Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

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

This site may earn commission on affiliate links.
Who can't love Warren for calling out that dips hit purveyor of FUD on E for Electric?

I've been following Tesla and the community for long enough now that I can see it coming. Usually it starts with some hurt feelings about someone disagreeing with you over something. Then they start to say things like "oh I love Tesla but I'm not a Tesla stan" or "I love the cars but hate the owners". Often that is listed right in their Twitter/YT profile. Then they begin every interaction with a negative or downright hostile attitude. When people snap back they say "See! Tesla stans (I can't roll my eyes enough at seeing fellow old people use that term) are horrible fanbois." After a while they slowly move to full on hater mode.
 
We have to step back and think about how the driver knows when to intervene.

The overwhelming majority of the time there is a visual clue that something isn't right, or the diver loses confidence that FSD knows what it is doing...

The vast majority of those visual errors, seem to be not recognising or correctly categorizing, something the car sees,,

These fall into 2 categories:-
  1. More training needed against a known label with a predefined course of action.
  2. New label needed and perhaps additional driving instructions..
There is no obvious indication Tesla is running out of headroom in terms of training with more data, and no obvious indication they are running out of capacity to add more labels.

In terms of the part of FSD that can be solved with vision alone, there is plenty of scope for further improvement, some of that being rapid improvement, some taking months rather than weeks.

The march of 9s is never 100%, any driving situation that requires sensors or information beyond what vision can provide, has a element of risk.
Vector maps, auditory cues and situational memory in training data can help.

But when I look at the videos, most of the time the driver intervened in response to a visual cue, that sort of problem is a NN training problem.
It seems like Tesla is going down the path of avoiding categorized objects. So it did not avoid the stopped fire truck with flashing lights because it had not categorized it. There’s also the recent examples of the concrete pillars in between lanes in Seattle. There’s also the case of Tesla gathering lots of data on wheels/tires in the road at various states of occlusion.

I would hope there is a way to just detect broadly that something is an object, or that the surface in the path is undriveable and that object categorization is only important insofar as predicting movement, or judging how bad it would be to hit. If it could have that capability, then at least it would be safe. E.g. It might be annoying that it brakes hard for a balloon in the road, because it is non-categorized, but again at least it would be safe.

James Douma stated that if you show enough pictures to the neural net, at some point, it just gets it. I wonder if one could just show enough frames of various objects in the road, or screwed up driveable paths and then the trained net would just get that there’s something there even if it’s a UFO;-) Otherwise, the problem seems nearly intractable.
 
But the dealer-network is supposed to be a boon for the conventional manufacturers! Having skilled people on the ground that take care of customers!

Not that i am surprised ;)
I see your point about the repair fail. I do think that for initial EV purchasers, being able to see an EV at a local dealer is going to be a positive for first time buyers particularly those on the fence. Whether it is enough to overcome the strong positive buzz among Tesla early adopters is hard to know. I think there is strong brand loyalty among many car buyers in the US and they may not go to Tesla until they have a fail with their regular dealer.

I used to just talk a lot about my Tesla but now I quickly try to get people to drive it. They will discuss the experience with their non-Tesla friends and they will inevitably (having real first hand knowledge) end up dismissing some false preconceptions and thereby "maybe" pry loose from conventional thinking enough to see Tesla and EVs more clearly.
 
I’m hoping for a cybertruck update during Q2 earnings call
Lack of update has made me move my optimistic delivery estimate of "some" being delivered late this year to more like late Q1 2022. The factory is moving along, but Y is the priority. Plus the CT will need entirely new production lines, crash testing etc.
 
I don't know if you guys saw the video where Gali's FSD beta seemingly went toward the pillars, but this work shows the underlying depth maps clearly detecting the pillars. This is bullish to me as again, camera based perception is the main challenge Tesla has to overcome vs. Lidar -sensored competitors.

Most likely a classification issue. I'd bet the next build has this fixed. Simple to fix as the annotation to classify this as a column and to treat it like a wall.

As I'm in Seattle, these columns are unique to be in the middle of two lanes going the same way. If I were Gali or anyone else, I'd drive this several times as I'm sure Tesla would be interested in more training data.
 
Lack of update has made me move my optimistic delivery estimate of "some" being delivered late this year to more like late Q1 2022. The factory is moving along, but Y is the priority. Plus the CT will need entirely new production lines, crash testing etc.
I agree. I ended up ordering a Model Y for our vacation home (hoping for August delivery after ordering early June), because I don't think CT is coming until next summer at the earliest. CT is going to be like the new Roadster - they can't make them because of battery shortage. I also think 600 mile try-motor will be delayed. Luckily I have two early reservations, so I will get the dual motor, and then upgrade a year or two later when the tri is finally released.
 
Interesting intraday price chart comparison between TSLA and NIO, meaning to me that the movement today has nothing to do with any company news.

002AF22A-6447-4AF4-9A94-18FEF45797B4.jpeg
 
Lack of update has made me move my optimistic delivery estimate of "some" being delivered late this year to more like late Q1 2022. The factory is moving along, but Y is the priority. Plus the CT will need entirely new production lines, crash testing etc.
I'm just hoping to have my Cybertruck by next summer. I'm already shopping for accessories like a travel trailer and 2 electric KTM dirt bikes - one for me and one for my son, to put in the back.
 

My brokerage's newsfeed - half hour ago:

Tesla's FSD Beta 9 Is Here And Testers Are Mostly Impressed

3:06 pm ET July 13, 2021 (Benzinga)

Tesla Inc (NASDAQ: TSLA) is working towards fully autonomous cars. The company says every car it sells today comes with the hardware necessary for full self-driving; the software just needs to catch up to the task.

Tesla says an optional software package called "Full Self Driving" for $10,000 that allows the vehicles to make more advanced moves on the road.

Tesla has rolled its FSDBeta program out to a small set of testers who have been sharing their thoughts and experiences across social media. The software has recently been updated to FSD Beta Version 9, which removes the use of radar and relies solely on the car's eight exterior cameras.

Tesla updated the visuals on the in-car display, and people are enjoying the new look. The car can display people, cars, trucks, bikes and even dogs.
Testers are describing the experience as smoother and more confident, noting that the cars, which still aren't fully autonomous, have become better at decision making in many instances.

On YouTube, videos are showing the V9 update in action. A video from YouTube channel Dirty Tesla shows the software successfully navigate a dirt road in the dark of night, successfully completing a left turn.

Another video from James Locke shows a drive from Santa Clarita to Northridge California. The car is able to handle most of the drive on its own, with the driver present monitoring the car and correcting few mistakes.

"Overall FSD Beta v9 did VERY well. It wasn't a perfect drive but we both were super impressed with the behavior and performance of the current FSD beta," James wrote in the description of the video.

But Tesla's FSD Beta V9 is not great for everyone. A video by channel HyperChange shows the Tesla trying to drive into a large concrete pillar around the 17-minute mark in the video. But the driver is doing their job, and stops the car from making the incorrect move.

Benzinga's Take: Overall, Tesla's Version 9 update of FSD Beta seems to be impressing drivers, while showing it's still not quite ready for wide release. Elon Musk said on Twitter that if this release goes well, a wider release for the general public could commence in about a month or so.
An announcement around July 26th?
 
Last edited:
  • Like
Reactions: madodel