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

[Nocturnal]

I've been hearing the argument that nobody has a chance against Tesla w/autonomous driving because Tesla has so much data. But this doesn't mean others can't catch up.

Here's some more thoughts on this.

View attachment 404641

How the F can you buy a $200 dashcam and have it turn your car into an autonomous vehicle?

Others can and likely will catch up, assuming Tesla is in the lead, but there are missing bits. How do you get your cars out there in large numbers quickly? How do you fuel them? Automated gas stations would be very costly. If they are EVs, how do you charge them? Would VW let GM use their electrify America charging system while VW destroys them on the FSD side?

FSD for robo taxis needs a lot of vehicles, it requires EVs, and requires a charging infrastructure. Tesla is the only company to have all of this.

This is very true. The radar can only see two cars ahead AFAIK. On a straight motorway in heavy, fast moving traffic, if the driver wants to keep up with what's going on ahead it's necessary to position their car within their lane so they can see several cars ahead and predict a slow-down or stop in plenty of time. A centrally-mounted camera looking forward is going to find this difficult, as will the radar I suspect.

Maybe one of the next improvements will need to be forward facing cameras in the door mirrors which can see around cars in front and look along the line of traffic?

Nobody actually does that though. The vast majority of the time everyone is looking at the car in front or maybe the one in front of them. ​

 

[mongo]

I've been hearing the argument that nobody has a chance against Tesla w/autonomous driving because Tesla has so much data. But this doesn't mean others can't catch up.

Here's some more thoughts on this.

View attachment 404641

If $200 dashcam footage is all it takes, Google already has YouTube (and street view)
This idea would create petabytes of unlabeled data with no driving control reference information. There is no driver provided feedback on the Google NN design. And even if Goggle did do this, it does not advance their fleet size at all.
This is not like airplanes which were reliant on engine HP/ weight ratios, there is no key technology that unlocks FSD for everyone.

Slowing sooner, not stopping faster.

Slowing sooner is more deceleration/ braking, stopping faster is also more deceleration/ braking. The only way to increase the following distance is more braking earlier.
If you have someone approaching you from behind quickly, you want to apply your brakes enough for the lights to come on, but you also want to stay as far away from them as possible (and flip a coin on avoiding an impact, being up against the next car at impact, or getting hit, then hitting the next car).

 

[TheTalkingMule]

Wonder if we will get an announcement this afternoon that the equity offering is fully subscribed. Thoughts?

WTF is this on-topic post doing in here???

 

[Cherry Wine]

I've been hearing the argument that nobody has a chance against Tesla w/autonomous driving because Tesla has so much data. But this doesn't mean others can't catch up.

Here's some more thoughts on this.

View attachment 404641

An interesting thought, but I'm not convinced. A competing Waze camera system is problematic for one primary reason: a 2-3 camera system that is not keyed into other driving metrics will have substantial deficiencies in the data provided by the fleet when contrasted with Tesla's 8-camera system. Furthermore, every additional camera that Google could offer as part of such a system in order to compensate for the data gap would, in my opinion, drive down the adoption rate as it would raise costs and increase installation difficulty.

 

[oneday]

This is very true. The radar can only see two cars ahead AFAIK. On a straight motorway in heavy, fast moving traffic, if the driver wants to keep up with what's going on ahead it's necessary to position their car within their lane so they can see several cars ahead and predict a slow-down or stop in plenty of time. A centrally-mounted camera looking forward is going to find this difficult, as will the radar I suspect.

Maybe one of the next improvements will need to be forward facing cameras in the door mirrors which can see around cars in front and look along the line of traffic?

This is probably a good idea. One could think that it wouldn’t be entirely necessary because the computer can react at the millisecond level, I can envision o few scenarios where it would still be helpful.

 

[DaveT]

How the F can you buy a $200 dashcam and have it turn your car into an autonomous vehicle?

You must not be following what the folks at comma.ai are doing. Please do your research on them.

 

[LJS22]

It will take a combined effort from multiple companies (auto and tech) to catch up to Tesla. Tesla being a jack of all trades is going to allow it to have a clear path on what it wants to achieve and what it will achieve.

Other companies will figure it out, but Tesla should like the position they are in currently.

 

[Dutchie]

Drove about 100 km with NoA yesterday. Since the last update this has been brilliant. I think I could have driven that stretch blindfolded without any problem!

At one point another car tried to merge into the freeway. My car held in a bit so the other car could merge. The lady in the car raised her hand as a "thank you". I thought to myself: "Don't thank me, thank the computer!"

The only negative is that it is slow merging onto HOV. There is only a certain spaice in which the line is not continues. The car is too slow to use that space. I blame the computer for that. I think with the new HW3 computer the car will be much more assertive.

 

[DaveT]

If $200 dashcam footage is all it takes, Google already has YouTube (and street view)
This idea would create petabytes of unlabeled data with no driving control reference information. There is no driver provided feedback on the Google NN design. And even if Goggle did do this, it does not advance their fleet size at all.
This is not like airplanes which were reliant on engine HP/ weight ratios, there is no key technology that unlocks FSD for everyone.

A Google dashcam/camera system would be completely different than Youtube and Streetview. The dashcam/camera system would have a suite of sensors like a smartphone, including gps, accelerometer, etc. Thus, the system would be taking video footage but also speed, location and various movements.

It would improve their Waze system.

And it would give a tremendous amount of data, especially if Google can subsidize it to get on millions of cars.

True, this is a bridge technology, but it can get lots of data for Google and allow them to train their neural nets.

Btw, just because Google is public with Lidar and such doesn't mean they aren't developing a completely separate/different software-only vision system similar to Tesla as well. Of course this would be a completely different system and team on it. But we don't know all that Waymo (and others) are doing right now.

 

[ReflexFunds]

I've been hearing the argument that nobody has a chance against Tesla w/autonomous driving because Tesla has so much data. But this doesn't mean others can't catch up.

Here's some more thoughts on this.

View attachment 404641

A vision self driving system needs a supercomputer, redundancy built into the car control systems, loads of wiring for the sensors etc. Retrofit cost for a real self driving solution will be extremely expensive so very hard to get into consumer's cars. Comma.AI is 12 people working on retrofit driver assist, they do not seem to have any credible path to full self driving - i'm not sure this is even their ambition.
Waymo could choose to admit their Lidar first approach is wrong and drop everything to try to copy Tesla, but this is still extremely difficult to do if you are not a vertically integrated EV manufacturer where all systems have already been designed for redundancy and are set up to be software controlled.
Traditional auto manufacturers are also extremely slow to change their car designs, taking many many years to go from design to production. I think if Waymo/other Lidar companies were to pivot to copying Tesla's approach, it would be several years before they got the hardware rolled out into consumer's cars in significant numbers, and then they would also be very far behind in building the fleet data collection & filtering systems and other software 2.0 infrastructure Tesla has already spent years building and improving, as well as behind on building up annotated datasets/training cases for 100k+ driving scenarios.

 

[Joe F]

This is very true. The radar can only see two cars ahead AFAIK. On a straight motorway in heavy, fast moving traffic, if the driver wants to keep up with what's going on ahead it's necessary to position their car within their lane so they can see several cars ahead and predict a slow-down or stop in plenty of time. A centrally-mounted camera looking forward is going to find this difficult, as will the radar I suspect.

Maybe one of the next improvements will need to be forward facing cameras in the door mirrors which can see around cars in front and look along the line of traffic?

That's the one thing I don't like about AP, and have stated as much here long ago. I'd like to be able to have AP drive off center in the lane, just so I can see around traffic. I'm not a fan of last second reactions, by AP or me.

I can see much further down the road and react faster than my Tesla can. It's far too easy to be distracted, and we all do, but I don't get why most drivers seem to fixate on the vehicle directly in front of them, and are far too close at highway speeds to react in time.

Defensive driving is a lost art I guess.

 

[Nocturnal]

You must not be following what the folks at comma.ai are doing. Please do your research on them.

That's why I asked!

Open source relying on integration with various companies...that seems unlikely. Interesting concept though.

 

[farzyness]

I've been hearing the argument that nobody has a chance against Tesla w/autonomous driving because Tesla has so much data. But this doesn't mean others can't catch up.

Here's some more thoughts on this.

View attachment 404641

Dave - the thing that I think you're missing here is that while yes - the self-driving problem can be solved by companies outside of Tesla, the overall supply chain needed to execute this self-driving problem is next to non-existent outside of Tesla. Google, Comma-AI, and others can get the software side solved in let's say ~3 years from today, but it'll take them an additional ~3-5 years to scale out the fleet necessary to compete with Tesla. This puts them in a best case scenario ~6 years behind. Let's not forget that battery supply will be the single biggest supply bottleneck for any manufacturer trying to scale out self-driving cars.

As long as Tesla maintains course in its pace of innovation then no one will catch up. Once that slows down, likely with Elon's departure or non-involvement with Tesla, then we will begin seeing that gap narrow. But until then, I'm having a tough time seeing your argument come to fruition.

 

[DaveT]

Others can and likely will catch up, assuming Tesla is in the lead, but there are missing bits. How do you get your cars out there in large numbers quickly? How do you fuel them? Automated gas stations would be very costly. If they are EVs, how do you charge them? Would VW let GM use their electrify America charging system while VW destroys them on the FSD side?

FSD for robo taxis needs a lot of vehicles, it requires EVs, and requires a charging infrastructure. Tesla is the only company to have all of this.

Nobody actually does that though. The vast majority of the time everyone is looking at the car in front or maybe the one in front of them. ​

FSD robotaxis don't require EVs. Fuel/charging costs are cheaper for EVs but ICE cars can FSD as well (again research comma.ai).

Here's the thing, if a FSD EV costs $35,000 and a used ICE that's retrofitted for FSD costs $10,000, then you're going to have a lot of ICE FSD robotaxis alongside the FSD EV ones. Regarding how you fuel them, they'll bring back full-service at gas stations.

 

[Tslynk67]

Can you try?

Let me put it politely: he's a distillation of the worst traits of human nature, rolled into one disgusting package, has no redeeming qualities or virtues whatsoever, represents everything I consider vile and makes me want to puke.

 

[DaveT]

An interesting thought, but I'm not convinced. A competing Waze camera system is problematic for one primary reason: a 2-3 camera system that is not keyed into other driving metrics will have substantial deficiencies in the data provided by the fleet when contrasted with Tesla's 8-camera system. Furthermore, every additional camera that Google could offer as part of such a system in order to compensate for the data gap would, in my opinion, drive down the adoption rate as it would raise costs and increase installation difficulty.

The main cameras that FSD relies on are the front cameras. All the other cameras are minor in comparison. Sure you need side cameras for lane-changing, etc. But the main cameras are the front ones, just like we rely on our eyes looking forward for 99%+ of driving.

I'm not saying a Google dashcam/camera system is the end-all for Google FSD, but what I'm saying is it's a bridge for them to get data to train their neural nets and allow them to build a somewhat decent semi-FSD solution that can get on cars and get to FSD.

 

[EVNow]

FSD robotaxis don't require EVs. Fuel/charging costs are cheaper for EVs but ICE cars can FSD as well (again research comma.ai).

Here's the thing, if a FSD EV costs $35,000 and a used ICE that's retrofitted for FSD costs $10,000, then you're going to have a lot of ICE FSD robotaxis alongside the FSD EV ones. Regarding how you fuel them, they'll bring back full-service at gas stations.

Regulations will probably say robotaxis have to be EVs. LA mayor already has made that rule (or said he will). I can definitely see a lot of states, as well as EU, doing that.

ps : Not sure ICE can be retrofitted that cheaply. Regulations will probably require lots of built in redundancies.

pps : Large robotaxi operators will at least start having CAFE, GHG like regulations to comply with.

 

[DaveT]

Fairly sure the regulations will say robotaxis have to be EVs. LA mayor already has made that rule (or said he will). I can definitely see a lot of states, as well as EU, doing that.

I'll take the opposite side on this. I don't think regulators will require robotaxis to be EVs. I think they'll look at it as purely an autonomous driving/safety issue and not as an environmental one combined.

 

[RobStark]

Really good points, but given Robo Taxis aren't realistic for another 2 years (if you'r on Elon time) or 10 years (if you're in the AI business and know wtf you're talking about) it might have made sense to build the potentially better-seller first, and roll out the Model 3 in 2020, just in time for FSD (apparently).

Now I love the Model 3, it's exactly the electric car I wanted, minus the panel gaps, and I might not have been so enthusiastic about the Y. But it seems most 'Muricans love their small SUVs and the Y potentially could have had bigger initial demand and a much wider tail in the US than the '3 has had.

Tesla couldn't have built more Model Ys than Model 3s to date.

Production constraints, battery cell constraints or whatever. Since Model Y will be less aerodynamic it will need more cells per vehicle to attain the same range as Model 3.

And Tesla would've needed to charge $4-5k more for Model Y, putting a rather large group of people outside of Tesla's price range for another ~4 years.

For Tesla it is always smart to build the sedan first (S and 3) then the CUV ( X and Y).

 

[ZeApelido]

Karpathy even appeared to wince when Musk started making his grandiose timeline claims.

And Karpathy doesn't even understand the full scope of the problem yet, because the cases he presented as "very unusual" are what I call "normal, everyday driving". In a couple of years they may start looking at the unusual cases.

I was thinking about this yesterday, I am really curious to analyze his body language whenever Musk talks boldy in the presentation. Anyone pick up on anything?