Elon: FSD Beta tweets

[Terminator857]

https://twitter.com/x/status/1439787397022199812

7 day safety evaluation starts after hitting the button. So soonest anyone gets it is Oct 3.

 

[Daniel in SD]

Of course they are. The question of whether they will be eventually successful is related but separate. I'm guessing that your statement stems from the whole California DMV email discussion. Tesla did not say they are not working on, or will never have, L4 i.e.unsupervised self-driving. They said that the feature being worked on for release is currently an L2 feature and is expected to remain so through release, i.e. in the foreseeable future.

As I and others have written several times before, that is an eminently sensible approach and does not legally, ethically or morally bind them to reporting incremental development data to California DMV.

The engineering knowledge gained during this development is perfectly open to be used for a subsequent L4 release. When and if this comes to fruition, and at the time that Tesla applies for CA approval to operate such a feature, they will be required to submit a performance data set that justifies the safety and performance of the driverless feature release.

Yeah, they test autonomous vehicles in other jurisdictions, just not California.

Tesla conducts testing to develop autonomous vehicles via simulation, in laboratories, on test tracks, and on public roads in various locations around the world.

 

[Bladerskb]

I too think that Waymo's self-driving technology is quite far along and seems impressive In many ways. I'm less sure that they're entirely barking up the right tree in their self-driving hardware+software+mapping architecture; this doubt comes not so much from a belief in specific and superior engineering concepts on my part, but from the apparently slowing pace of advancement - especially in light of their head-start on almost everyone else outside of academia.

As a close follower of Waymo, what is your opinion about the various upheavals in management and, to some degree, sales focus (the recently unsuccessful / aborted attempt to market their in-house Lidar IP)? The question is not an attack on Waymo, it's a very reasonable topic for anyone trying to assess their chances going forward.

What do you think of the slowing pace of advancement of Tesla's FSD? They have been working on it for over 6 years and with the release of the BETA last year we were faced with a software that couldn't even go a-couple minutes without a safety disengagement. A far cry from the marketing, pr and promises from Elon, Tesla and tesla fan base for the past 6 year that said that Tesla was 5-10 years ahead.

Are you less sure that they are barking up the right tree in their self-driving-low resolution cameras with blind spots, limited cleaning and no multi-modal sensor or compute redundancy?

 

[Bladerskb]

I can't wait to try it, but I'm waiting for it to exist in a place worth visiting.

Will you fly to SF at launch next year? I know i will. Tempted to fly to phoenix now and check out some sites though.

 

[Terminator857]

What do you think of the slowing pace of advancement of Tesla's FSD?

The last percents are more difficult. I wouldn't say pace of advancement is slowing.

They have been working on it for over 6 years and with the release of the BETA last year we were faced with a software that couldn't even go a-couple minutes without a safety disengagement.

Makes it safer as a level 2 system, forcing people to pay attention.

A far cry from the marketing, pr and promises from Elon, Tesla and tesla fan base for the past 6 year that said that Tesla was 5-10 years ahead.

Sounds ridiculous, and I didn't see those claims.

Are you less sure that they are barking up the right tree in their self-driving-low resolution cameras with blind spots, limited cleaning and no multi-modal sensor or compute redundancy?

Waiting for HW4. After HW4 arrives I'll be waiting for HW5.

 

[Cheburashka]

https://twitter.com/x/status/1439787397022199812

7 day safety evaluation starts after hitting the button. So soonest anyone gets it is Oct 3.

The barrage of FSD tweets from Musk over the past week or so actually makes me think it will finally happen. I wish we could get a better understanding of how the process will work though. Does everyone who is deemed a safe driver get the FSD Beta access, or is it assigned based on your YouTube subscriber count or something else?

 

[StealthLSU]

I haven't seen one yet, but has there been any indication if vision only cars will have the button or not? Or are we still stuck on a different branch for a while?

 

[Cheburashka]

I thought FSD Beta doesn't use anything but the cameras, so the availability of radar should not matter.

 

[StealthLSU]

That is my hope, but those of us in vision only cars for the most part are stuck on older software versions so I dind't know if that affected updates to FSD and the button.

 

[momo3605]

That is my hope, but those of us in vision only cars for the most part are stuck on older software versions so I dind't know if that affected updates to FSD and the button.

This is likely because vision code hasn’t been merged in to production branch yet. That’s probably why all the updates recently have been going to radar cars. But like @Cheburashka mentioned, FSD Beta is pure vision, and there is no reason vision-only cars wouldn’t get it.

 

[Doggydogworld]

Sounds ridiculous, and I didn't see those claims.

https://twitter.com/x/status/686279251293777920

Well, by that standard it is doubtful Tesla will ever achieve success. They aren’t even working on fully driverless cars.

Of course they're working on driverless. Musk refuses to report disengagements to CA so his lawyers prevaricate. That's their job.

Thanks, and a couple of follow-up questions. Perhaps this should move to the Waymo thread, but oh well:

1. Is there any indication, from conferences or interview answers etc., that Waymo has rethought any major part of the AV technical strategy or architecture over the last several years?

I've seen no major rethink. They continually improve sensors and move more functionality into NNs, like everyone else. But five years ago they were already pretty far along, doing drives without safety drivers and such. Five years ago Tesla's code base was a joke, just an Inception-based vision NN to follow lane lines and a crappy radar to (sometimes) avoid hitting the car in front. Oh, and sonar for side-collision avoidance. So it's not really fair to give Tesla much credit for a major rethink. They literally had no other choice but to toss their early crap and start fresh.

Also, hackers like Green show their "re-writes" aren't nearly as comprehensive as claimed.

1. Your explanation implies that that the last CEO, Krafcik, was too much of a marketer and not enough bound to the engineering reality

Krafcik was a large company exec, not an entrepreneur or a marketer. His was there to do deals with automakers. Both short term deals to buy cars for the Waymo One robotaxi service and long term deals to put the "Waymo Driver" in the OEM's passenger cars. He did get a couple of toothless MOUs signed. But their Robo business model doesn't work and neither he nor the current co-CEOs have a clue how to fix it. And OEMs were rightfully leery of "Waymo Inside". They saw what Intel and Microsoft did to the PC makers.

I don't think Krafcik got out ahead of his engineers. To the contrary, Dolgov called autonomy a solved problem years ago. Waymo just lacks a startup mentality.

 

[rdunniii]

Elon needs to drive his Model S from the Fremont factory to the the Nevada factory in the dark during the winter and post the trip. Using V9 certainly keeps you more alert and awake than just driving being prepared for the curves in the dark where it just goes beep, beep, beep take over immediately, I have just disengaged. Or the stop and go traffic west of Sacramento on 80 where it does just fine enough to lull you into amazement only to all of a sudden go beep, beep, beep take over immediately, I have just disengaged.

Will V10 be any better?

 

[tmoz]

https://twitter.com/x/status/686279251293777920

Of course they're working on driverless. Musk refuses to report disengagements to CA so his lawyers prevaricate. That's their job.


I've seen no major rethink. They continually improve sensors and move more functionality into NNs, like everyone else. But five years ago they were already pretty far along, doing drives without safety drivers and such. Five years ago Tesla's code base was a joke, just an Inception-based vision NN to follow lane lines and a crappy radar to (sometimes) avoid hitting the car in front. Oh, and sonar for side-collision avoidance. So it's not really fair to give Tesla much credit for a major rethink. They literally had no other choice but to toss their early crap and start fresh.

Also, hackers like Green show their "re-writes" aren't nearly as comprehensive as claimed.

Krafcik was a large company exec, not an entrepreneur or a marketer. His was there to do deals with automakers. Both short term deals to buy cars for the Waymo One robotaxi service and long term deals to put the "Waymo Driver" in the OEM's passenger cars. He did get a couple of toothless MOUs signed. But their Robo business model doesn't work and neither he nor the current co-CEOs have a clue how to fix it. And OEMs were rightfully leery of "Waymo Inside". They saw what Intel and Microsoft did to the PC makers.

I don't think Krafcik got out ahead of his engineers. To the contrary, Dolgov called autonomy a solved problem years ago. Waymo just lacks a startup mentality.

Yep, my June 2015 model was all set to come get me from a trip to NY, in case I only wanted to drive oneway, if I only were patient enough to wait 2 yrs.

 

[EVNow]

"In ~2 years,"

What we need to understand is that the Std Dev. for Muskestimates is very high

 

[EVNow]

To be fair, Waymo actually has a service that works right now. Anyone can use it.

To be fair 100s of thousands are using Tesla AP right now.

ps : Aren't there Waymo like services running in China ?

 

[powertoold]

Elon being 3-5 years late isn't unusual with big projects.

Model 3 ramp was late 2-3 years and never got to the 10k a week (in Fremont) Elon said in 2017

Falcon Heavy was planned to launch in 2013 but ended up launching in 2018

Roadster 2 is still delayed 2+ years now

Tesla Semi still delayed...

Holiday fire fire update...

List goes on if you look closely

At the end of the day, none of it matters because Elon pushes technology further than any other person or company right now. When he's on time, others are already 5+ years behind, just look at Starlink.

 

[nativewolf]

https://twitter.com/x/status/686279251293777920

Of course they're working on driverless. Musk refuses to report disengagements to CA so his lawyers prevaricate. That's their job.


I've seen no major rethink. They continually improve sensors and move more functionality into NNs, like everyone else. But five years ago they were already pretty far along, doing drives without safety drivers and such. Five years ago Tesla's code base was a joke, just an Inception-based vision NN to follow lane lines and a crappy radar to (sometimes) avoid hitting the car in front. Oh, and sonar for side-collision avoidance. So it's not really fair to give Tesla much credit for a major rethink. They literally had no other choice but to toss their early crap and start fresh.

Also, hackers like Green show their "re-writes" aren't nearly as comprehensive as claimed.

Krafcik was a large company exec, not an entrepreneur or a marketer. His was there to do deals with automakers. Both short term deals to buy cars for the Waymo One robotaxi service and long term deals to put the "Waymo Driver" in the OEM's passenger cars. He did get a couple of toothless MOUs signed. But their Robo business model doesn't work and neither he nor the current co-CEOs have a clue how to fix it. And OEMs were rightfully leery of "Waymo Inside". They saw what Intel and Microsoft did to the PC makers.

I don't think Krafcik got out ahead of his engineers. To the contrary, Dolgov called autonomy a solved problem years ago. Waymo just lacks a startup mentality.

Many truths there, it is moving so slow, very frustrating if you are someone like Krafcik and there is nothing to actually sell.

 

[nativewolf]

Elon being 3-5 years late isn't unusual with big projects.

Model 3 ramp was late 2-3 years and never got to the 10k a week (in Fremont) Elon said in 2017

Falcon Heavy was planned to launch in 2013 but ended up launching in 2018

Roadster 2 is still delayed 2+ years now

Tesla Semi still delayed...

Holiday fire fire update...

List goes on if you look closely

At the end of the day, none of it matters because Elon pushes technology further than any other person or company right now. When he's on time, others are already 5+ years behind, just look at Starlink.

Or look at EVs. He was 2 years behind Nissan and Chevy getting a mass market car out. They both flubbed because they did not understand what the market wanted and was willing to pay. I think AI and FSD will be the same, the program now has recognizable urgency that resembles other Tesla/SpaceX efforts, progress builds on progress and they'll have the first wide release of a system into generalized commercial fleet. Maybe to 400k units overnight. It will then start planning for L4 and Dojo will be doing the simulations with all the edge cases 400k drivers accumulate.

 

[S4WRXTTCS]

Will you fly to SF at launch next year? I know i will. Tempted to fly to phoenix now and check out some sites though.

Absolutely.

Especially since there is a flight out of Paine field which is one of my favorite little airports.

Edit: Maybe not at launch, but when I can be reasonably certain that I can try it out.

 

[Terminator857]

https://twitter.com/x/status/1440361722473156611

FSD beta visualizations coming to Europe in a month+.