How will Tesla demo FSD?

[Electroman]

Elon says that Teslas will be capable of "hands free" driving in six months so I guess we'll know soon how right he is about that...

And if it happens in say 6 months 1 day, you get to say 'see I told ya. He was lying'.

Me thinks we are about 2 to 3 years away from any drive that does not require human supervision in any stretch of the road, highway or not. And thats all right. And that only means Elon is very bad at predicting time lines, and not achieving his stated goals.

 

[Daniel in SD]

And if it happens in say 6 months 1 day, you get to say 'see I told ya. He was lying'.

Me thinks we are about 2 to 3 years away from any drive that does not require human supervision in any stretch of the road, highway or not. And thats all right. And that only means Elon is very bad at predicting time lines, and not achieving his stated goals.

I don’t really have a problem with overly optimistic schedules. I tend to make them myself and I think it help motivate me to finish projects faster.
There is a point where it becomes delusion and I think Elon is at that point when it comes to FSD. If they were really close to hands free there would be some evidence.

 

[diplomat33]

If they were really close to hands free there would be some evidence.

Well, we might just get some evidence in just 2 days, on April 22.

 

[CarlK]

The author who I have proven has no clue so he misinterprets it to mean Mobileye is AGAINST deep learning. Because the author thinks only AlphaGo = deep learning.

IEEE is THE top engineering organization in the world that, among other things, organizes conferences which most important new discoveries in the field are published and the primier publisher of peer reviewed papers. I've been a member since i was in graduate school. The author was an accomplished technical journalist not just your run of the mill blogger either. I'd take their words much more serious than your usual gibberish about MobileEye.

 

[CarlK]

I don't know if this Lex Fridman state of self driving technolgy art video had been posted here. Nothing there contradicts what Tesla has said. Unlike what those detractors always want to argue he seemed to give a pretty good blessing to the Tesla camera approach too (starts at 41:50).

Also interesting is what he said neural network running inside Tesla cars (50:05 having a tons of aweson nueral running inside the car). Again exactly what those few detractors insist that Tesla does not do.

 

[Daniel in SD]

Well, we might just get some evidence in just 2 days, on April 22.

I don't think there is anything they could show that I would take as evidence that they are close to releasing a level 3-5 system. And I suspect that you would take a demo equivalent to the Google demos circa 2010 as evidence that they are. So I guess it's really a matter of faith.
Now, if they present data from their employee testing program, their methodology for determining how safe their system is relative to a human and how close they are to achieving that goal, I would be impressed.

I don't know if this Lex Fridman state of self driving technolgy art video had been posted here. Nothing said there contradicts what Tesla has said. Unlike what those detractors always want to argue he seemed to give a pretty good blessing to the Tesla camera approach too (starts at 41:50).

There is nothing wrong with a camera only approach. Everyone is using cameras and I'm sure they'll work to remove lidar once they get their systems working. Lidar does provide an extra margin of safety for when the vision system fails to detect objects so it would seem silly to remove it from test vehicles until you've determined that you can get the same level of safety without it.

 

[Bladerskb]

Also interesting is what he said neural network running inside Tesla cars (50:05 having a tons of aweson nueral running inside the car). Again exactly what those few detractors insist that Tesla does not do.

Ofcourse NN models are running inside the car, that's called inference, just like any EyeQ4 cars.
What people are saying is that they are not trained in the car.
You just showed that you really have no clue what you are talking about.

 

[CarlK]

I don't think there is anything they could show that I would take as evidence that they are close to releasing a level 3-5 system. And I suspect that you would take a demo equivalent to the Google demos circa 2010 as evidence that they are. So I guess it's really a matter of faith.
Now, if they present data from their employee testing program, their methodology for determining how safe their system is relative to a human and how close they are to achieving that goal, I would be impressed.

There is nothing wrong with a camera only approach. Everyone is using cameras and I'm sure they'll work to remove lidar once they get their systems working. Lidar does provide an extra margin of safety for when the vision system fails to detect objects so it would seem silly to remove it from test vehicles until you've determined that you can get the same level of safety without it.

The only problem is you can't just remove Lidar from software developed with it. Machine needs to relearn the hard way when switched to camera only. At least that's what Elon said and why he called using Lidar is like on crutches.

Ofcourse NN models are running inside the car, that's called inference, just like any EyeQ4 cars.
What people are saying is that they are not trained in the car.
You just showed that you really have no clue what you are talking about.

Did I just mention your usual Mobileye gibberish and you are now putting NN and EyeQ4 in the same sentence? How pathatic!

 

[diplomat33]

I don't think there is anything they could show that I would take as evidence that they are close to releasing a level 3-5 system.

Really? Not even test drives if the car actually does self-drive around city streets with no driver disengagements at all?

Now, if they present data from their employee testing program, their methodology for determining how safe their system is relative to a human and how close they are to achieving that goal, I would be impressed.

I am a data and numbers guy. I would love to see some data too. Hopefully, Tesla does provide some data on April 22. Definitely data like number of miles between disengagements is very useful for showing how reliable the self-driving system actually is. Sounds like you are more of a data driven guy. You don't really care about seeing the car in action, you care about numbers that prove reliability.

I like data but I am also a visual guy. I like some eye candy. I know that quick test drives or demos don't prove a system is L4 but it's still nice to see the system in action.

 

[Daniel in SD]

The only problem is you can't just remove Lidar from software developed with it. Machine needs to relearn the hard way when switched to camera only. At least that's what Elon said and why he called using Lidar is like on crutches.

Crutches is actually a great analogy. They're great for when your leg is not functional yet. Camera only vision systems still make more mistakes than systems where you combine them with lidar and radar. I would argue that is not nearly as difficult as you think to transition to a camera based system since you can run both systems in parallel to validate the accuracy of the camera only system.
When I watch the Cruise video of driving in San Francisco I question whether the vision system is really what's limiting autonomous vehicles right now.

 

[Daniel in SD]

Really? Not even test drives if the car actually does self-drive around city streets with no driver disengagements at all?

This was done nearly a decade ago and there are still no autonomous vehicles. I want to hear more about why Tesla thinks they will succeed where everyone else has failed.

I am a data and numbers guy. I would love to see some data too. Hopefully, Tesla does provide some data on April 22. Definitely data like number of miles between disengagements is very useful for showing how reliable the self-driving system actually is. Sounds like you are more of a data driven guy. You don't really care about seeing the car in action, you care about numbers that prove reliability.

I like data but I am also a visual guy. I like some eye candy. I know that quick test drives or demos don't prove a system is L4 but it's still nice to see the system in action.

Hearing that Waymo is 11,000 miles between disengagements I thought they were getting close but then I watched the video of the journalist following them and I realized that that number is very misleading. It appears that they are not counting the planned disengagement they do pulling out of their own parking lot!
Seeing self driving demos was cool a decade ago but it should have become abundantly clear to everyone that being "feature complete" means you're just getting started.

 

[diplomat33]

This was done nearly a decade ago and there are still no autonomous vehicles. I want to hear more about why Tesla thinks they will succeed where everyone else has failed.

Hearing that Waymo is 11,000 miles between disengagements I thought they were getting close but then I watched the video of the journalist following them and I realized that that number is very misleading. It appears that they are not counting the planned disengagement they do pulling out of their own parking lot!
Seeing self driving demos was cool a decade ago but it should have become abundantly clear to everyone that being "feature complete" means you're just getting started.

Thanks. Your reply really helps me get where you are coming from. Basically, we don't have L4 autonomy yet despite many nice looking demos and even some seemingly good data so you want to see more to actually convince you that FSD is really happening. That makes sense.

 

[diplomat33]

@Daniel in SD Tesla is very secretive about their FSD work. They don't release any disengagement report so we have no data there. We don't get any public glimpses of Tesla cars testing FSD. All we have are software updates which can show us what Tesla is ready to release, Tesla's word and of course Musk's infamous tweets and interviews. So I am totally with you that demos and such are not enough to prove L4, but at the same, since Tesla is so secretive, I think it is even more important for them to show something so that the public can get a meaningful update on their FSD work. That's probably a big reason why Tesla is doing the event on Monday. So while test drives and videos are not enough, they are still needed. Companies like Tesla have to show their work at some point and test drives and videos are a visual way of doing that.

 

[Daniel in SD]

@Daniel in SD Tesla is very secretive about their FSD work. They don't release any disengagement report so we have no data there. We don't get any public glimpses of Tesla cars testing FSD. All we have are software updates which can show us what Tesla is ready to release, Tesla's word and of course Musk's infamous tweets and interviews. So I am totally with you that demos and such are not enough to prove L4, but at the same, since Tesla is so secretive, I think it is even more important for them to show something so that the public can get a meaningful update on their FSD work. That's probably a big reason why Tesla is doing the event on Monday. So while test drives and videos are not enough, they are still needed. Companies like Tesla have to show their work at some point and test drives and videos are a visual way of doing that.

I find it odd that they don't test on public roads in California and no one has been able to figure out where they are testing. I don't see any reason for the secrecy, it's not like you can learn Waymo's and Cruise's trade secrets by watching their cars drive around.

 

[CarlK]

Crutches is actually a great analogy. They're great for when your leg is not functional yet. Camera only vision systems still make more mistakes than systems where you combine them with lidar and radar. I would argue that is not nearly as difficult as you think to transition to a camera based system since you can run both systems in parallel to validate the accuracy of the camera only system.
When I watch the Cruise video of driving in San Francisco I question whether the vision system is really what's limiting autonomous vehicles right now.

I have also learned something from the Lex Fridman video linked above. It confirmed what we thought that camera is more challenging to program but it made up for that by able to provide a lot more machine learning experiences. I can't think of a way to avoid redo the learning from Lidar-centric system to camera-centry system becasue more complex NN is needed for the camera system. You have to learn how to walk without crutches.

@Daniel in SD Tesla is very secretive about their FSD work. They don't release any disengagement report so we have no data there. We don't get any public glimpses of Tesla cars testing FSD. All we have are software updates which can show us what Tesla is ready to release, Tesla's word and of course Musk's infamous tweets and interviews. So I am totally with you that demos and such are not enough to prove L4, but at the same, since Tesla is so secretive, I think it is even more important for them to show something so that the public can get a meaningful update on their FSD work. That's probably a big reason why Tesla is doing the event on Monday. So while test drives and videos are not enough, they are still needed. Companies like Tesla have to show their work at some point and test drives and videos are a visual way of doing that.

I find it odd that they don't test on public roads in California and no one has been able to figure out where they are testing. I don't see any reason for the secrecy, it's not like you can learn Waymo's and Cruise's trade secrets by watching their cars drive around.

Tesla is not deliberately trying to be scretive. They just don't develop programs with test cars like Waymo or Cruise does. It still needs to do final verifications but it looks employee owners are doing that part with their own cars. Pretty smart if you ask me. Elon himself probably is the first one to test any new features. That should keep the AP team honest.

I don't think Tesla has an unfair advantage because it does not report disengagement data. It eventually still has to put cars to test in customers' hands that no one else is doing at this moment. Besides only California requires disengagement report and that could easily be gamed. Not saying they all do this but companies could develop the program elsewhere and in California only run test cars in same familar and repeatedly tested routes. They could get a very nice but not representative data that way if they want to. Tesla could do that too if it wanted to cheat.

 

[diplomat33]

Tesla is not deliberately trying to be secretive. They just don't develop programs with test cars like Waymo or Cruise does. It still needs to do final verifications but it looks employee owners are doing that part with their own cars. Pretty smart if you ask me. Elon himself probably is the first one to test any new features. That should keep the AP team honest.

I get that. But the lack of disengagement reports or any meaningful public testing certainly affects public perception. It gives ammo to the critics who say it is evidence that Tesla has nothing. And it's not just public perception. Without any public data, we literally don't know. It's a big reason why a lot of the discussions on this forum are so speculative about the state of Tesla's FSD. Hopefully April 22 will answer some questions.

 

[CarlK]

I get that. But the lack of disengagement reports or any meaningful public testing certainly affects public perception. It gives ammo to the critics who say it is evidence that Tesla has nothing. And it's not just public perception. Without any public data, we literally don't know. It's a big reason why a lot of the discussions on this forum are so speculative about the state of Tesla's FSD. Hopefully April 22 will answer some questions.

I'm not too worried about it. If there are disengagement data those people will still find something else to discredit Tesla. I trust Elon who although often optimistic but is always honest in what he says. Like I said if he wants to game the CA reporting system he's more than smart enough to do it. There are too much noises out there most people just don't have the ability to pick out real stuff amidst them. Not saying I wasn't confident before but that video from Lex Fridman, someone who is as knowledgeable and objective in this subject as any, did give me quite a warm and fuzzy feeling. Almost want to shout I told you so.

April 22 will answer a lot of technical questions for people like you and I. As to whether they could do a good PR job to help suppress some of those noises remains to be seen. I hope they will.

 

[diplomat33]

Looks like Tesla has filed a restraining order against individuals who tried to sabotage the FSD demo:
Nafnlaus on Twitter

Not cool to try to sabotage Tesla!

 

[electronblue]

Looks like Tesla has filed a restraining order against individuals who tried to sabotage the FSD demo:
Nafnlaus on Twitter

Not cool to try to sabotage Tesla!

Look on the bright side, Autopilot is approaching Waymo levels of outsider harrasment.

 

[diplomat33]

Look on the bright side, Autopilot is approaching Waymo levels of outsider harrasment.

True. Maybe that's a good sign for Tesla's FSD effort.