FSD - Level 2

[shrineofchance]

Yes, the DMV emails are consistent with Tesla saying that FSD would roll out as L2 and require driver supervision at first. We all know that FSD Beta will roll out as L2 and will require driver supervision when it goes wide to the fleet. That's not the issue.

Some people are, quite wrongly, making that the issue.

The question is when will FSD go from L2 to L5? Elon has said that FSD will reach L5 this year. Do you read the DMV emails as saying that FSD will reach L5 later this year? Or are the emails saying that Tesla will not achieve L5 this year as Elon has said but will be L2 for years to come?

Obviously, the emails provide no new information on that subject. They merely reiterate what Tesla has said publicly many times in the past. It really couldn't be more clear:

“Please note that Tesla’s development of true autonomous features (SAE Levels 3+) will follow our iterative process (development, validation, early release, etc.) and any such features will not be released to the general public until we have fully validated them and received any required regulatory permits or approvals.”

You should take Elon's predictions with the same grain of salt now that you did before these DMV emails came out. The emails change nothing.

 

[mikes_fsd]

Do you read the DMV emails as saying that FSD will reach L5 later this year? Or are the emails saying that Tesla will not achieve L5 this year as Elon has said but will be L2 for years to come?

LOL, you are really trying, I'll give you that.
The DMV emails are Tesla answering specific DMV questions, not the future state of where they (Tesla) want to bring FSD to.

Like Elon says, ask the right questions, or figure out what question to ask first!

 

[diplomat33]

Some people are, quite wrongly, making that the issue.



Obviously, the emails provide no new information on that subject. They merely reiterate what Tesla has said publicly many times in the past. It really couldn't be more clear:

“Please note that Tesla’s development of true autonomous features (SAE Levels 3+) will follow our iterative process (development, validation, early release, etc.) and any such features will not be released to the general public until we have fully validated them and received any required regulatory permits or approvals.”

You should take Elon's predictions with the same grain of salt now that you did before these DMV emails came out. The emails change nothing.

The previous paragraph says that Tesla is not planning to add any more OEDR to FSD Beta before "final release". The line you quote about L3+ is a new paragraph. So I think it is fair to ask if Tesla is planning to add more to FSD Beta after "final release" to make it L3+ or if FSD beta will remain L2 after final release and Tesla will "move on" to L3+. That's what some people are worried about, that Tesla will release FSD Beta to the fleet and then "move on" and not make FSD beta more than L2.

 

[shrineofchance]

The previous paragraph says that Tesla is not planning to add any more OEDR to FSD Beta before "final release". The line you quote about L3+ is a new paragraph. So I think it is fair to ask if Tesla is planning to add more to FSD Beta after "final release" to make it L3+ or if FSD beta will remain L2 after final release and Tesla will "move on" to L3+. That's what some people are worried about, that Tesla will release FSD Beta to the fleet and then "move on" and not make FSD beta more than L2.

"Final release" just means the public beta version of FSD that will be made available to all customers. The pipeline goes something like:

1. QA
2. Alpha
3. Early Access (private beta)
4. Final release (public beta)

In other words, Tesla is merely saying that the version of FSD Beta in Early Access that is Level 2 will remain Level 2 when it is released to the public (obviously).

Much ado about nothing.

 

[diplomat33]

"Final release" just means the public beta version of FSD that will be made available to all customers. The pipeline goes something like:

1. QA
2. Alpha
3. Early Access (private beta)
4. Final release (public beta)

In other words, Tesla is merely saying that the version of FSD Beta in Early Access that is Level 2 will remain Level 2 when it is released to the public (obviously).

Much ado about nothing.

Yes I get that. But what about after "final release"? Will Tesla continue to add more to FSD Beta after the "final release" to make it more than L2 later? Based on how Tesla does things, I would say yes, Tesla will continue to improve FSD Beta after "final release". But I think it is a legit question to ask.

 

[M109Rider]

Yeah we know that. It's just that the marketing for FSD was misleading because Tesla made it sound like FSD would be L5 relatively soon. Heck, the CEO is saying that L5 will happen this year. Yet, now we see that FSD will be L2 for awhile and L5 is just what Tesla hopes FSD will eventually become someday. For people who purchased FSD in 2016 when Tesla was describing it as essentially L4/L5 and just needing some validation and regulatory approval, it is understandable why they would be pissed.

Well, we clearly don’t know that, thus the OG posting. ...

The OG post was about an article pulled from the DMV that he felt implied Tesla was never going to deliver anything above level 2.
That’s not what Tesla was saying of course.

Your reply is not addressing the context of the original post. You shifted to maybe imply this is about the same old song that FSD’s delivery date keeps getting moved.
I get that particular complaint, and it’s valid, but it’s not the context of the original post, it just seems like another segway to whine about FSD delivery dates. Why do the same folks here keep going back to this over and over at every opportunity. ?
It must just put a smile on their face, or make their day somehow I guess.

I’m Canadian, we don’t make cars here. Wish we did. Some of the folks here that seem to enjoy bashing Tesla should really truly try to see a big picture here.
I have nothing against Chinese, Japanese, or European auto makers. I’ve bought many myself.
But the Asian market is likely soon going to engulf the electric car market, because the American auto makers, (EXCEPT FOR TESLA) are really dropping the ball again. I say again, because that’s quite simply why the Japanese car market took off in the U.S in the first place. Same reason, and their doing it again on the electric home front. Just look at the products the Asian market is coming out with. The exception being THE AMERICAN MADE TESLA) of course.
Every time I see a post that bashes Tesla, (who is the only American auto maker not only competing head to head with the Asian and European market, but dominating), I wonder why folks here are doing their part in helping the Asian market get a foot hold by bashing Tesla.
For such a patriotic country, I sure don’t understand that ...
Not saying we shouldn’t hold them accountable, but we can still support American made at the same time. I don’t know how relentlessly jabbing at the same rib is supporting them....

Anyway, just a disclaimer so things are clear, I love the USA, Europe, and Asia. Visited them all many times. Owned European cars, American cars, and Asian built cars. However, I still believe in supporting your own where it makes sense. Tesla isn’t perfect, but they (Elon) represents the first change in this industry (for the better) I’ve seen in my lifetime. I’m choosing to support them, and do my part to help what I see as positive change even with their flaws, and I’ve been doing that on this forum for a while.
I wish the many American members that regularly bash Tesla here would support their own American made auto makers as well. You can get your car fixed, solve problems, address issues, and still support them. It doesn’t have to be one or the other.

 

[EVNow]

As such, a final release of City Streets will continue to be an SAE Level 2, advanced driver-assistance feature

I'm surprised this would be a surprise to anyone.

I mean highway NOA is Level 2 after several releases. Why would anyone think first release of City NOA would be anything but L2 ?!

Apart from that - Tesla is telling DMV why they don't have to report anything more - so even if the "next" release is L3, they will continue to highlight that the upcoming release is L2.

I don't expect them to get out of L2 for a few years - only when it is so good that everyone already uses the feature like L3.

 

[Dan D.]

Yes I get that. But what about after "final release"? Will Tesla continue to add more to FSD Beta after the "final release" to make it more than L2 later? Based on how Tesla does things, I would say yes, Tesla will continue to improve FSD Beta after "final release". But I think it is a legit question to ask.

This ^

Of course Tesla will release FSD as Level 2. No question.

My concern is that "final release" having satisfied their obligations, they no longer have to upgrade any existing cars to the hardware needed for Level 3+. Sure they will continue to work on their autonomous systems but it will be in new hardware cars and new monthly subscription payments.

 

[shrineofchance]

Yes I get that. But what about after "final release"? Will Tesla continue to add more to FSD Beta after the "final release" to make it more than L2 later?

Obviously, yes.

But I think it is a legit question to ask.

No, it isn’t, because Tesla has very clearly and explicitly stated their plans and intentions many, many times in many different places (DMV emails, earnings calls, Autonomy Day, Twitter, the Tesla website, interviews and press conference calls, etc.).

This is truly much ado about nothing.

 

[Tam]

...Sure they will continue to work on their autonomous systems but it will be in new hardware cars and new monthly subscription payments.

These visions, goals, and hopes for Robotaxis are not proven facts (because there's no Tesla Robotaxis right now picking up rides to make money for us) but the projected guess was based on what Tesla had in 2016.

It's tempting to think that the hardware in 2016 would be the same in 2021, and what we have now should be the same in the future also.

Most likely not. It's the nature of technology. As time goes by, Tesla would get access to more technology, not just in software but in hardware too.

Thus, these visions, goals, and hopes for Robotaxis will no longer be based on what Tesla had in 2016 but what it has now in 2021 and also in the future.

So when the hardware changes from the original 2016, it's tempting to think that Tesla wouldn't want to squeeze some more money out of those 2016 era owners but it found a loophole to charge them anywhere from $1,500 to $2,500 for MCU1 replacement.

 

[Dan D.]

No, it isn’t, because Tesla has very clearly and explicitly stated their plans and intentions many, many times in many different places (DMV emails, earnings calls, Autonomy Day, Twitter, the Tesla website, interviews and press conference calls, etc.).

This is truly much ado about nothing.

Maybe. But there is a strong sentiment that they can't do Level 3+ with the sensor suite they have now. No side-looking front camera and zero redundancy for a start. If they ever get going on Level 3+ deployment testing they will never pass regulation.

At best they will have to keep these HW2.5 & HW3 cars at Level 2, and if there continue to be problems with detecting crossing traffic I would bet they will be made to remove those features from FSD. Sure, Tesla will just blame "the regulators", naturally.

That is why the reality IMO is very different from their plans and intentions.

 

[AlanSubie4Life]

Again, lets look at what Tesla tells us about FSD - Autopilot


That wording has been there since the update post Autonomy Day in April 2019...

Again, you've posted this stuff before, so I'll post it again here, the comparison... (please correct me if I have mislabeled anything - I just lifted it directly from a prior post of yours and am taking your description as accurate).

Nothing below is novel at all - it's all been discussed here many times before, ad nauseum I think. So those without interest can stop reading now:

Did not want to discuss in the inappropriate thread, but this thread seems much more appropriate. The way I see it, the 2021 description has that key phrase difference, that many early buyers of FSD (I am not one of them; I bought after the verbiage change, and I am fully expecting level 2 indefinitely, which I am fine with, since that fully meets my expectations for FSD) use as their potential ammunition against Tesla, should they not implement a Level 3 or better system at some point in time (though the way I read it, there is no clear timeline in the 2016 promise, so not sure how that would play out in court - I think it would come down to what was a "reasonable" delivery timeframe expectation of these buyers, based on contemporaneous public statements by Tesla).

The key phrase difference in 2021 is:

"The future use of these features without supervision is dependent on achieving reliability far in excess of human drivers..."

To me, this clearly means that if Tesla cannot satisfy their various caveats, Tesla is under no obligation to deliver a level 3 system or better to owners who purchased with that verbiage in place. They are saying that if they can't achieve the reliability necessary, for any reason, but have delivered the described 2021 features at some level they deem "good enough" with supervision, they have still delivered on FSD. I just can't see another way to read this. As soon as they have sufficiently validated features, possibly less safe than human drivers, that are also ok with the regulators (the regulators may insist that they are supervised, and Tesla may want them to be too), they can be released, and the promise has been kept.

I hope you don't just slap a disagree on this and call it good. I really genuinely want to know how this could be read otherwise.

The way I see it, the people who purchased under the prior description or similar (the key wording changed in 2019 I think?), have a much stronger claim to being promised a level 3+ system. There simply is no indication that the approved and validated features can require supervision. In the event that Tesla thinks that supervision is needed, or regulators insist upon it, the promise will still not be kept (since that wouldn't satisfy the description in 2016).

So my expectation is:

1) We may get Level 2 that's pretty decent, but not safe enough to be unsupervised. They can release that and satisfy their obligations to most buyers, but not the early FSD buyers. I could see this happening in the next year or so. How that plays out safety-wise & with the regulatory bodies, I'm not willing to predict, but obviously people have concerns.

2) Tesla will continue development, using newer sensors, etc. (for sure there will be newer and better sensors because that's just what happens, and they may well enable higher capabilities). They're obviously going to continue to try to get to Level 3+; it's essential to be competitive! They will have to settle any claims by early buyers, in the event they cannot release level 3+ with the existing hardware (or retrofit if it's possible) in a reasonable timeframe (which I think will be within the next 2 years for most early FSD buyers).

3) I think eventually Tesla will provide Level 3+ functionality, but I have no idea whether it will prove to be possible with existing hardware & sensor locations. I would guess not, if I had to guess, but I don't have a strong position. I think they'll probably need better resolution (and possibly different wavelength) sensors, and possibly additional locations for sensors, to make it sufficiently safe. They may need more powerful hardware as well (HW4+).
Just my opinion & interpretation of the statements & promises from Tesla.

 

[Tam]

...I think they'll probably need better resolution (and possibly different wavelength) sensors, and possibly additional locations for sensors, to make it sufficiently safe. They may need more powerful hardware as well (HW4+)...

For hardware:

"There’s no LiDAR or RADAR being used here. This is just using passive optical which use essentially what a person uses. The whole road system is meant to be navigated with passive optical or camera, so once you solve cameras, or vision, then autonomy is solved. If you don’t solve vision, it’s not solved. That’s why our focus is so heavily on having a vision neural net that’s very effective for road conditions. You can absolutely be super-human with just cameras. You could probably do ten times better than humans with just cameras" (Elon Musk, TED 2017. Vancouver).

The challenge is: Human has a brain to process vision so when will Tesla's software and hardware be good enough to avoid collisions (from scrapes at the curbs or parking lots to high-speed fatalities).

 

[AlanSubie4Life]

For hardware:

"There’s no LiDAR or RADAR being used here. This is just using passive optical which use essentially what a person uses. The whole road system is meant to be navigated with passive optical or camera, so once you solve cameras, or vision, then autonomy is solved. If you don’t solve vision, it’s not solved. That’s why our focus is so heavily on having a vision neural net that’s very effective for road conditions. You can absolutely be super-human with just cameras. You could probably do ten times better than humans with just cameras" (Elon Musk, TED 2017. Vancouver).

The challenge is: Human has a brain to process vision so when will Tesla's software and hardware be good enough to avoid collisions (from scrapes at the curbs or parking lots to high-speed fatalities).

Yes! You’ll note that I didn’t suggest or even imply that they would need radar or lidar. I don’t think they need to at all, though it is possible that at some point additional sensor modes could enhance safety even further. Humans definitely have problems with vision sometimes, so it might well be nice to have superhuman capabilities down the road! Depends on how computer vision advances probably, and the capabilities of the cameras.

But that is not what I was discussing above. It sounds like you were focusing in the hardware part, but just wanted to be clear that I was not suggesting those other sensors were required.

 

[Dan D.]

A key indicator of their progress will be when they achieve their full capabilities in seamless vision, where the blind spots are, and how they can overcome any limitations. There may be some interesting tricks they can do to mitigate such problems, perhaps some inexpensive camera modifications. No prediction or guesstimate software can overcome a true blind spot or nearsightedness though.

It's difficult at the moment to decide if disengagements are the fault of vision/sensors, software, or GPU processing. I wonder what happens when the scene complexity exceeds the processing ability? Does it drop frames, drop objects, skip decision steps?

There may be a limit to what is needed for Level 2. If the driver is "attentive" anyway, why bother solving fully autonomous behaviour for now, maybe just have a great driver assist package and forget about fully dealing with the driving environment?

 

[Tam]

...There may be a limit to what is needed for Level 2. If the driver is "attentive" anyway, why bother solving fully autonomous behaviour for now, maybe just have a great driver assist package and forget about fully dealing with the driving environment?

The answer is obvious: Because $10,000 FSD owners might get $30,000 paid a year by enrolling (and not driving) in Robotaxi program:

Can Elon Musk's robotaxi plan help Tesla owners make $30,000 a year?

Tesla CEO Elon Musk said he is “very confident” Tesla would be able to offer robotaxis next year as a rival to ride-hailing apps like Uber and Lyft.

www.usatoday.com

...blind spots...true blind spot or nearsightedness though...

That might just be too advanced before asking for basic stuff like just to avoid colliding into a stationary obstacle:

2016: Semi-trailer truck in Florida. It was huge and right in front.

2017: Cement median divider in Mountain View, CA: It's easily seen, not a blind spot.

2019: Semi-trailer truck in Florida: Just like 2016 in broad daylight, good weather, good visibility and it's a big truck right in front!

 

[diplomat33]

Obviously, yes.


No, it isn’t, because Tesla has very clearly and explicitly stated their plans and intentions many, many times in many different places (DMV emails, earnings calls, Autonomy Day, Twitter, the Tesla website, interviews and press conference calls, etc.).

This is truly much ado about nothing.

Oh I am not surprised by the DMV emails. We all knew FSD Beta would be released as L2. The question is what will happen after release? I don't believe Elon's timeline of L5 happening later this year. I think it is very likely that FSD Beta will stay L2 for awhile.

 

[boonedocks]

I'm surprised this would be a surprise to anyone.

I mean highway NOA is Level 2 after several releases. Why would anyone think first release of City NOA would be anything but L2 ?!

Apart from that - Tesla is telling DMV why they don't have to report anything more - so even if the "next" release is L3, they will continue to highlight that the upcoming release is L2.

I don't expect them to get out of L2 for a few years - only when it is so good that everyone already uses the feature like L3.

... and this is why I did NOT order FSD for a 4th time on my Model S refresh on order. Pre-paid on a '17 S & X and a '19 S and won't do that again.... It is really pretty sad that Tesla doesn't transfer it to a new order for current customers when the features paid for are FAR from being released...

 

[mikes_fsd]

1) We may get Level 2 that's pretty decent, but not safe enough to be unsupervised. They can release that and satisfy their obligations to most buyers, but not the early FSD buyers.

The only problem with your interpretation is it ignores the middle paragraph altogether -- that middle paragraph has stayed the same from 2016 to today... word-for-word

All you will need to do is get in and tell your car where to go. If you don’t say anything, the car will look at your calendar and take you there as the assumed destination or just home if nothing is on the calendar. Your Tesla will figure out the optimal route, navigate urban streets (even without lane markings), manage complex intersections with traffic lights, stop signs and roundabouts, and handle densely packed freeways with cars moving at high speed. When you arrive at your destination, simply step out at the entrance and your car will enter park seek mode, automatically search for a spot and park itself. A tap on your phone summons it back to you.

This paragraph describes how you the owner/user of the car will use FSD, and the operation in the 2016 & 2021 version is identical - so the goal remains identical.

Again, you've posted this stuff before

And I will keep posting it, because that is the official definition of FSD Capability from 2016 to now... so having the differences noted is also important.

The first paragraph has one common sentence about hardware capable of "full self-driving in almost all circumstances".

There is one safety measuring difference that is introduced in the newer text.
2016 has "a probability of safety at least twice as good as the average human driver"
2019-today has "use of these features without supervision is dependent on achieving reliability far in excess of human drivers as demonstrated by billions of miles of experience"

The last paragraph in both instances explains it will be a piecemeal release (features at a time not big bang) and that there is no concrete timeline tied to the release (because of software validation and/or regulatory approval).

Back to that ignored middle paragraph...
I do not see how you can read both 2016 definition and the 2019 revision and say that Tesla does not have to deliver a system that is capable of doing what that middle paragraph describes.

Safety that is at least 2x as safe as human is still encompassed by 'reliability far in excess of human drivers'.
I see a commitment to safety in the newer description not a way to get out of delivering a product.

You seem to see it as a way for Tesla to get out of delivering a product that they describe, or at least that is the vibe I am getting from the following.

Tesla is under no obligation to deliver a level 3 system or better to owners who purchased with that verbiage in place.

Tesla never committed to deliver L3, there is a clear distinction that it is either L2 supervision or L4+ ..... no supervision in almost all circumstances.

They are saying that if they can't achieve the reliability necessary, for any reason, but have delivered the described 2021 features at some level they deem "good enough" with supervision, they have still delivered on FSD.

They are saying no such thing. You are saying that based on your interpretation of the texts -- but as I pointed out -- you're ignoring that middle paragraph (the largest and word-for-word identical paragraph of the entire FSD Capability definition).

 

[AlanSubie4Life]

The only problem with your interpretation is it ignores the middle paragraph altogether

No, I am not ignoring it. The features described there are the same in both cases.

This paragraph describes how you the owner/user of the car will use FSD, and the operation in the 2016 & 2021 version is identical - so the goal remains identical.

Yes, features are identical, but the following paragraph and caveats are laid on top of this set of features, in both cases. The paragraphs are not separable. In one case the caveat clearly says use without supervision is dependent on safety being achieved - the other does not include that as a “gate” - it just is assumed that twice a good human safety level will be achieved.

I do not see how you can read both 2016 definition and the 2019 revision and say that Tesla does not have to deliver a system that is capable of doing what that middle paragraph describes.

They do need to deliver all those features, subject to the caveats in the final paragraph. You can get all of those features in the middle paragraph, capable of doing what is described, while requiring supervision of the vehicle. It’s not inconsistent, since the text is not separable.

You seem to see it as a way for Tesla to get out of delivering a product that they describe

No, I don’t. I see them describing over three paragraphs a product that will require supervision until it is safe enough, even if every feature described is present and works very consistently without intervention.

Tesla never committed to deliver L3, there is a clear distinction that it is either L2 supervision or L4+ ..... no supervision in almost all circumstances.

Yeah I can agree with that. I’m not much of a believer in L3 in general, but was trying to make the distinction in the huge leap needed to go from L2 to L3 and beyond. I think L3 is pretty dangerous but it still requires substantial capability beyond a driver assist system. Once the safety level is achieved, it seems certain to be L4+ - in most circumstances. I do think the “in most circumstances” clause opens up the possibility of something degrading to L3 (and shortly thereafter to L2) in some extreme conditions. But there is no reason to debate these details, as it’s not really important to the core discussion about supervision requirement. We both know all the definitions by heart.