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

OFFICIAL BUTTON WATCH

This site may earn commission on affiliate links.
From earnings press release:

Our team continues to focus on the V9 version of FSD City Streets beta, which will soon become more widely available in the United States. We believe that a vision-only system is ultimately all that is needed for full autonomy. Our AI-based software architecture has been increasingly reliant on cameras, to the point where radar is becoming unnecessary earlier than expected. As a result, our FSD team is fully focused on evolving to a vision-based autonomous system and we are nearly ready to switch the US market to Tesla Vision.
 
From Tesla.com/autopilot
A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead.
I wonder when they are going to remove that line, and what issues it will cause when they sold people cars with radar for 4.5 years and they disable it in a SW update and it can no longer see the car ahead.
 
Last edited:
From earnings press release:

Our team continues to focus on the V9 version of FSD City Streets beta, which will soon become more widely available in the United States. We believe that a vision-only system is ultimately all that is needed for full autonomy. Our AI-based software architecture has been increasingly reliant on cameras, to the point where radar is becoming unnecessary earlier than expected. As a result, our FSD team is fully focused on evolving to a vision-based autonomous system and we are nearly ready to switch the US market to Tesla Vision.

I'm sure the person who's been writing this gibberish for years now had a good chuckle this time around as well.
 
Feb 2019 earnings report was about Q4 2018 where they reported a profit. Lots of production in Q3 2018 and also a profit.
[QUOTE="DanCar, post: 5481089, memb...rt/9245/tesla-revenue-and-net-profit/[/QUOTE]
You are right - I was getting a little confused, because of the below.


That drop from ~ 350 to ~300 killed all my heavily leveraged options. Q4 ER itself was ok (until in the end he announced Deepak was (again) leaving Tesla), but the overhang from massive layoffs and other problems Tesla was internally facing as Model 3 demand in US apparently dried up and it wasn't clear they would be able to deliver cars to Europe meant a very difficult cash situation.
 
Blah blah blah, doesn't sound like it'll be available in May, let alone June lol. It seems there are still many unsolved challenges in video labeling.

Auto-labeling is a pie in the sky endeavor. Karpathy has talked about auto-labeling in 2019. Auto-labeling leads to bias in the data set, minor drifts in labeling will lead to poorer performance vs full and meticulous human labeling.
 
I think it was very telling during the call when someone asked a two prt question:

1) Status update on v9
2) Status update on FSD subscription

Followed by ~20 seconds of awkward silence while everyone (including Elon) waited for someone else to answer. I didn’t recognize the voice of the person who eventually responded, but he just gave a vague answer about the benefits of FSD recurring revenue.

And nothing about v9. If it was about to drop in the near future, Elon wouldn’t have been able to resist that question.
 
Tesla AI is definitely making the right moves, but I'm doubtful about the timing now. Software in general is notoriously difficult to estimate timing, and this is novel software ("hardest technical problems that exist, that has ever existed" lol per Elon), so barring a miracle, unlikely to see them widely release V9 by July 1.

I'd personally be pumped if they do a limited (to current FSD beta testers) V9 release by June.

As for auto-labeling, I guess it's possible to do geometry-based auto-labeling. Almost all of the geometries we encounter during driving are only subject to translation and rotation. The video labeler would label the initial and ending geometry / bounding box and then the software would predict the geometry change vs time. The labeler would adjust the bounding box if the prediction is off. This approach would still be very problematic because the prediction is likely based on a NN, which takes a long time to train, so you can't adjust the NN model on the fly.
 
Last edited:
From earnings press release:

Our team continues to focus on the V9 version of FSD City Streets beta, which will soon become more widely available in the United States. We believe that a vision-only system is ultimately all that is needed for full autonomy. Our AI-based software architecture has been increasingly reliant on cameras, to the point where radar is becoming unnecessary earlier than expected. As a result, our FSD team is fully focused on evolving to a vision-based autonomous system and we are nearly ready to switch the US market to Tesla Vision.

For years, they told us radar > lidar, lidar was a fool's errand, etc...but *now* even radar isn't necessary. Lol ok.
 
From earnings press release:

Our team continues to focus on the V9 version of FSD City Streets beta, which will soon become more widely available in the United States. We believe that a vision-only system is ultimately all that is needed for full autonomy. Our AI-based software architecture has been increasingly reliant on cameras, to the point where radar is becoming unnecessary earlier than expected. As a result, our FSD team is fully focused on evolving to a vision-based autonomous system and we are nearly ready to switch the US market to Tesla Vision.


All vision huh? Ooooofa.
 
Tesla’s will come with a professional driver when purchasing the full self driving subscription. For $1000/mo your own personal driver will use pure vision to drive you to any location, hands free. Read

I was thinking Elon will vertically integrate Neuralink into Telsa like he did with SolarCity. Then he'll put a monkey in the frunk that is neuralinked to the cameras and that will steer the car.
 
For years, they told us radar > lidar, lidar was a fool's errand, etc...but *now* even radar isn't necessary. Lol ok.
That doesn't sound right. I do remember many times him saying that lidar was trash, but he was comparing it to passive video (or video with headlights and marker lights on the other vehicles in low vision situations). I don't remember them saying anything about radar v. lidar.

I don't even remember them saying "Hey, our radar bounces under the car ahead of you and can see the car ahead of him" That saved my butt again yesterday when the car I was following too close swerved out of the lane we were in revealing a stopped car, but my car had already sounded the alarm before he swerved and I was consequently already slowing, so great feature of radar.

I just remember him saying that lidar is for bitches who can't make a 3D map of the environment w/ just video the way his team could. That lidar is expensive and requires you to make HD maps to match that are larger to store on board the car and to transfer over the cellular network and that if you haven't updated can break with construction or even weather events. And if you don't have them you can't use the feature.

Removing radar I think is because, as he said, what do you do when radar says one thing and the cameras say another? Who do you believe? And right now everyone is always complaining about the phantom braking, and the plowing into stopped objects on the road that radar is failing us on. Just give up, maybe save the already installed units for that bouncing under trick on the shorter cars that cannot see over. Save money on all the new cars that really will be better served by Teslavision.
 
For me, the most nonsensical thing Elon says about FSD relates to the business aspect of it. He says almost all cars will be autonomous soon, and that the leaders will take a "quarter":

When vast amounts of manufacturing are needed, as in robotaxis, this slows down rate of introduction, so maybe more like winner-takes-a-quarter. Still great.

He then goes on to say that FSD is the [most difficult technical challenge ever to exist]. So unless the winner shares (or others steal) their technology with other car makers, these two statements seem to contradict.
 
He did say that they intend to share with others when their big supercomputer for self driving training comes online. I imagine that would be sharing with other car manufactures, I don't think slurpee machine manufacturers will need that tech