General Discussion: 2018 Investor Roundtable

[zmarty]

Disagree.

Ok. Here are my reasons for saying it's bizarre that Tesla has tested 0 miles in California:

• Pretty much the entire AP team is in California. I find it simply unbelievable that you can develop self-driving software without testing it seriously on public roads, like your competition does. If they did test something that is not classified as self-driving by California it's pretty much as bad in my eyes. I do not believe for a second they periodically shipped software to a totally separate team for testing out of State, let's be serious.
• As always with Tesla, I find their choice of words very deliberate. The statement prepared by their counsel was obviously meant to reassure the public about the state of their self-driving program. I can guarantee they have foreseen how badly this would look, that is why they wrote the statement this way. Legally they did not have to explain themselves this much to the California DMV.

 

[RiverCard]

ABS securitisation get's pretty complex. I think the majority of the $3.8b is unlikely to be available for securitisation. Some of the reasons for this are, in no particular order:

• Tesla would have warehoused or securitised these assets earlier if they could. The WA coupon on this securitisation is likely to be well under 2% when they just went out and offered junk bonds for 5.3% not that long ago. The rates on the warehouses are also well under the 5.3% of Tesla's bond offering. Why hasn't Tesla gone out and taken a $3b warehouse line? There would be plenty of banks out there ready to offer these facilities. Tesla can't get cheaper money than an efficient warehouse/securitisation.
• Looking at how hard Moody's hit them on this deal (50% RV loss assumption, <70% AAA note rating) it is likely that they haven't been able to get a rating any earlier than this. Fitch and S&P tend to treat new issuers harder than Moody's - this doesn't mean that they won't be able to securitise more in the future, but they will not get anywhere near the execution of a more established player
• Only one RA rated this deal. The standard is two or more, and it would be strange for Tesla not to try for more than one rating as it would provide greater confidence to the market.
• If the $3.8b of leases is global, they would have to build up a large enough pool to securitise in each jurisdiction. Min size is about a quarter billion USD for the economics to work - there are substantial establishment fees.
• Tesla Finance LLC (the sponsor and Tesla's financing arm, established in 2013) only has $830m of leases - why are the rest not included in this company? Are they required to stay within Tesla's main entity to meet covenants on other bond issuances? Are they leased on non-US vehicles? Who knows - but it seems unlikely that they are available to securitise in the US. My guess is that the difference between the $600m sold in this securitisation and the $230 left behind is that these loans need to season for a month or two, or are not currently performing.
  
• One of the big no-nos of structured finance is that the sponsor cannot provide additional credit support to a deal - Does the resale value guarantee provide this and therefore exclude these leases from being securitisable.
• Did Tesla have the correct wording in the T&Cs of their early leases (able to transfer/sell the loans without the lessee's consent)?
• There is extreme geographic concentration in this securitised pool. Are the remaining loans all in CA and geographic concentration would kill the rating if more CA loans are tipped in?

Both Resale value and Residual value guaranteed vehicles are in the $3.8 B on the Balance Sheet because the guarantees require "lease" treatment under GAAP. Neither type of transaction can be securitized by Tesla simply because Tesla is not any of the vehicles' owner (as on the cars' titles). The Resale VG vehicles are owned by the original individual purchasers; the Residual VG vehicles are owned by 3rd party financial institutions.

Moody's Pre-Sale shows that as of 9/30/17 TFL had 9,773 leases outstanding of which 8,879 will be transferred to the Tesla Lease Trust-68% S/ 32% X. The average remaining term on the leases transferred is 20 months (out of 35 months total).

Most of the $546 MM proceeds from the ABS will likely be used to re-pay the Warehouse Line creditors (who will release their security interest in the vehicles in the pool (10% of the pool is retained as over-collateralization.)) Presumably the re-payment will be made available by the Warehouse lenders for M3 leases.

With the 50% haircut on the residual value of vehicles in the pool in order to get the ratings, what is the disposition of the vehicles when the lease terms expire. Does the titling trust send them to auction or does Tesla re-purchase them for CPO inventory?

 

[schonelucht]

Well that disengagements report at least puts to bed the rumors that Tesla was actually trying coast to coast trips on a permanent basis and got as far as doing one with only 30 disengagements. Unless someone wants to claim their coast to coast trip would not start or end in California. But that would be really bizarre.

Honestly I too am in the camp that this is not good. Take a look at the other reports here https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2017 and tell me. Had Tesla reported what Cruise reported would there not have been posts on here, electrek and reddit about how far ahead Tesla is and how it massively improved? And would you then still have said : that is a big no burger?

I still remember there was a time that the main argument went that traditional carmakers won’t stand a chance because they weren’t physically located were all the bright minds are( Bay Area ) But now Tesla would inflict itself that same penalty voluntarily by having the test team anywhere but there?

Before this report my position was that Tesla had a massive lead coming into 2016 but lost some of it during the turmoil in the AP program the last two years, while still retaining that lead. Today the only consistent position is to assume Tesla is behind.

The only parameter on which Tesla is ahead is live data capture. Even there the lead isn’t that massive. Data send off happens only sporadically (1 in 100 miles not even), is very short bursts of activity and only started first half of 2017. Billions of miles driven turns into millions of miles captured. That’s the kind of scale of data capturIng that other manufacturers with dedicated cars with far more sensors can attain in a relatively short time.

 

[mongo]

Here's a 2018 image from Google Maps:

That image may say copyright 2018, but it is not from this year. No automated storage tower, and the rail spurs are still there.

 

[AudubonB]

MODERATOR COMMENT:

After studying the links provided in post #1810 above, it seems prudent to alert the date-unwary that all the references therein are from 2016 and the beginning of 2017. Without poring through the entirety of the reddit post, it appears none of its participants caught the seeming time-warp.

 

[dc_h]

Here's a 2018 image from Google Maps:

That is what I see when I zoom to street level, it is older than the image when you zoom out a bit. If you look to the south, you’ll see half of the model 3 parking lot isn’t thee yet and the grade work to the north has not started, but when you zoom out the expanded north and south lot are complete and the grade work is in full tilt. I think the second half of the model 3 lot was done in December, so the new grading on the north side is in the last few weeks.

 

[Curt Renz]

That image may say copyright 2018, but it is not from this year. No automated storage tower, and the rail spurs are still there.

Thanks for the update, but my point was that the image appeared newer than the one presented by dc_h and may have been helpful to him/her. Perhaps I was wrong and the area in the north has since been reworked.

 

[dc_h]

Thanks for the update, but my point was that the image appeared newer than the one presented by dc_h and may have been helpful to him/her.

No, I see your your image/view when I zoom down and it is definitely older. Look south of the plant and you will see no new parking lot for 3s in your view of the plant.

 

[mongo]

Well that disengagements report at least puts to bed the rumors that Tesla was actually trying coast to coast trips on a permanent basis and got as far as doing one with only 30 disengagements. Unless someone wants to claim their coast to coast trip would not start or end in California. But that would be really bizarre.

Honestly I too am in the camp that this is not good. Take a look at the other reports here https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2017 and tell me. Had Tesla reported what Cruise reported would there not have been posts on here, electrek and reddit about how far ahead Tesla is and how it massively improved? And would you then still have said : that is a big no burger?

I still remember there was a time that the main argument went that traditional carmakers won’t stand a chance because they weren’t physically located were all the bright minds are( Bay Area ) But now Tesla would inflict itself that same penalty voluntarily by having the test team anywhere but there?

Before this report my position was that Tesla had a massive lead coming into 2016 but lost some of it during the turmoil in the AP program the last two years, while still retaining that lead. Today the only consistent position is to assume Tesla is behind.

The only parameter on which Tesla is ahead is live data capture. Even there the lead isn’t that massive. Data send off happens only sporadically (1 in 100 miles not even), is very short bursts of activity and only started first half of 2017. Billions of miles driven turns into millions of miles captured. That’s the kind of scale of data capturIng that other manufacturers with dedicated cars with far more sensors can attain in a relatively short time.

As long as the test drivers keep their hands on the wheel (like any other AP user) it is not considered automonomous mode and does not get reported. The software can automatically log disagreements between the driver and AP for training purposes. This is the safest way to run real world tests due to having the least reaction time between bad AP driving and ability of driver to take over.

Further, if the SW requires monitoring by a natural person, it is not considered an autonomous vehicle. I would put forth that a button one must press at regular intervals (or continuously hold aka "dead man's switch") would fall into this category as does the hands on wheel nag.

Therefore, I do indeed claim that Tesla can run FSD with a nag added and not be required to report anything.

Link to CA legislation

Regarding data uploads: Tesla's internal test fleet can be instrumented to log the complete raw data set for further training.

 

[Skryll]

I don't follow the autopilot issue closely but others here are more competent and up-to-date. Please correct any misunderstandings.

By some measures, including comparison with autopilot 1.0, the current autopilot falls behind or is more betaish than we would like.[...]

I have been driving a lot of AP1 in our Model X90D from 2016 (45xx vin range), and more recently AP2.5 in our model 3 from 2017 (31xx vin range).

While they are different and both not perfect in all situations, I would not say that 2.5 is inferior to 1.0, in fact I much prefer 2.5 because it does not follow the car in front of it out of the center of the lane just because the driver in front of me isnt paying attention.

Also Elon said that we will see some significant updates this year, which will leave AP1 in the dust.

I am not worried.

Cheers, Michael

 

[Curt Renz]

As long as the test drivers keep their hands on the wheel (like any other AP user) it is not considered automonomous mode and does not get reported. The software can automatically log disagreements between the driver and AP for training purposes. This is the safest way to run real world tests due to having the least reaction time between bad AP driving and ability of driver to take over.

Further, if the SW requires monitoring by a natural person, it is not considered an autonomous vehicle. I would put forth that a button one must press at regular intervals (or continuously hold aka "dead man's switch") would fall into this category as does the hands on wheel nag.

Therefore, I do indeed claim that Tesla can run FSD with a nag added and not be required to report anything.

Link to CA legislation

Regarding data uploads: Tesla's internal test fleet can be instrumented to log the complete raw data set for further training.

Thanks for that information. I too suspect that Tesla does not want its test results to be made available by California to competitors at this time.

EDIT: I now see that Ford, BMW, Volkswagen, Honda and some others are taking a similar approach to that of Tesla regarding California's request for autonomous driving information.

 

[schonelucht]

Therefore, I do indeed claim that Tesla can run FSD with a nag added and not be required to report anything.

Link to CA legislation

Regarding data uploads: Tesla's internal test fleet can be instrumented to log the complete raw data set for further training.

Sure. Anything is possible. Don’t disagree there. But in the context of everything else like continued missed deadlines, team leadership turnover and repeated failure of even the smallest promised deliverables, at some point we need to employ Occam’s Razor and go with the simpelst explanation that fits all the facts.

At this point the brightest sliver of hope for the FSD program is @verygreen ’s conclusion that they finally started testing in December based on the appearance of high precision data mapping. That means Tesla is roughly 10 months behind Cruise and a year or two behind Waymo wrt testing.

 

[mongo]

But in the context of everything else like continued missed deadlines, team leadership turnover and repeated failure of even the smallest promised deliverables, at some point we need to employ Occam’s Razor and go with the simpelst explanation that fits all the facts.

Agree, they have had churn in AP land. But something to keep in mind is that whereas functional/ procedural software has incremental improvements that build on and add to previous versions, NN is a huge set of coefficients in an also large combination of processing blocks.
Segmenting our small deliverables doesn't fit with how the NM stuff works (other than totally independent functions like rain sense) as they can't necessarily maintain any specific functionality while working on the whole.

They could release a beta version, then realize the configuration is a dead end and scrap all the tuning performed and start over with the learning process on a new data flow. This would reset the timing. That has likely already happened at least once based on the Tweet regarding front end vision processing.

 

[tinm]

MEANWHILE...

Some New Mexico news. In case you’re not aware (the media certainly isn’t), New Mexico is a State in the Southwestern U.S. It’s historically been one of the most anti-Tesla states in the union thanks to harsh franchise-dealer laws that have banned Tesla from having ANYTHING in this state: no service centers, no stores, no galleries. To be an owner here is to know what it’s like to have to book a hotel when taking one’s car in for service... in CO or AZ.

And now the NEWS!

It’s taken three years but we finally have the governor, the state senate, the state house of representatives, the Tesla team, Tesla’s lobbyists, my group the Tesla Owners Club of New Mexico, and the New Mexico Automotive Dealers Association (NMADA) all talking.

We have a shiny new BILL, SB 255, presented to the legislature today. And we have HEARINGS featuring owner testimony, next week! We are asking for the law to be changed to allow stores AND service centers!

Wish us luck!

 

[Curt Renz]

Sure. Anything is possible. Don’t disagree there. But in the context of everything else like continued missed deadlines, team leadership turnover and repeated failure of even the smallest promised deliverables, at some point we need to employ Occam’s Razor and go with the simpelst explanation that fits all the facts.

At this point the brightest sliver of hope for the FSD program is @verygreen ’s conclusion that they finally started testing in December based on the appearance of high precision data mapping. That means Tesla is roughly 10 months behind Cruise and a year or two behind Waymo wrt testing.

Of course your fears could be valid, but my stab of Occam's Razor indicates that Tesla keeping information from competitors is the simplest explanation. The same for some other automakers who are dealing similarly with California's request.

In any event, being first shouldn't really matter. Governments are undoubtedly going to drag out the time required for complete approval of fully autonomous driving. By the time the necessary regulations become implemented, I expect that many companies will be able to comply.

 

[Mike Smith]

Tesla is the only car company even attempting to make self driving cars. Everyone else is working on limited use autonomous taxis. When Teslas become self driving they will also be by far the best autonomous taxis since they'll work everywhere.

The reason nobody else is working on self driving cars is that their customers won't allow them to gather the billions of miles of real-world data necessary to achieve it.

 

[verygreen]

At this point the brightest sliver of hope for the FSD program is @verygreen ’s conclusion that they finally started testing in December based on the appearance of high precision data mapping.

Actually They started the coast to coast FSD video in December. The maps (Very spotty!) have many segments in California and some are much older. There's a spot in Los Angeles, spot in Edgewood, other random Bay area bits.

 

[verygreen]

Teslas become self driving they will also be by far the best autonomous taxis since they'll work everywhere

Current code only works if you have the maps = same as all the others.

 

[zmarty]

Therefore, I do indeed claim that Tesla can run FSD with a nag added and not be required to report anything.

Link to CA legislation

The document you linked to says no such thing, as several other people have stated in the autonomous forum.

Here is what the document says:
"An autonomous vehicle is operating or driving in autonomous mode when it is operated or driven with the autonomous technology engaged."

It clearly does not matter if the driver has the hands on the wheel or anywhere else on the car. If the technology is running and controlling the car on California roads, then they need to report it.

What I definitely would believe is happening instead is that they are testing EAP with what Mobileye coined Level 2+, which is a combination of lane sensing and HD maps to better steer the car in the lanes. But that is still level 2 and nowhere near FSD. And they don't need to report that.

 

[schonelucht]

MODERATOR COMMENT:

After studying the links provided in post #1810 above, it seems prudent to alert the date-unwary that all the references therein are from 2016 and the beginning of 2017. Without poring through the entirety of the reddit post, it appears none of its participants caught the seeming time-warp.

While I appreciate the very difficult task that is moderating and the hard work that goes into it, I object very strongly against using the moderator priviliges to make an argument on the content. Even if you want to make a factual correction, you should do it as any other poster on this board would. Through a regular reply and ABSOLUTELY not by modifying the original post (or headlining your reply with 'moderator comment' which is also bad but less bad). Please reserve those privileges when dealing with personal attacks, off topic posts and deliberate trolling, not when discussing content that is perfectly acceptable according to the rules.

On to the substance, and this makes it even more painful : you are also wrong. The electrek article is from early 2017 but the reddit reference is from 11 hours ago and links to recently released data discussing performance up to December 2017.

I want to re-iterate my very strong objections against you using moderator privileges to make a point in the discussion. That your interjection as a moderator had factual errors too makes it even worse. Can we have an official reply from the moderators what their policy is going forward on modifying user posts on the sole basis that moderators feel they must warn against the content?