Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[Krugerrand]

Because if it misses a car in the adjacent lane going 70 mph, it will have just swerved the car moving a few MPH in front of a car going 70 mph, perhaps causing multiple fatalities. Nope. Not buying it.

But two people have already told you the car has performed the maneuver (one of those people experienced it multiple times) and a third person has video of it. Therefore it must not exist because you don’t think they’d do it? Um...ok.

 

[Dutchie]

Deliveries don't start in the UK until June and ordering was only available in May. So presumably April is too early for VIN registration for the UK? Apart from a few showroom models. Not sure how that works.

Remember Model X and S?

 

[Bladerskb]

Yes, I think that's the correct perspective. Tesla does have a 2-3 lead right now, and they have a chance to be the winner in a huge autonomous transport market. However, there are many scenarios that can play out. And I think it's far from guaranteed that Tesla has the largest autonomous transport platform in 10 years.

What I do believe is likely is that there will be a lot of competition and fierce competition, and yes there will be a price war and it will become ridiculous cheap to transport people/things. Margins will be super low and most companies will go out of business in this field. The winner in the end will win by outlasting everyone else and charging the least. I don't think it will be pretty. But in the end, whoever has the dominant autonomous transport platform will be able to monetize through all the services that can be built on and through that platform.

2-3 years lead in what exactly? Neural networks? Chips? software? mapping? sensor redundancy? Because they trail in all of that.

 

[EVNow]

Actually EyeQ5 will be out Q1 2020 and currently Mobileye beats Tesla in development and deployment of neural net models. When you compare the NN in eyeq4 and what Tesla currently has in production. Its not even close.

Secondly, Its actually been Tesla that has been copying Mobileye from the very start, the tri-focal camera invented by mobileye. The 8 camera configuration system also invented and patented by mobileye. There has to be dozens of mobileye patents that Tesla is using.

Mobileye literally has almost all the computer vision patents for self driving cars out there. For example the path prediction you saw Tesla show in their Autonomy Day is a mobileye patent. The only one doing any copying its Tesla. Mobileye copies no one.

Thirdly Mobileye does have their own complete system with 360 Vision, Mapping and Driving Policy with RSS Safety Envelope that they will start offering..

Lastly EyeQ5 is more efficient than Tesla's FSD chip.

There is no such thing, the amount of misinformation being posted in this thread is astonishing. I initially attempted to correct some of them, but there's just too many.

Can you provide links for all your claims ?

 

[SPadival]

When you compare the NN in eyeq4 and what Tesla currently has in production. Its not even close.

So where are the customer YouTube videos of eyeq4 in action? Can you please link some of them here.

 

[DaveT]

I’ve been watching Musk in just about everything since the late 2000s.

I have never seen him more assertive and confident (and more frustrated that others don’t see it) than at the autonomy event.

Probably the last time I saw Elon this confident was when he tweeted this.

 

[dc_h]

Finally, the point is not about Lidar or/and vision.

The question is how is anyone going to deal with edge cases. Lots and lots of them, to make FSD highly reliable.

If a lot of data is needed to handle edge cases, how does one get the data? Tesla’s solution is to put the sensors in all cars and get the data. Waymo’s is to simulate.

I think Tesla is simulating also, but primarily from edge cases. Run a sim NN and treat the data the same as physical data, but have the system work on every disengage scenario and run sims until the car has the right response every time. If they can automate the process of capturing thousands or millions of disengage events and run sims, they can solve the edge cases pretty quickly. If they solve for recurring edge cases with common attributes they will solve high value cases very aquickly.

 

[NicoV]

Google already has a system. You use it every day when you check the box that says.. I am not a robot. You then tag pictures of buses, fire hydrants, storefronts and cross walks. Your google car will never run into a fire hydrant or a store front, I can assure of that. I cannot vouch for much beyond that.

Unless Tesla makes a ‘Blues Brothers’ mode in addition to ‘Mad Max’ mode.

 

[CorneliusXX]

Some people here would argue Elon doesn't have a credibility problem. Go figure.

The sad thing is that his credibility problem is largely manufactured by others, and because credibility is predominately a subjective matter it is difficult to change it while such a powerful short campaign is being waged.

If all I did was read the secret master plan part 1 and then see what is being produced, it would be impossible not to see Elon as credible. However the tweets and optimistic timelines have taken a relatively small set of issues and been spun out or proportion to the point that there is a genuine credibility problem for many.

 

[Cattledog]

Tale of Two Teslas

As a long term shareholder (since 2011) and car owner (since 2012) there's something about the recent shift in emphasis on FSD that has me troubled:

When I bought our first Model S, and then every subsequent Tesla, while the reasons were multiple, the overwhelming reason was it offered a superior driving experience. Quicker, quieter, road-hugging, single-footing, aerodynamic, visually striking, extending door handles (extending door handles!) - all of it redefined the driving experience for me. I want Tesla to still be about that for a long time, and I think it will.

But...

They are now selling themselves as the future of the superior riding experience. That's ok as part of the story, because sometimes I want that story. And many people want that story. But the recent shift to primary emphasis on it causes the narrative about the driving experience to be lost. That's a shame. because its the more visceral reaction, and it has great play.

Tesla - be bold about the dual experience you get out of any of your cars - currently, and in the future, the best driving experience thanks to a substantial lead in electrification. In the future, and a right now a little more every day, your same car will give you the ultimate riding experience, relaxation in travel, time on your hands, etc.

Sell both, hard. You'll then have created the ultimate ownership experience, where you get all of the above and AirBnB cash on wheels if you choose, without sheets to wash.

 

[Carl Raymond]

The sad thing is that his credibility problem is largely manufactured by others, and because credibility is predominately a subjective matter it is difficult to change it while such a powerful short campaign is being waged.

If all I did was read the secret master plan part 1 and then see what is being produced, it would be impossible not to see Elon as credible. However the tweets and optimistic timelines have taken a relatively small set of issues and been spun out or proportion to the point that there is a genuine credibility problem for many.

Yes. Like pointing a weapon at somebody and saying “You have a problem”.
Not false, but the name of the problem is the name of the one holding the weapon.

 

[30seconds]

Actually EyeQ5 will be out Q1 2020 and currently Mobileye beats Tesla in development and deployment of neural net models. When you compare the NN in eyeq4 and what Tesla currently has in production. Its not even close.

Secondly, Its actually been Tesla that has been copying Mobileye from the very start, the tri-focal camera invented and patented by mobileye. The 8 camera configuration system also invented and patented by mobileye. There has to be dozens of mobileye patents that Tesla is using.

Mobileye literally has almost all the computer vision patents for self driving cars out there. For example the path prediction you saw Tesla show in their Autonomy Day is a mobileye patent. The only one doing any copying is Tesla. Mobileye copies no one.

Thirdly Mobileye does have their own complete system with 360 Vision, Mapping and Driving Policy with RSS Safety Envelope that they will start offering..

Lastly EyeQ5 is more efficient than Tesla's FSD chip.

There is no such thing, the amount of misinformation being posted in this thread is astonishing. I initially attempted to correct some of them, but there's just too many.

So Intel is just fine that Tesla is violating and monetizing multiple MobileEye patents?

Yes, sounds just like Intel.

 

[RobStark]

explanation: New Jersey is as far as I know the only state that outlaws self-service gas pumps.

Ditto Oregon.

 

[Maitri982]

So where are the customer YouTube videos of eyeq4 in action? Can you please link some of them here.

Haven't you seen all the mobileye cars signalling on their own and changing Lanes, etc.?

And geez, who would ever think that we should put 3 cameras looking forward!?! Those mobileye guys are genius. Let's file patent for that!!

 

[Singer3000]

The sad thing is that his credibility problem is largely manufactured by others, and because credibility is predominately a subjective matter it is difficult to change it while such a powerful short campaign is being waged.

If all I did was read the secret master plan part 1 and then see what is being produced, it would be impossible not to see Elon as credible. However the tweets and optimistic timelines have taken a relatively small set of issues and been spun out or proportion to the point that there is a genuine credibility problem for many.

It’s highly likely that in a few short years SpaceX will put humans in orbit around or even on the moon. It is likely that they will thereafter take enough plant and equipment to Mars that a manned mission seems credible rather than science fiction. It is possible that this manned mission may follow before 2030.

It is highly likely that Tesla will in a few years produce more vehicles than BMW. It is likely that they will push at least one major auto producer into bankruptcy. It is possible that they will become the face of the autonomous transport revolution.

People hear are sweating the small stuff. Elon’s “credibility” will do just fine with each step of successful execution.

 

[BioSehnsucht]

I am a software project manager. I know the drill.

No way is Tesla and Elon going to have THIS in the release notes:

“We’ll slightly correct steering if you start to go out of your lane”

But not this:

“If you get rear ended, the car will steer into adjacent lanes if possible to avoid obstacles ahead and attempt to minimize front end damage”.

Common sense guys. Now you’re starting to make ME feel like some TSLA bulls are blinded by the light...and I’ve probably been in the club longer than most of you!

(Joined TMC 2009, bought TSLA shortly after IPO).

There doesn't have to be "I've been hit from behind" logic just "try to stay in drivable space" logic. For whatever reason, the car suddenly accelerated and was likely to hit the vehicle in front, so it steers away to open space. This doesn't even have to rely on considering whether the Tesla is accelerating or the vehicle in front is decelerating, just that the distance is closing and a collision is imminent - the relative motion is essentially the same regardless of which is changing velocity.

Many have reported having the car move to avoid being hit by a vehicle encroaching on it's space (whether lane changing into them or merely drifting over the line). I don't see how it's a stretch that it could also steer to the side of an obstruction if it was about to hit it. Depending on how specific the logic was, it could be all the same set of code, and not an additional variation of collision avoidance.

 

[schonelucht]

So Intel is just fine that Tesla is violating and monetizing multiple MobileEye patents?

Yes, sounds just like Intel.

Why sue now and give Tesla the ability to avoid using your patents with little to no financial gain to yourself when you can wait until there are millions of cars on the road, it’s too burdensome for Tesla to overhaul the installed base and you can extract fees for every single robot driven mile? Trust me, the lawyers will come out.

 

[Fact Checking]

Why sue now and give Tesla the ability to avoid using your patents with little to no financial gain to yourself when you can wait until there are millions of cars on the road, it’s too burdensome for Tesla to overhaul the installed base and you can extract fees for every single robot driven mile? Trust me, the lawyers will come out.

Five comments:

1)

I find it probable/possible that Tesla cross-licensed most of MobilEye's patents that resulted from their early partnership, in which case Intel couldn't sue Tesla based on those patents. Tesla was MobilEye's main source of income and it's routine to not build a sole IP dependency on such partners. (In a similar vein key battery tech between Panasonic and Tesla is possibly cross-licensed as well which makes it harder for the parties to betray each other.)

The assumption that MobilEye was allowed to undermine Tesla via submarine patents financed by Tesla income is IMHO naive.

(Anyway, it's pure speculation: I don't know whether they did so (they might not have), and the agreements would generally be confidential.)

2)

Also, some of the earliest and broadest patents of the MobilEye founders have expired last year already, such as "Synthesizing virtual two dimensional images of three dimensional space from a collection of real two dimensional images" (US6219444B1).

3)

Tesla is likely aware of how patent-encumbered certain MobilEye technologies are: for example one feature that even the latest Autopilot version curiously doesn't have is reading and interpreting speed limit signs - which might be affected by a related "late" MobilEye software patent that issued after their partnership ended. (For example: "Detecting and recognizing traffic signs" US8064643B2, EP2383713B1)

I.e. Tesla might be actively working around existing patents (in this case by using GPS indexed speed limit data), to reduce patent litigation exposure.

4)

Intel delaying such a lawsuit would reduce their potential windfall from any patent lawsuit: the moment Tesla is warned of alleged patent infringement the damages increase.

5)

Tesla has numerous defensive patents. They do license them freely to everyone (see Tesla's patent pledge), under the condition that those patents are used in good faith. Suing Tesla is not good faith use and the moment any of the big manufacturers sues Tesla they can counter-sue on the basis of the countless EV patents Tesla owns, and shut down any carmaker's EV program that uses MobilEye technologies.

This is why patent litigation is usually done by 'patent trolls' who have no real business activities and have no skin in the game.

Intel could still sue Waymo and Lyft - although Waymo is possibly a dangerous target, due to their historic deep learning and NN roots.

So I'd be surprised if Intel sued Tesla based on the MobilEye patents. Maybe they'll make an exception to monopolize a trillion dollar market though, so what you are suggesting isn't an implausible outcome - but I think the complete picture is more nuanced.

 

[aubreymcfato]

Interesting article by Brad on LIDAR vs vision (it defends Lidar, and I don't trust everything he says, but it's good summary and a fair commentary for a mainstream newspaper)
Elon Musk's War On LIDAR: Who Is Right And Why Do They Think That?

 

[Fact Checking]

More along the lines of having both chips on the board (as was initially planned)

Tesla saw the point. The plan to keep both on the board was theirs. All I can do is speculate. Maybe skip NVidia completely and go directly from MobilEye to HW3 NN chip? Maybe continue to use functionality that they still don't have today (reading speed signs the most obvious one). I am sure all the usual posters can add their own speculation.

Do you have a source for this claim? I do not doubt that Tesla said something along those lines, I just think you are misreading it, so I'd like to double check.

Tesla could use the Mobileye chip output as a labeling engine for their new NN. It would also avoid the feature cliff that occured going to AP2. Once the NN could do everything the Mobileye chip could, they would stop installing it.

Literally 99%+ of the NN processing capacity is dedicated to the "labeling engine", the first step of interpreting the input of 8 cameras sending about 1000 megapixels of imagery per second (!). That step is the one that requires an ASIC that can compete with the human brain's visual processing performance.

You could write the remaining NN that is processing the already labeled visual output in JavaScript, running in a virtual machine running an emulated CPU, and still be well below the power budget.

So I'm highly dubious of the claim that two NN ASICs would be sensible in any board design. What I can imagine is that Tesla initially communicated their chip efforts as a plain old ARM based SoC development (not NN ASIC development), and this is what @schonelucht misinterpreted - but without seeing the source he is relying on it's hard to tell.