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

[Discoducky]

Since it is the weekend, I'm typing out more...sorry if this is better in another thread...

TL;DR - Tesla is at the top of the iceberg with integration of non-real-time ground truth data augmenting the real-time camera data and I hope subsequent builds will make huge progress to elimination of hesitations, lateral jerk, ghosting and lane changes.

Longer version...

I like solving problems and I'm hoping this helps as well...

Hesitations due to current real-time software pixel space limitations can be solved with higher confidence when the planned path is no longer an issue. Higher confidence is achieved with *some* amount of non-real-time ground truth (vector space) data of the next road segment at challenging areas. This is way forward I think they are taking and I think it will work and not reach a local maxima prior to solving all blocking challenging areas. This is is analogous to driving with only the on-screen visualization vs supposed 'super-human' memory (aka vector space ground truth) of previous trips where you've navigated the same road segment prior augmenting the current visualization.

High lateral jerk (quick succession, rapid steering wheel movements) at slow speeds (roughly 1 to 10 mph) are an issue I'm not as certain on how to fix. These are currently caused by the lane centering code which used to be solely based on Kalman smoothing/filtering and it seems it is still done that way as I'm noticing the exact same behavior since I was there in 2015. This type of control method is great for higher speeds, but totally falls over at slow speeds due to the calculations never settling. The signal to noise ratio gets out of whack. At its worst, the steering wheel will wildly move for several seconds at slow speeds as it thinks it is off the center of the lane (i.e. off center from where the path planner is commanding it to be) by an inch or less. This has nothing to do with confidence of occupancy or path planning in general, it is simply a control smoothing issue. I'd love to have a discussion with control engineers on how to make this better... The problem is that high jerk is needed to make/complete parking maneuvers, tight right/left turns

Lane changing issues will get better with feeding in a different type of up-coming road segment data (it seems like the real-time system is being fed with deterministic data on which lane it needs to be in and the path planner acts on this with very high confidence where it never did this before), which I see it doing better with the current build. It is worse in some areas however (regressions always occur with big step changes) but I'm not worried. Trained model weights need to be tweaked most likely.

That does make sense in terms of the precision gained by mapping markers to vector space as opposed to pixel space. And it meshes well with the previous AI day info we have regarding merging the multiple camera sources in to a single vector space used by the NN's.

But your info above seems to be a description of what FSD does with the data once it has it. What I was trying to understand was your statement that: "Creep wall and Median box which is powered by non real-time data that was collected by normal FSD cars and turned into ground truth."

I read that as indicating that the creep wall and median box areas are based on data collected by previous cars (perhaps incorporated in to some map-based data markers used by FSD cars?), rather than gleaned by the current car at the time it encounters that scene.

Are you saying the fact that the creep wall and median box features appear to be vector-space objects that is evidence of previously collected data?

Or maybe I'm mis-understanding what you are trying to say.

(again, thanks for your insights)

You got it!

 

[Silent Ludicrosy]

The creep wall gives me confidence the car isn’t going to dart into traffic. But the car needs to creep a little slower to give oncoming traffic the same confidence. I have a lot of disengagements to not annoy other traffic.

 

[Discoducky]

Is this kind of like HD mapping?

First off - I don't like the term "HD mapping" as it is too generic, but shares some attributes, yes.

Secondly - What Tesla is doing is inherently much more radically dynamic (HD maps are inherently static and useless IMO as they can't be trusted for path planning) as I believe this ground truth is going to be updated very often. For instance, like @scaesare well re-stated, as Tesla's pass through these road segments and see what previous Tesla's have seen then there is no change to the ground truth for that road segment. **IF** enough Tesla's go by to change that knowledge then it will be recalculated (ground truth will then be reestablished) and fed back to the fleet (aka fleet ground truth learning cycle). My hope, back in the day, was to do this within 1 to 2 seconds (i.e. the next car that approaches that segment would get an instantaneous ping NOT to trust the upcoming segment at some level of trust or even get a new set of ground truth). The data flow suggests that this is the first principles limit. I put a lot of time into that spec, but the high level is that each vehicle would have a trigger or a level of delta data (i.e. something different like a pothole or repainted lane line or even water on roadway...stuff like that) that would cause it to save off more robust data when it encounters a road segment that sets off the trigger. That data is sent back for further analysis, pre-processing occurs and it spits out new ground truth for that segment to feed into cars in the near vicinity thus allowing any new travel to have this updated knowledge. It is only bound by speed of travel as this is small bits of pixel data that can be quickly turned into vector data by a large compute cluster.

 

[Knightshade]

2nd, we also learned from the Tesla order page that Tesla is no longer taking new orders for LR/AWD Model 3. This implies that all Fremont Model 3s are now LFP/SR+ variants except for the highly profitable M3P LR/Dual-motor (these were about 10% of total Model 3 volume previously). So maybe 600/wk MP3 before, and maybe a bit more now since more people who wanted a 3LR now make their 2nd choice an upgrade to M3P rather than a downgrade to SR+ version. So maybe 1K/wk LR packs to Model 3 after? That's only if Tesla can get 5K LFP packs/wk for SR+ which is unverified.

While it may take some time to work through the backlog of 3LR orders (4-6 wks?), after those ~4K/wk LR packs could be freed up to go to Austin to support Model Y Production (very rough guess; large error bars). But none-the-less, Tesla is clearly prepared to continue the Model Y ramp in Austin w. or w/o 4680s. I think both will be made for quite a while.

Cheers!

Right before they stopped taking orders wasn't the delivery window for new ones late 22/early 23? That seems more like 4-6 months, not 4-6 weeks to fill backlog... which would kinda entirely negate your theory there. There'd be no "freed up" packs as they'll be used to fill already existing M3LR orders all year.

The best theory about the actual reason they pulled new orders for that config I've seen is they're working on a plan to reintroduce it (possibly with a tiny feature reduction- which'd also add more "value" to the P upcharge) at 55k to qualify for the new tax credit Jan 1.

 

[Stretch2727]

The creep wall gives me confidence the car isn’t going to dart into traffic. But the car needs to creep a little slower to give oncoming traffic the same confidence. I have a lot of disengagements to not annoy other traffic.

Having been in a 2 accidents where someone actually pulled out and ran into the side of my car (I live in NJ) I am very sensitive to people creeping into an intersection. I tend to slow down and move left if I see a car moving as I don't know whether they are going to pull out or if they are trying to get a better view. Also, Some are just trying to get a head start with acceleration and are timing the turn to an open space.

To me the proper behavior is to stay still if there is no possibility to pull out. Creep when no cars are coming to get a better view down the road and then make the turn.

 

[Tim S]

Tesla Twitter, 10,000 Model Ys built at Giga Texas to date

We picked up my wife's Made In Texas Model Y Thursday and the VIN ends with 0116xx.

 

[wipster]

We picked up my wife's Made In Texas Model Y Thursday and the VIN ends with 0116xx.

I see you posted back on Aug. 30th that your wife had gotten the VIN of 0116xx, so that means they must have produced 10,000 a week or so prior eh?

Excellent... just curious, do you have any idea if it has 4680's?

 

[garrett5688]

My first time driving a Tesla showed me why people are so obsessed with Elon Musk's electric cars

The Tesla Model Y is quick, packed with fun tech, and simple to charge. But not everyone will love its big touchscreen and lack of regular buttons.

www.businessinsider.com

How can a journalist covering EVs have never driven a Tesla in 10+ YEARS?? Are these the monkeys we are up against?

 

[Tim S]

I see you posted back on Aug. 30th that your wife had gotten the VIN of 0116xx, so that means they must have produced 10,000 a week or so prior eh?

Excellent... just curious, do you have any idea if it has 4680's?

Anyone know of a way to tell if it has 4680's?

 

[DarthPierce]

Anyone know of a way to tell if it has 4680's?

Is it an LR or a SR?

 

[Singuy]

Anyone know of a way to tell if it has 4680's?

I thought all 4680s are the Sr+ Ys model

 

[jerry33]

My first time driving a Tesla showed me why people are so obsessed with Elon Musk's electric cars

The Tesla Model Y is quick, packed with fun tech, and simple to charge. But not everyone will love its big touchscreen and lack of regular buttons.

www.businessinsider.com

How can a journalist covering EVs have never driven a Tesla in 10+ YEARS?? Are these the monkeys we are up against?

You might be surprised at how many reviews are written where the reviewer never drives the car. I recall several 2001 Prius reviews that had an error obvious to anyone who drove the car. So one reviewer made an error and the other reviews were copying the first review.

 

[Tim S]

Is it an LR or a SR?

LR

 

[ZeApelido]

Found the philosopher for me.

https://twitter.com/x/status/1571337302416437252

 

[Singuy]

More general hate toward Tesla. His final conclusion was "I would take this over a Model S hands down, I just wish it would work".

So apparently the car would just break at random, leaving you stranded for 15 mins at a time. Also everything else that's software related barely work.

 

[Gigapress]

My first time driving a Tesla showed me why people are so obsessed with Elon Musk's electric cars

The Tesla Model Y is quick, packed with fun tech, and simple to charge. But not everyone will love its big touchscreen and lack of regular buttons.

www.businessinsider.com

How can a journalist covering EVs have never driven a Tesla in 10+ YEARS?? Are these the monkeys we are up against?

The founder, CEO and Editor-in-Chief of Business Insider has serious securities fraud allegations from 20 years ago for which he settled with the SEC. $4 million dollars of penalties paid (in 2002 dollars) and permanent ban from ever participating in the securities industry again. A few years later he started BI.

Henry Blodget - Wikipedia

en.wikipedia.org

This is the same publication that put out the anonymous friend-of-a-friend sexual assault allegation story earlier this year, for which there is still, months later, no additional evidence that has come to light.

I have not seen anything resembling journalistic integrity from this website.

 

[DarthPierce]

LR

Then it's not 4680 as far as we know.

 

[Singuy]

 

[Electroman]

My first time driving a Tesla showed me why people are so obsessed with Elon Musk's electric cars

The Tesla Model Y is quick, packed with fun tech, and simple to charge. But not everyone will love its big touchscreen and lack of regular buttons.

www.businessinsider.com

How can a journalist covering EVs have never driven a Tesla in 10+ YEARS?? Are these the monkeys we are up against?

Great point. I had the exact same question pop in mind when I read that sentence

 

[WADan]

Great point. I had the exact same question pop in mind when I read that sentence

And it tells me that I don’t need to waist time reading the article!