/x/status/1561607860101996544

At least for "FSD wide release" later this year (unclear if "Beta" was intentionally omitted), I would expect it to officially no longer be "early limited access" but there will still be a prompt to confirm with different wording but more like "Autosteer (Beta)" and "Navigate on Autopilot (Beta)" which is a different kind of "Beta."

In terms of functionality for this wide release. I would expect the functionality to be not too different from what we have now with 10.69.x.

 

[Mardak]

I wonder with wide release, will everybody that's in the FSD Beta early access will update to regular production release and future FSD updates come as part of normal software releases. If so, I guess we'll lose the video snapshot button?

Then again, Tesla has maintained early access program software updates in parallel to regular releases, and this smaller group gets updates maybe weeks ahead of wide release. (E.g., right now some people have 2022.36 with the new energy graph.) So maybe some / all of us will still get FSD updates a little bit earlier still with video snapshot button?

 

[momo3605]

I wonder with wide release, will everybody that's in the FSD Beta early access will update to regular production release and future FSD updates come as part of normal software releases. If so, I guess we'll lose the video snapshot button?

Then again, Tesla has maintained early access program software updates in parallel to regular releases, and this smaller group gets updates maybe weeks ahead of wide release. (E.g., right now some people have 2022.36 with the new energy graph.) So maybe some / all of us will still get FSD updates a little bit earlier still with video snapshot button?

I assume so, since FSD builds are happening so last minute, they aren't being added into the latest production releases. Once things settle down, it should reach a point where everyone gets the same release...but depending on if you've paid for FSD, you'll have the enable FSD button in your autopilot settings page

 

[KArnold]

I wonder with wide release, will everybody that's in the FSD Beta early access will update to regular production release and future FSD updates come as part of normal software releases. If so, I guess we'll lose the video snapshot button?

Not sure I see the relationship between "Future FSD updates as part of normal <and to all who paid> software releases" and "lose the video snapshot button"? It could roll wide and still have the snapshot button, no? If nothing else, it gives an idea of what the customer preceives to be a problem as opposed to what their AI says is a problem. Those may not be the same thing.

I voted "L2 but 90% better than average human, nags".

I'm with you but honesetly would be disappointed if that's what happens. It also sounds as if we're already making excuses for Telsa to deliver whatever they want to deliver and call it "done". I guess in retrospect, that's likely true.

As of this writing, we're about 50% of guessers landing on a form of L2 and 50% L3+. Interesting.

 

[diplomat33]

I'm with you but honestly would be disappointed if that's what happens. It also sounds as if we're already making excuses for Tesla to deliver whatever they want to deliver and call it "done". I guess in retrospect, that's likely true.

I think reality is settling in and many are re-aligning their expectations with reality. I know I have. Part of it is that Elon created incredibly high expectations with talk of "solved FSD", "L5 capable", "robotaxi", "summon your car cross country" etc... Now it is becoming more and more obvious that reality will fall short. And rather than admit that the whole FSD thing was a big lie (which they can't do for legal reasons), I think Tesla's best course of action now is to quietly release FSD. Once they release "autosteer on city streets", they can technically claim they released all the features on the website. Then they can start work on the next FSD with better hardware.

 

[nvx1977]

I'm with you but honesetly would be disappointed if that's what happens. It also sounds as if we're already making excuses for Telsa to deliver whatever they want to deliver and call it "done". I guess in retrospect, that's likely true.

I don't think they would consider L2 at 90% human safety "done." They will keep working on it (march of 9's). But L2 @ 90% human safety is arguably a good threshold for wide release. Wide release w/ HW3, more improvements with HW4, 5, etc. Also not suggesting that the hardware is the sole bottleneck for improvement.

 

[Daniel in SD]

L5 (beta)
1000x worse than a human without a safety driver (which is why it’s still beta)

 

[Tam]

...50% L3+...

Mercedes has already sold L3 since May in Germany. It's waiting for L3 permits for USA sales.

Mercedes Drive Pilot Level 3 Autonomous Tech Officially On Sale In Germany

The Stuttgart-based automaker became the first in the industry to receive international approval for Level 3 autonomous tech in December last year.

www.motor1.com

It's doubtful that Tesla will allow its system to be classified as L3 or above because while Mercedes takes responsibility for accidents caused by its L3 system, Tesla blames its owners.

Tesla can claim it's L5, but when talking to the government, it says it's classified as L2.

Tesla Tells California DMV that FSD Is Not Capable of Autonomous Driving

After years of touting its long-awaited Full Self-Driving feature, Tesla is telling California regulators a different story about its capabilities.

www.caranddriver.com

 

[momo3605]

I voted "L2 but 90% better than average human, nags". That is the option closest to what I think it will be. I think the disengagements will continue to go down as Tesla improves the software and it can handle more situations. So I think it will be better than "L2 with slight improvement". But I am not sure it will be "L2 but with >99.99% better than human". I definitely don't think it will be L4 or L5. Conceivably, Tesla could do L3 highway, I think.

I am curious about the complacency problem because if FSD Beta gets to the point where it can handle city driving with very few disengagements, drivers will likely get complacency which will make it harder for them to intervene in time for that (rare) but potentially serious issue.

How far away do you think FSD beta is from 90% better than average human? Tesla has already released it to safety score 80+ group (which is about 50% of users). I expect FSD will go wide soon (like within the next year). No way it gets to 90% better than average human that soon. Based on their progress so far, i expect it'll be maybe a tiny bit improved in 1 year.

 

[diplomat33]

How far away do you think FSD beta is from 90% better than average human? Tesla has already released it to safety score 80+ group (which is about 50% of users). I expect FSD will go wide soon (like within the next year). No way it gets to 90% better than average human that soon. Based on their progress so far, i expect it'll be maybe a tiny bit improved in 1 year.

It's really hard to quantify since we don't have any statistical data on FSD Beta's actual reliability. Maybe ~1 year from "90% better than average human"? But I am leaving some room for the possibility that in the next 3 months, FSD beta could improve quite a bit. After all, the software will continue to get better at different tasks with more training. Plus, I would expect that Elon will want to do wide release when FSD beta is much better than just "slight improvement" over what it is today. FSD Beta today is not ready for wide release IMO. But wide release could really be anything between "current reliability" and "90% better than human". Honestly, I don't think we really know what Tesla's standard is for wide release.

 

[Yelobird]

My Opinion only with no fact or concern of what was said, assumed, or promised is that they will get to L3 for most all driving but maintain Beta title as a stop gap to liability to protect the company. To many people Will want to fail this system and Will find a way to do so. I do also believe in the distant future the system will be segmented allowing L4 maybe even L5 for highway commute only switching back to L3 liability for any other driving path. Interstate driving has Way fewer variables compared to city use with directions, pedestrians, and unknown events. Just a guess but I suspect we will get advanced operations but only in more controlled (predictable) environments.

 

[momo3605]

My Opinion only with no fact or concern of what was said, assumed, or promised is that they will get to L3 for most all driving but maintain Beta title as a stop gap to liability to protect the company. To many people Will want to fail this system and Will find a way to do so. I do also believe in the distant future the system will be segmented allowing L4 maybe even L5 for highway commute only switching back to L3 liability for any other driving path. Interstate driving has Way fewer variables compared to city use with directions, pedestrians, and unknown events. Just a guess but I suspect we will get advanced operations but only in more controlled (predictable) environments.

Out of curiosity, why do you think Tesla will go L3 route? I’ve seen 0 mentions of L3 in FSD DMV documents nor from Elon himself. He’s always used the word L5. Is there something I’ve missed? I would love to have L3, but I haven’t seen anything to suggest Elon even has L3 on his radar

 

[Daniel in SD]

Out of curiosity, why do you think Tesla will go L3 route? I’ve seen 0 mentions of L3 in FSD DMV documents nor from Elon himself. He’s always used the word L5. Is there something I’ve missed? I would love to have L3, but I haven’t seen anything to suggest Elon even has L3 on his radar

Tesla told the DMV that the version of FSD shown in the demo video at Autonomy Day was L3.
But yeah they’ve never given any indication that they want to develop an L3 system, it’s robotaxi or bust!
I do think that if another automaker releases a full speed divided highway L3 system then Tesla will reconsider.

 

[diplomat33]

Out of curiosity, why do you think Tesla will go L3 route? I’ve seen 0 mentions of L3 in FSD DMV documents nor from Elon himself. He’s always used the word L5. Is there something I’ve missed? I would love to have L3, but I haven’t seen anything to suggest Elon even has L3 on his radar

I think L3 is just wishful thinking. FSD won't be L5. So what will it be? So Tesla owners create this "head canon" that Tesla will do L3. But you are right, Tesla has not said anything about L3.