PSA: Please don't expect any FSD features this year, or even early next year, really

[OPRCE]

So you're saying that if Waymo allowed their test drivers to hold the steering wheel they would be exempt from all licensing and reporting requirements for self driving testing in CA?

1. No, because by operating them thus far under the program they have declared them as >= L3, so they remain L3 even if the safety drivers behave differently. It would just mean that the number of disengagements reported would spike, which is the opposite of what they are trying to achieve, hence an entirely improbable scenario.

My point is that to test a L3 system in California the driver is required to continuously monitor the system. I'm not sure where you're getting that a L2 system requires driver input. L2 only requires driver input to maintain safety which is exactly what is required when testing a L3-5 system in CA.

2. The safety driver under the AV >=L3 regulations =/= driver of L2, because the systems are designed and declared to assign them different functions, review SAE graphic above. The fact that they both do "some sort of monitoring" is a declaration too vague to assist understanding.

3. I note you do not address the point (as regards Tesla) about this whole issue being resolved since yesterday.

 

[Daniel in SD]

1. No, because by operating them thus far under the program they have declared them as >= L3, so they remain L3 even if the safety drivers behave differently. It would just mean that the number of disengagements reported would spike, which is the opposite of what they are trying to achieve, hence an entirely improbable scenario.

It's easy to hold the wheel and not cause disengagements. I'm not sure hat you're saying.

2. The safety driver under the AV >=L3 regulations =/= driver of L2, because the systems are designed and declared to assign them different functions, review SAE graphic above. The fact that they both do "some sort of monitoring" is a declaration too vague to assist understanding.

Look at how Tesla describes FSD. They are describing a L5 system that still requires monitoring because it is not yet reliable. I'm saying that a safety driver testing a L3-5 system is exactly the same as a driver of Tesla's FSD system as they have described it. The argument that they're not testing a FSD system just because their system doesn't work very well yet doesn't make any sense to me and I doubt it would make much sense to the DMV.

5. Time will tell!
Surely the question is now settled, since yesterday's rejigging of AP/FSD has made it abundantly clear that the latter will remain a L2 dev program for the foreseeable future?

Yes, I agree that Tesla is nowhere near releasing a functional L3-5 system. However I believe the system they are planning on releasing this year would qualify as a L3-5 system under California rules (and other states too) and would be subject to testing requirements. At what point do you think Tesla would be subject to FSD testing requirements? Only when the system reaches a certain level of functionality?

 

[OPRCE]

It's easy to hold the wheel and not cause disengagements. I'm not sure hat you're saying.

1. I'm saying that however Waymo safety drivers choose to hold (or not) the wheel does not change the vehicle from being L3, thus they would not relieved from the requirement to report disengagements, however oft they occur.

Look at how Tesla describes FSD. They are describing a L5 system that still requires monitoring because it is not yet reliable. I'm saying that a safety driver testing a L3-5 system is exactly the same as a driver of Tesla's FSD system as they have described it. The argument that they're not testing a FSD system just because their system doesn't work very well yet doesn't make any sense to me and I doubt it would make much sense to the DMV.

2. I think here we have belaboured our respective points sufficiently.

Yes, I agree that Tesla is nowhere near releasing a functional L3-5 system. However I believe the system they are planning on releasing this year would qualify as a L3-5 system under California rules (and other states too) and would be subject to testing requirements. At what point do you think Tesla would be subject to FSD testing requirements? Only when the system reaches a certain level of functionality?

3. I see no possibility of Tesla releasing anything they would want to qualify (and allow to be officially tested) as L3 this year. Musk stated "feature complete" by end 2019, then +1 year until "sleep to destination", which I am taking to mean he imagines approval by regulators as L3 a year after all FSD features are available for real-world validation. However, it is never as easy in reality, so I would estimate L3 approval not before Dec 2022, then L4, if at all feasible with current sensor suite, by start of 2025.

 

[Daniel in SD]

1. I'm saying that however Waymo safety drivers choose to hold (or not) the wheel does not change the vehicle from being L3, thus they would not relieved from the requirement to report disengagements, however oft they occur.

Yes, and Tesla FSD is a level 3-5 system in exactly the same way Waymo's system is. It is not yet validated, is still under development and requires a driver to monitor it for safety. Look at all of Tesla's descriptions of the system. Elon Musk repeatedly describes FSD as a level 5 system. By your logic Tesla will only have to conform to California's testing requirements after the system is already complete and can pass the validation tests necessary to sell it to the public. It's strange that all these other companies are bothering to adhere to the rules if that is the case. It is absurd to think that is what the DMV intended when they designed the testing rules.
Now Tesla claims that they don't need to test on public roads to validate their system yet. That's fine, maybe they don't. My point is I don't see how they can release what they're talking about releasing with violating the self driving system testing rules. They probably don't either, that's why they're talking about approval from regulators all the time.

 

[commasign]

From the conference call yesterday...

Dropbox - Tesla Call Transcript (2.28.19).pdf

 

[Knightshade]

... he can't possibly be basing the idea they've got full self driving in all conditions in all areas cracked because it works well in his personal car in one part of CA.... right?

 

[commasign]

... he can't possibly be basing the idea they've got full self driving in all conditions in all areas cracked because it works well in his personal car in one part of CA.... right?

He probably is. Same thing happened with AP1 autosteer and NoA. At some point, he's happy with how it works in his development car on his commute, and then they push it out to other internal testers, then early access program, etc., Then wide release followed by lots of forum and Facebook posts and youtube videos showing how it doesn't work in other conditions, then over time (months or years) it gets better and better.

 

[Knightshade]

He probably is. Same thing happened with AP1 autosteer and NoA. At some point, he's happy with how it works in his development car on his commute, and then they push it out to other internal testers, then early access program, etc., Then wide release followed by lots of forum and Facebook posts and youtube videos showing how it doesn't work in other conditions, then over time (months or years) it gets better and better.

....all I can think of is this....


Saturday Morning Breakfast Cereal - 2012-03-25

 

[Daniel in SD]

From the conference call yesterday...

View attachment 382108

Dropbox - Tesla Call Transcript (2.28.19).pdf

I am really curious how this doesn't count as autonomous vehicle testing in California.
From the autonomous vehicle testing regulations:

(b) “Autonomous test vehicle” is a vehicle that has been equipped with technology that is a combination of both hardware and software that, when engaged, performs the dynamic driving task, but requires a human test driver or a remote operator to continuously supervise the vehicle’s performance of the dynamic driving task.

(1) An autonomous test vehicle does not include vehicles equipped with one or more systems that provide driver assistance and/or enhance safety benefits but are not capable of, singularly or in combination, performing the dynamic driving task on a sustained basis without the constant control or active monitoring of a natural person.

(2) For the purposes of this article, an “autonomous test vehicle” is equipped with technology that makes it capable of operation that meets the definition of Levels 3, 4, or 5 of the SAE International’s Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles, standard J3016 (SEP2016), which is hereby incorporated by reference.

(3) The presence of a natural person who is an employee, contractor, or designee of the manufacturer in the vehicle to monitor a vehicle’s autonomous performance shall not affect whether a vehicle meets the definition of autonomous test vehicle.

I guess the argument Tesla would make is that their system is not able to perform the dynamic driving task on a sustained basis?

 

[OPRCE]

Yes, and Tesla FSD is a level 3-5 system in exactly the same way Waymo's system is. It is not yet validated, is still under development and requires a driver to monitor it for safety. Look at all of Tesla's descriptions of the system. Elon Musk repeatedly describes FSD as a level 5 system.

1. LOL, just no, there is literally no evidence that Tesla FSD exists at all, apart from the various claims and confused mutterings of Mr Musk, who has vacillated widely between describing it (long ago in late 2016 on the Tesla website, when the staged demo video was released) as L5 capable of forming an autonomous Tesla Network robotaxi fleet and (in the present) a mere L2 NoAP where the driver is persistently nagged every 15 seconds and has to confirm lane-change suggestions. To state FSD is L3 like Waymo's system is ludicrously unsupported, indeed counterfactual to everything we know about Waymo and contradictory of everything Musk is currently saying, but suit yourself if that is what you need to believe for some reason. You are, however, fooling only yourself at this stage.

By your logic Tesla will only have to conform to California's testing requirements after the system is already complete and can pass the validation tests necessary to sell it to the public.

2. As of yesterday it also seems to describe Musk's logic, see the conference-call transcript in answer to Ben Sullins, where Elon painfully chews up his own tongue trying to describe a staged process (seemingly invented ad-hoc as he goes along) to get from current L2 to a L3 approval:

3. Again, make of this double-dutch what you will, but I have already made my position clear and IMHO this new evidence only confirms it.

Now Tesla claims that they don't need to test on public roads to validate their system yet.

4. Please link to where Tesla has ever claimed this?

My point is I don't see how they can release what they're talking about releasing with violating the self driving system testing rules.

5. Your cognitive dissonance arises from the unreasonable insistence that Tesla already has a L3 FSD to test or release. Just get it through your head, as Musk has now crabwise admitted, that this is not the case, and it all becomes very simple to understand.

 

[scottf200]

Currently EAP can't handle traffic lights or stop signs or zipper merges. The first two are the easy ones.

Are you serious about the traffic lights and stop sign comment?!?! Have you seen the various Mobileye videos on this or other discussions like below? There are many hard examples. This is not a simple + sign looking intersection problem. 100s of variations, angles, 3+ intersecting streets, trees blocking view of signs, etc.
comma.ai: announces stopsign/red-light 'AP' - needs 10 cm GPS loc

 

[kbecks13]

Hey - we just got a new FSD feature (for people who had it before "the announcement") in the form of an invitation to the early access program! That's worth paying a few months early for me

Upgrading to Autopilot and Full Self-Driving Capability

 

[OPRCE]

Hey - we just got a new FSD feature (for people who had it before "the announcement") in the form of an invitation to the early access program! That's worth paying a few months early for me

Upgrading to Autopilot and Full Self-Driving Capability

Great, it looks like I am now promoted to the privileged caste of EAP testers! YESSSS!
[presuming this also applies in Europe?]

"Customers who previously purchased Full Self-Driving will receive an invitation to Tesla’s Early Access Program (EAP). EAP members are invited to experience and provide feedback on new features and functionality before they are rolled out to other customers."

 

[OPRCE]

By your logic Tesla will only have to conform to California's testing requirements after the system is already complete and can pass the validation tests necessary to sell it to the public.

Another part of your difficulty here is in assuming Tesla needs to get FSD approved as L3 before it can be sold, whereas in reality they had been selling it, and have now resumed doing so, before it even exists in any tangible form available to the customers.

BTW, sorry if I may have come across as a bit short earlier, but this discussion is bordering on sterile repetition which I find somewhat exasperating.

 

[uNki]

Could someone help me to understand why there are several official Tesla Autopilot videos from 3-4 years ago, demonstrating a pretty decent L3 system which seems to work better than the system they have in place right now?

like this one (I can't find the older one with the Model S)

Is the current Autopilot system just limited on purpose and could they introduce a working L3 system for city driving like they say on their website later this year?

 

[AlanSubie4Life]

I am really curious how this doesn't count as autonomous vehicle testing in California.
I guess the argument Tesla would make is that their system is not able to perform the dynamic driving task on a sustained basis?

I think this is exactly the argument. It’s not really an argument so much as the obvious truth of the matter! I’m fairly sure the system they are testing is absolutely incapable of doing this task on a sustained basis, and I figure that won’t change for a few years.

 

[OPRCE]

@Daniel in SD, here is the final clincher that FSD is being rolled out as L2 until eventual approval as L3 without nags:

So, the FSD upgrade just re-appeared in my account and it has a different fine print than the order page which makes more sense (emphasis mine):

"The current features require active driver supervision and do not make the vehicle autonomous. The future 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, as well as regulatory approval, which may take longer in some jurisdictions. As these self-driving features evolve, your car will be continuously upgraded through over-the-air software updates."

I don't think Tesla can spell it out any more clearly than that, at the point of sale, as reviewed by their lawyers rather than someone loaded up on Ambien & Vino.

 

[OPRCE]

Could someone help me to understand why there are several official Tesla Autopilot videos from 3-4 years ago, demonstrating a pretty decent L3 system which seems to work better than the system they have in place right now?

like this one (I can't find the older one with the Model S)

Is the current Autopilot system just limited on purpose and could they introduce a working L3 system for city driving like they say on their website later this year?

The most charitable interpretation is that this video was made to demonstrate to worried customers (after the split with MobilEye in mid-2016) that the new AP2 system had all the hardware sensors needed for eventual FSD, i.e. offering reassurance that the AP development program would continue and not just implode.

Another is that is was basically a lie used on a loop in showrooms (from Nov. 2016 until today) to give potential customers the false impression that L3 FSD was a more or less finished product only awaiting regulatory approval to roll out, and that Tesla salespersons verbally reassured them of this, thus tending to boost sales.

Also, both may be true.

 

[Knightshade]

The most charitable interpretation is that this video was made to demonstrate to worried customers (after the split with MobilEye in mid-2016) that the new AP2 system had all the hardware sensors needed for eventual FSD, i.e. offering reassurance that the AP development program would continue and not just implode.

Another is that is was basically a lie used on a loop in showrooms (from Nov. 2016 until today) to give potential customers the false impression that L3 FSD was a more or less finished product only awaiting regulatory approval to roll out, and that Tesla salespersons verbally reassured them of this, thus tending to boost sales.

Also, both may be true.

I think the obvious answer is it was a very custom FSD software designed to handle that exact and specific location, route, and set of circumstances.

As Musk noted a while ago, they could have done the coast to coast autonomous drive when promised, but would've had to pick a very specific route and program to just that route, rather than have a general-case solution that he wants.


Same reason Waymos only "production" self driving product only works in one specific town in one specific state with a very specific and narrow range of weather.

 

[OPRCE]

Could someone help me to understand why there are several official Tesla Autopilot videos from 3-4 years ago, demonstrating a pretty decent L3 system which seems to work better than the system they have in place right now?

like this one (I can't find the older one with the Model S)

Is the current Autopilot system just limited on purpose and could they introduce a working L3 system for city driving like they say on their website later this year?

I think the obvious answer is it was a very custom FSD software designed to handle that exact and specific location, route, and set of circumstances.

As Musk noted a while ago, they could have done the coast to coast autonomous drive when promised, but would've had to pick a very specific route and program to just that route, rather than have a general-case solution that he wants.


Same reason Waymos only "production" self driving product only works in one specific town in one specific state with a very specific and narrow range of weather.

Would add that they also had several AP project leaders quit (Sterling Anderson left and was sued, Chris Lattner lasted 6 months) during this time and there were several rewrites of the software from scratch according to current kapo Karpathy, so the whole department has been stood on its ear at regular intervals, not exactly conducive to progress.