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

Tesla posts FSD demo video of Model 3

This site may earn commission on affiliate links.
There is literally no indication that they are "in use" by the NN, it's just a video frame from each. :rolleyes:


move goalposts much? :)

Your original post was:

T
It has the cameras but does it actually use them yet? Or are they like the internal driver-facing camera, there but not doing anything."

The interior camera isn't recording regularly- thus not doing anything- you asked if that's true of the others.

It's not. They're all taking video, unlike the interior camera. And you were shown video evidence of that fact.

But on top of that it's pretty obvious that side video footage is "doing something" since the NN is using it, right now, for things like displaying nearby cars, automatic lane changes and blind spot detection.
 
By the way, anyone find it interesting that Tesla used a Model 3 this time for their FSD demo instead of the Model S like they did in the 2016 FSD video? Also, Tesla always uses the Model 3 when they talk about the Tesla Network. It seems like the Model 3 is Tesla's go-to car now for FSD.
 
But on top of that it's pretty obvious that side video footage is "doing something" since the NN is using it, right now, for things like displaying nearby cars, automatic lane changes and blind spot detection.

Do you have any actual evidence of that? It's what I asked for but you didn't appear to comprehend.
 
By the way, anyone find it interesting that Tesla used a Model 3 this time for their FSD demo instead of the Model S like they did in the 2016 FSD video? Also, Tesla always uses the Model 3 when they talk about the Tesla Network. It seems like the Model 3 is Tesla's go-to car now for FSD.

No one who can afford a Model S/X is going to use it as a robotaxi, that's why.

Also the Model 3 has an internal camera, which will probably be essential for taxi service. Otherwise there is no way to monitor what passengers are doing.
 
Do you have any actual evidence of that? It's what I asked for but you didn't appear to comprehend.


You mean besides the multiple specific, working-right-now, features I listed that are only possible because it's using the side cameras that you didn't appear to read or comprehend?

Sure, here's even more evidence you're wrong-

Here’s what Tesla Autopilot can see with the new v9 neural net

One of the main new capabilities is the fact that the neural net is camera agnostic, which enabled Tesla to deploy it on all cameras around the car.

Notice how the NN is boxing and tracking various objects in all the cameras?
 
Notice how the NN is boxing and tracking various objects in all the cameras?

Great, you could have just posted that the first time. So what does the NN actually do with that information? Where is the 3D environment map and overlay showing that the car is predicting vehicles coming from the side will cross its path?

You know like Waymo demonstrated back in 2016.

That's what makes this Tesla demo even more suspicious. If they had this data why did they not show it? Prove that it's not smoke and mirrors.
 
  • Funny
Reactions: tracksyde
Great, you could have just posted that the first time.

The guy you were replying to to begin with did post feeds from all the cameras the first time. You still insisted maybe they weren't actually running like the interior camera isn't running.... which I pointed out was impossible since he just showed you the cameras running

Then you changed your story to how can we be sure it's USING all that video (WTF would it bother wasting power capturing it if it were not BTW?)- so I pointed out multiple features of the car that are only possible if it's using that data.

You didn't like that, so I showed you actual data from the computers showing it using those camera feeds to ID and track objects (which, again, we already knew by the fact it has currently-working features that are only possible if it's using that data).

Now you're moving the goalposts yet again to wanting details on what it's doing with the data you didn't even believe it was using, after first not believing it was even capturing.... (and again SOME of the things it's doing with the data is obvious... blind spot detection, automatic lane changes, displaying nearby cars including by general vehicle type, etc)



So what does the NN actually do with that information? Where is the 3D environment map and overlay showing that the car is predicting vehicles coming from the side will cross its path?

You know like Waymo demonstrated back in 2016.

That's what makes this Tesla demo even more suspicious. If they had this data why did they not show it? Prove that it's not smoke and mirrors.


They already showed a 3D environment map at the recent presentation you apparently didn't bother to watch.

They displayed a 3D environment map based on a 6 second video capture from vehicle cameras to explain why LIDAR was so stupid a technology- since cameras could do detailed depth maps without it.

They also showed various clips of the system making predictions about pathing, objects, etc including beyond the actual range of the cameras based on the data it had in the moment.

Maybe go watch all the presentations from Monday before deciding where you want to move the goal posts next time?


Also, this has been out there for a while-
A rare look at what Tesla Autopilot can see and interpret

Radar data overlayed with camera/labeling data, a detailed list of objects autopilot sees and tracks. It reports things like the dimensions, relative velocity, confidence that the actual object exists, the probability that the object would be an obstacle and so on.

And remember this is just what hackers have been able to see- obviously folks with real developer access can see a lot more.
 
  • Like
Reactions: tracksyde
Notice how they filmed this on a calm low traffic freeway and roads, there is no way I would use this in real life.....not yet anyway. My AP2 is still reallly iffy at best.
Electrek had an article showing before Easter. Bet that was practice and the actual taping was on Easter for the low traffic. Keep in mind the chip has only been available a few months and the software is VERY pre alpha. It is AMAZING that it is already this competent with so little time, debugging and reprograming. This is VERY promising at this extremely early point.
 
  • Informative
Reactions: diplomat33
Notice how they filmed this on a calm low traffic freeway and roads, there is no way I would use this in real life.....not yet anyway. My AP2 is still reallly iffy at best.

I find the video and test to be encouraging, but so far, "hands off the wheel" is the only feature that would make a difference to me in real-life driving. And I'd pay the $5,000 for that if it became available. But all the lane changing and freeway-to-freeway changing involve such a small part of my driving, even when I was back on the mainland, that they confer small advantage. On my drive from Spokane to Seattle, I used EAP nearly the whole way, and once in Seattle I took over for freeway changes, and that was no big deal.

Hands-off-the-wheel on highways would be a big deal, and NAVonAP on city streets would be a bigger deal. But that's when it can handle real-life driving. I think that's several years away at least.
 
Great, you could have just posted that the first time. So what does the NN actually do with that information? Where is the 3D environment map and overlay showing that the car is predicting vehicles coming from the side will cross its path?

You know like Waymo demonstrated back in 2016.

That's what makes this Tesla demo even more suspicious. If they had this data why did they not show it? Prove that it's not smoke and mirrors.

Man you are a piece of work. At least I’ve got you calibrated now, thought you may have had somewhat valid opinions in the past but no more.
 
Great, you could have just posted that the first time. So what does the NN actually do with that information? Where is the 3D environment map and overlay showing that the car is predicting vehicles coming from the side will cross its path?

You know like Waymo demonstrated back in 2016.

That's what makes this Tesla demo even more suspicious. If they had this data why did they not show it? Prove that it's not smoke and mirrors.

Tesla does not need to prove anything to anyone. If you don't believe them, don't. That's fine. Now if you had a billion dollars you wanted to invest in Tesla you could get a personal meeting with Elon and I'll bet he'd show you data. But these demonstrations are for the rest of us who understand their limitations: They are best-case scenarios demonstrating a system far from FSD but years ahead of the competition.

The car in the video cannot make that drive without a human driver to take over. But it's still an impressive accomplishment.
 
Tesla does not need to prove anything to anyone.

They do though, which is why they did this. They are trying to keep the cash flow going and keep investors happy after all the delays. Teasing an additional source of revenue for next year is designed to attract money.
 
They do though, which is why they did this. They are trying to keep the cash flow going and keep investors happy after all the delays. Teasing an additional source of revenue for next year is designed to attract money.

A teaser demonstration is not proof. It's an advertisement or a publicity stunt. As I said, I find it useful because for one, I don't believe they cheated and I don't need them to prove they didn't cheat, and for another, I understand what the demonstration was not: It was not a performance that can be repeated on any arbitrary section of road.

There will be people who will say that Tesla "cheated" by doing this or that. Those people don't matter to Tesla. Tesla does not need to prove to them that it didn't. The rest of us who understand the demonstration in context don't need proof.

Now, given that Tesla has never met a deadline in its entire history, I won't believe anything they say about when, but given what my car can do today, their word carries a lot more weight when they talk about what.
 
A teaser demonstration is not proof. It's an advertisement or a publicity stunt. As I said, I find it useful because for one, I don't believe they cheated and I don't need them to prove they didn't cheat, and for another, I understand what the demonstration was not: It was not a performance that can be repeated on any arbitrary section of road.

There will be people who will say that Tesla "cheated" by doing this or that. Those people don't matter to Tesla. Tesla does not need to prove to them that it didn't. The rest of us who understand the demonstration in context don't need proof.

Now, given that Tesla has never met a deadline in its entire history, I won't believe anything they say about when, but given what my car can do today, their word carries a lot more weight when they talk about what.

my car cant summon straight into the my garage.
 
I think it will take a while until any autonomous vehicle can interpret something like this correctly. :p


xd4KtXQ.jpg

Never hold up in court - there's no mention of Daylight Saving Time changes, drivers not informed of that particular school's days.
Then again maybe the sign is meant for a person by the name of 'School Days'.