Q3 Earnings Call: Questions about FSD

[diplomat33]

It seems that investors will be asking Elon for an update on FSD at the Q3 Earnings Call coming up on Nov 4. Here are the relevant FSD questions that will be asked:

- Can you provide more detail on the DeepScale acquisition, its importance, and whether Tesla is still on track to recognize and respond to traffic lights and stop signs with automatic driving on city streets at the end of 2019?

- There is skepticism regarding your comment that Full Self Driving will be ‘feature complete’ by year end, likely resulting from confusion about what ‘feature complete’ means. Could you please talk to this, perhaps give us a list of the features that establish the FSD baseline?

- Could you update us on the progress being made towards FSD, the trajectory of revenue recognition associated with this and the level of performance at which you see full revenue recognition for FSD? Additionally, how do you see pricing developing for this product?

- Can you provide an update on FSD package attach rates. As FSD attach rates improve, will you let the financial benefits manifest in higher gross margins for the company/shareholders; or will you lower vehicle pricing to drive deliveries volume.
Top 5 things Tesla (TSLA) investors want to know from the Q3 2019 earnings call

Let's save this thread and use it to discuss whatever FSD updates/information we get in the Earnings Call.

 

[thewishmaster]

The date might be off a bit, here's the Q3 call:

 

[DBeckwith]

If I understand Elon correctly, he is saying FSD will be feature complete at level 3. I can't see how this will work as robotaxi.

 

[thewishmaster]

If I understand Elon correctly, he is saying FSD will be feature complete at level 3. I can't see how this will work as robotaxi.

I only skimmed the responses but it sounded to me that he said it might be hands-off and robotaxi-ready end of next year rather than at the first "feature complete" checkpoint.

 

[diplomat33]

If I understand Elon correctly, he is saying FSD will be feature complete at level 3. I can't see how this will work as robotaxi.

Because "feature complete" is just Elon's "Level 1". It is just the first step. It is not the level when Tesla will do robotaxis. Robotaxis will happen later when they reach Elon's "level 3" when regulators rate the car as reliable autonomous driving.

 

[Daniel in SD]

• Basic functionality (feature complete?) aspirationaly by the end of this year.
• Level 4-5 by the end of 2020. "Reliable enough that you don't need to pay attention in our opinion by the end of next year."

He's not backing down! haha.

 

[diplomat33]

If I understand Elon correctly, he is saying FSD will be feature complete at level 3. I can't see how this will work as robotaxi.

I only skimmed the responses but it sounded to me that he said it might be hands-off and robotaxi-ready end of next year rather than at the first "feature complete" checkpoint.

@electronblue looks like we got Elon's definition of "feature complete L5 nogeofence".

Here is a summary of Elon's answer in the earnings call:

"feature complete" = the car can drive from your house to your workplace "most likely" without any driver intervention. Will still require driver supervision. "Feature complete" will fill in the gaps between "low speed autonomy" (ie smart summon), "high speed autonomy" (ie highway) and "intermediate speed autonomy" (ie city driving, traffic lights etc). "Feature complete" will most likely be able to do those 3 things without intervention but still require supervision.

Timeline for FSD:

3 "levels:
"level 1" = feature complete. Car is autonomous but requires supervision and may require driver intervention at times. Will work "most of the time", not every situation or every edge case.
"level 2" = Tesla deems FSD safe enough to remove driver supervision.
"level 3" = regulators agree that FSD is safe enough to remove driver supervision.

---------------------------

I am assuming that robotaxis will happen at Elon's "level 3".

 

[diplomat33]

• Basic functionality (feature complete?) aspirationaly by the end of this year.
• Level 4-5 by the end of 2020. "Reliable enough that you don't need to pay attention in our opinion by the end of next year."

He's not backing down! haha.

Personally, I don't think Tesla will meet these deadlines. They got most of what Elon calls "high speed autonomy" (NOA Highway), basic functionality for "low speed autonomy" (smart summon) and nothing released yet for "intermediate autonomy" (automatic city driving). The only way they could declare "feature complete" done by the end of this year is if the dev team secretly actually have traffic lights and automatic city driving done and they suddenly release it in December. I doubt it.

Having said that, I do think that once Tesla achieves what Elon is calling "feature complete", the vast amount of fleet data should allow Tesla to improve it pretty quickly. So I do suspect that we could see a lot of progress after "feature complete" is released. But no, I don't think we will have L4/5 autonomy by end of 2020. Just my opinion.

 

[DBeckwith]

So if you use the classic level definition, "feature complete" is about level 2.5, but by end of 2020, it will be level 5 (Elon's level 3) awaiting regulatory approval.

 

[Daniel in SD]

@electronblue looks like we got Elon's definition of "feature complete L5 nogeofence".

I would agree with Musk's definition.

I am assuming that robotaxis will happen at Elon's "level 3".

I wish he hadn't said "level 3". That's going to be super confusing for people since there's an actual definition for Level 3 autonomous vehicles. He was saying that the 3rd step is to get regulatory approval. If FSD is to be used in robotaxis it will be SAE Level 4 or SAE Level 5 since SAE level 3 requires a driver to be in the car to take over when the system requests.

I think in Arizona no regulatory approval is required to operate vehicles without a driver so in some jurisdictions Tesla should be able to skip step 3. We should all thank Arizonians for being guinea pigs

 

[diplomat33]

I would agree with Musk's definition.


I wish he hadn't said "level 3". That's going to be super confusing for people since there's an actual definition for Level 3 autonomous vehicles. He was saying that the 3rd step is to get regulatory approval. If FSD is to be used in robotaxis it will be SAE Level 4 or SAE Level 5 since SAE level 3 requires a driver to be in the car to take over when the system requests.

I think in Arizona no regulatory approval is required to operate vehicles without a driver so in some jurisdictions Tesla should be able to skip step 3. We should all thank Arizonians for being guinea pigs

Agree about the confusion part. By the way maybe we should call it EL1/2/3 for "Elon Level 1,2,3" to distinguish from the SAE levels?

I would amend my previous answer. If States don't require regulatory approval, Tesla could probably do robotaxis when they reach "Elon Level 2" instead of "Elon Level 3".

 

[drtimhill]

Personally, I don't think Tesla will meet these deadlines. They got most of what Elon calls "high speed autonomy" (NOA Highway), basic functionality for "low speed autonomy" (smart summon) and nothing released yet for "intermediate autonomy" (automatic city driving). The only way they could declare "feature complete" done by the end of this year is if the dev team secretly actually have traffic lights and automatic city driving done and they suddenly release it in December. I doubt it.

Having said that, I do think that once Tesla achieves what Elon is calling "feature complete", the vast amount of fleet data should allow Tesla to improve it pretty quickly. So I do suspect that we could see a lot of progress after "feature complete" is released. But no, I don't think we will have L4/5 autonomy by end of 2020. Just my opinion.

Agreed. I think FSD is going to come incrementally, not so much in terms of specific "tricks" (red lights, stop signs), but more in terms of starting with simple "follow the road rules" basic street-level driving that then gradually gets better and better at handling more exceptional circumstances. Much like a human learns to drive in fact

Remember that AI is NOT "artificial intelligence", it's really "artificial imitation" .. it can mimic things it has been taught, but not use deduction or induction to apply that to new situations.

 

[DBeckwith]

Without a regulatory stamp of approval, Tesla would be exposed to a lot of liability. Though that hasn't stopped Tesla yet.

 

[Daniel in SD]

Without a regulatory stamp of approval, Tesla would be exposed to a lot of liability. Though that hasn't stopped Tesla yet.

I suspect that most self driving regulations will expose Tesla to a lot of legal liability. I think it makes sense for the manufacturer of an autonomous vehicle to be liable for flaws in the software just like they are for other vehicle defects.
It would have been interesting to hear Elon's opinion on this.

 

[DBeckwith]

Elon has previously speculated that Tesla would be liable if FSD was at fault.

 

[Bladerskb]

• Basic functionality (feature complete?) aspirationaly by the end of this year.
• Level 4-5 by the end of 2020. "Reliable enough that you don't need to pay attention in our opinion by the end of next year."

He's not backing down! haha.

He is..in his own deceptive BS way.

This is typical push back by elon that he has been doing since 2015.
Months ago he said ppl will be able to look out the window by Q2 2020, now he's saying by end of 2020.
Ask him 6 months from now and he will say by Q1 2021. Rinse and Repeat.

 

[heltok]

So what Elon was saying is that he believes(guides) that FSD will be able to drive from home to work most(>50%) of the time this year(before 31dec) with only driver monitoring [for EAP users].

As Elon is the head of Autopilot and has not been changing the guidance over the last quarter I would say that it is pretty likely that they will do it in Q4 or at least early Q1. Which even without robotaxi would be a crazy cash cow, lot’s of upgrades of pure margins and insane PR. Interesting times!

 

[paaltio]

The only way they could declare "feature complete" done by the end of this year is if the dev team secretly actually have traffic lights and automatic city driving done and they suddenly release it in December. I doubt it.

If they can somehow encourage the driver to verify the traffic lights on the visualization each time, I could definitely see it on early access at least soon.

They’re not going to get to 100% anyway since they’re relying on maps right now, so the feedback to the driver is pretty essential. And the driver needs to pay attention. If they can solve that human problem over any other right now, they’ll move forward this year as promised I think.

 

[EinSV]

On the call Elon suggested that what they were hoping to release to early access by the end of the year was the full suite of features needed to drive on city streets. Timeline aside, I had thought they would introduce things one step at a time -- for example starting with stop signs and stop lights before moving on to turns.

Since the demonstration we saw back in April had a full set of basic features, I would have been very surprised if the dev mode they are working on was anything other than "the whole enchilada" in terms of features. But I was not expecting they would release them all at once. Should be interesting to see how it all plays out.

 

[electronblue]

@electronblue looks like we got Elon's definition of "feature complete L5 nogeofence".

Here is a summary of Elon's answer in the earnings call:

"feature complete" = the car can drive from your house to your workplace "most likely" without any driver intervention. Will still require driver supervision. "Feature complete" will fill in the gaps between "low speed autonomy" (ie smart summon), "high speed autonomy" (ie highway) and "intermediate speed autonomy" (ie city driving, traffic lights etc). "Feature complete" will most likely be able to do those 3 things without intervention but still require supervision.

This does not answer the question at all, really, but it does provide some credibility to the recent poster that claimed Tesla is measuring feature complete by — what was it — 20% of successful commutes from some testing location to another?

But the things is: These answers do not provide any insight into the system’s abilities beyond reliability. I get it that it needs supervision because it is not reliable, but does that mean all the abilities required for Level 5 no geofence feature complete (the English meaning of those words in the usual professional contexts) are there... just not very reliable?

For example, does it mean Tesla has the ability to park seek in a parking lot with humans directing traffic? Does it mean it can read all traffic signs and follow all traffic legislation? Does it mean it can detect all obstacle types and react to them (just not necessarily very reliably yet)? Does it mean the car has the ability to reach a minimal risk condition everywhere (just not very reliably) ie reach parking etc safely from any place on the road?

Or does it just mean some minimum viable urban Level 2 product that can complete some pre-selected commutes, following some basic lane types and a few traffic signs or signals, but not actually react to anything out of the ordinary or beyond the very, very basic of tasks? This would not be anything like Level 5 no geofence feature complete in their common professional meanings.

Level 5 no geofence feature complete is a massive undertaking — and we do not yet know if Tesla told the truth when they claimed it as the goal for this year. These answers unfortunately shed no light on that, at least no light that would confirm they are aiming for actual SAE Level 5 no geofence feature complete.

Timeline for FSD:

3 "levels:
"level 1" = feature complete. Car is autonomous but requires supervision and may require driver intervention at times. Will work "most of the time", not every situation or every edge case.
"level 2" = Tesla deems FSD safe enough to remove driver supervision.
"level 3" = regulators agree that FSD is safe enough to remove driver supervision.

---------------------------

I am assuming that robotaxis will happen at Elon's "level 3".

A friendly suggestion: Let us not invent new meanings for levels in an autonomous vehicles forum. SAE has the definitions for those. If you must, call yours steps or something else. This just obfuscates things.

In any case, I think the best I can add is that we don’t know what Tesla is doing — and until we see it, all this is pretty much moot.