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

[izemize]

So far absolute zero report on Trump tariff threat in Chinese domestic media, must be the Chinese government, IMHO looks more like they are working out the differences than a break-down on trade deal, otherwise I should have seen tons of Trump bashing in Chinese domestic media.

The Chinese stock market seems to know about it...
-5.9% is a big drop in my book.

 

[sundaymorning]

I have had a few experiences where the car took over properly. It made, in my opinion, accurate decisions. Last experience was on a city street where the car thought I was too close to a bicycle rider, as I passed the bike my Model 3 moved over a foot or two to the left and then brought me back to center of lane.--this was without AP.

A dear family friend on a bike, riding in a bike lane, was hit and killed by an H2 Hummer. Had it been a Tesla, the young man would still be alive today.

So when Autopilot is NOT engaged the car still takes over when necessary?

 

[Carl Raymond]

That said, I am still undecided at how the public will respond when the first FSD accident takes place. I think a large factor in that will be how many miles FSD drives without an accident before that first accident occurs.

I think an impossible to ignore stat will be the fault ratio. Every collision will have video. If FSD has been deemed good enough to gain regulatory approval, it will very quickly become apparent that of every 100 accidents, more than 90 were the fault of another party (driver, cyclist, biker, pedestrian, road crew).

Insurers will notice that they pay out to Tesla (insurance division), but rarely if ever collect from.

 

[Pezpunk]

So when Autopilot is NOT engaged the car still takes over when necessary?

not only that, these safety features are active even if the car does not have Autopilot.

Tesla Announces New Safety Features For Non-Autopilot Driving

 

[Doggydogworld]

I think it is very clear Tesla plans to move as much of its prediction and driving policy to NNs as possible.

Sure. Everyone does.

I would guess at the moment Tesla’s 3D scene is almost entirely populated by Vision and radar neural nets, including NN distance and velocity labels.

Pretty sure their NN is camera only.

They are increasingly moving object future path prediction to NNs too. This includes predicting the future curve of the road and predicting future “cut ins” from NNs which have been trained to identify them before they happen.

All done from a single 2D frame. They trained an Inception-style NN to label objects as car/truck/pedestrian/etc along with a distance estimate. They also do a little left/right path prediction using location and pose, probably just on certain objects. They apparently also run each frame through a similar NN that's been trained to identify driveable space and obstructions, along with a "curve" hint. These NNs feed bounding boxes and labels to the driving policy heuristics. This delivers the stuff Karpathy claims and is consistent with what Bowers, Bannon and Musk said.

Heuristics still take care of driving policy, but if a NN tells Tesla there will be an object in its current planned path in 3 seconds...

I doubt these NNs do time predictions. They seem to give location in terms of pixels and distance. Heuristics convert these 2.5D objects into a 3D view using radar pings and probably other tricks to cross-check. Heuristics can also translate the left-right prediction labels into 3D path predictions, though this seems to be in early stages.

Dojo and "video in, steering out" are fun and games at this point. Nothing in the work Karpathy discussed seems amenable to that kind of approach. I see an incremental path, e.g. adding NNs trained to give more 3D-ish velocity labels, adding multiple frames and/or RNN features.​

 

[oneday]

Dojo and "video in, steering out" are fun and games at this point. Nothing in the work Karpathy discussed seems amenable to that kind of approach. I see an incremental path, e.g. adding NNs trained to give more 3D-ish velocity labels, adding multiple frames and/or RNN features.

It sounded to me like Dojo was part of their path to have a NN determine driving behavior, instead of heuristics. It also sounded to me like they are at infancy (give or take) for this project. Would you agree?

 

[Buckminster]

Tesla Insurance
New thread on insurance. I have included 2 recent great posts.

 

[PANN]

The Chinese stock market seems to know about it...
-5.9% is a big drop in my book.

Yeah everybody in the market know about it, but the media is forbidden to talk about it at all so far.

 

[Fact Checking]

by an order of magnitude

Ready for this phrase to die

What does this mean ? You don't understand the phrase or do you think I used it wrong ?

Overused and fanboyish.

It's like how middle school girls try and sound like a Kardashian when they speak. Folks here trying to sound like Elon.

"Order of magnitude" is a routine term of art in physics, and it often comes with the understanding that your related guess might be off by plus minus half an order of magnitude, i.e. by a factor of 5.

I routinely used it well before Elon traveled to the U.S. for the first time, and I'll continue to use it - even though I am admittedly a big Tesla fanboy.

I.e. statements like "2-3 orders of magnitude" or "an order of magnitude" in informal communications often implicitly come with very, very large error bars. Which is how physicists like Elon are using it - and @EVNow's usage was correct in that sense as well.

Furthermore there's no equivalent replacement for "orders of magnitude": saying "10x larger" implies far more precision than intended.

So, to use a down-under term of art in response to your misinformed and unsolicited language policing attempts and insults: bugger off.

 

[vigleik]

And we’re down from $255 to $248 in early premarket trading. I guess it’s not unexpected, based on the bad macros.

 

[PANN]

Just saw first Chinese domestic media report came out on trade deal -- China government speaker said the negotiation team is preparing to go to the US for next round negotiations. Seem setting the narratives now, after all day silence.

 

[gene]

So when Autopilot is NOT engaged the car still takes over when necessary?

Correct, even when AP is not on. My Model 3 has AP, but my wife's Model 3 does not have AP. In her car which does not even have AP we had the car steer towards the left a couple of feet where there was a construction barrier jutting a little into a lane in San Francisco.

 

[Fact Checking]

Just saw first Chinese domestic media report came out on trade deal -- China government speaker said the negotiation team is preparing to go to the US for next round negotiations. Seem setting the narratives now, after all day silence.

My take: Trump kept this card and he uses it like he used it last time: to distract from the Mueller testimony which is likely going to include highly damaging bits like the conclusion of the Mueller report that Donald Trump engaged in felony obstruction of justice when they destroyed Russia evidence, gave false testimony, threatened and tried to influence both investigators and witnesses, etc.

I.e. the likely conclusion of the Mueller report was that while prosecutor didn't find clear-cut evidence of Russia collusion but found clear-cut evidence of Trump trying to hide, destroy or distort evidence of Russia collusion and other crimes.

Robert Mueller (a Republican appointed by another Republican - i.e. hard to accuse of bias in a credible fashion) is possibly going to testify in the next two weeks, while Team Trump is trying to keep him from testifying - or should that fail, I expect fireworks of distraction.

 

[mongo]

All done from a single 2D frame.

The NN is fed two frames per @jimmy_d

The V9 network takes 1280x960 images with 3 color channels and 2 frames per camera from, for example, the main camera. That’s 1280x960x3x2 as an input, or 7.3MB. The V8 main camera processing frame was 640x416x2 or 0.5MB - 13x less data. The extra resolution means that V9 has access to smaller and more subtle detail from the camera, but the more interesting aspect of the change to the camera interface is that camera frames are being processed in pairs. These two pairs are likely time-offset by some small delay - 10ms to 100ms I’d guess - allowing each processed camera input to see motion. Motion can give you depth, separate objects from the background, help identify objects, predict object trajectories, and provide information about the vehicle’s own motion. It's a pretty fundamental improvement to the basic perceptions of the system.

 

[SPadival]

Mercedes EQC production launched. It beats Model X standard on price and range. However, I bet the production will be low and Daimler will lose money hand over fist selling these puppies (battery size is 80kwh).

Mercedes-Benz EQC sales release & start of production: Electrified Mercedes hits the road | marsMediaSite

 

[EVMeister]

Mercedes EQC production launched. It beats Model X standard on price and range. However, I bet the production will be low and Daimler will lose money hand over fist selling these puppies (battery size is 80kwh).

Mercedes-Benz EQC sales release & start of production: Electrified Mercedes hits the road | marsMediaSite

The Model X is a significantly larger vehicle though. It should cost more and given the extra weigh it's not unexpected for the range to be lower. Or am I missing something here?

 

[Fact Checking]

Mercedes EQC production launched. It beats Model X standard on price and range. However, I bet the production will be low and Daimler will lose money hand over fist selling these puppies (battery size is 80kwh).

Mercedes-Benz EQC sales release & start of production: Electrified Mercedes hits the road | marsMediaSite

So, regarding range, it says this about the Mercedes EQC, according to the old and discredited NEDC testing standard:

"In combination with the extensive services and a range of 445 - 471 km (NEDC)1"​

While the Standard Range Model X is listed with a WLTP range of 375 km.

I believe the multiplier from NEDC to WLTP is around 0.887, so the EQC has approximately a WLTP range of ~397 km.

Long Range Model X has a range of 505 km WLTP - crushing the EQC by about 25% if I calculated it all correctly.

An interesting detail:

"Prices for the Mercedes-Benz EQC start at €71,281[2] for the already generously equipped base version. Its net basic price of less than €60,000 makes it eligible for an environmental bonus in Germany – regardless of individual equipment."​

Is the "net" price without the 19% VAT, i.e. net price is €59,900, which with VAT becomes €71,281?

And yes, Tesla needs the Model Y.

 

[mutle]

So how can pricing start at 71k while still eligible for the 60k bonus? Some loophole?

The €60k bonus is based on pre-tax (brutto) price, excluding MwSt. And in this case the price seems to be right around €59999.

 

[Fact Checking]

The €60k bonus is based on pre-tax (brutto) price, excluding MwSt. And in this case the price seems to be right around €59999.

Yeah, figured that out and edited my comment meanwhile.

 

[SPadival]

The Model X is a significantly larger vehicle though. It should cost more and given the extra weigh it's not unexpected for the range to be lower. Or am I missing something here?

Ah yes.. I wasn't aware of the size difference.

And yes, Tesla needs the Model Y.

And Maxwell acquisition