FSD rewrite will go out on Oct 20 to limited beta

[Matias]

What I find most interesting is that prior to fsd beta, many of us were saying the sensor suite is impossible or pointing out some hardware limitation, but now with V9, we're not talking about hardware inadequacy anymore. Now we're pointing out disengagements or dumb decisions that have nothing to do with the sensors but rather the software or perception not seeing or acting on something that is obviously there.

As I wrote here

Post in thread 'FSD Beta Videos (and questions for FSD Beta drivers)'
FSD Beta Videos (and questions for FSD Beta drivers)

and here

Post in thread 'FSD Beta Videos (and questions for FSD Beta drivers)'
FSD Beta Videos (and questions for FSD Beta drivers)

I’m not completely sure, that the hardware (specifically forward looking side cameras) is enough for unprotected left turns with fast incoming traffic.

 

[powertoold]

As I wrote here

Post in thread 'FSD Beta Videos (and questions for FSD Beta drivers)'
FSD Beta Videos (and questions for FSD Beta drivers)

and here

Post in thread 'FSD Beta Videos (and questions for FSD Beta drivers)'
FSD Beta Videos (and questions for FSD Beta drivers)

I’m not completely sure, that the hardware (specifically forward looking side cameras) is enough for unprotected left turns with fast incoming traffic.

Daniel in SD responded to you about this. It's definitely a NN perception issue rather than a sensor suite issue... It's very obvious to me that one or more of the front cameras would see the oncoming cars, especially since Chuck's wide angle front facing camera can see it at low YouTube resolutions.

 

[Knightshade]

The problem is that we have almost no data because Tesla has not released any real data.

Yes.

But making up #s, for either side of the argument, helps nobody.

 

[Matias]

Daniel in SD responded to you about this. It's definitely a NN perception issue rather than a sensor suite issue... It's very obvious to me that one or more of the front cameras would see the oncoming cars, especially since Chuck's wide angle front facing camera can see it at low YouTube resolutions.

Wide forward camera‘s distance is only 60 meters, I think that the oncoming car is further away than that. Main forward camera‘s angle is much more narrow and I think that when you look at Chuck’s car position from the drone image, the oncoming car is not in its visual field.

Edit: From the drone image, the oncoming car is in the 22 degrees angle to Chuck‘s car. Maybe I was wrong. What is the width of the main camera‘s visual field?

edit2: If the picture is Tesla’s page is accurate, the width of the main camera is approximately 23 degrees both sides, so the car might be in the visual field.

Edit3: the lane closest to Chuck’s car is outside camera’s view.

 

[Bladerskb]

You think a disengagement per mile number over 3x lower is "virtually the same"

Glad you're not in charge of anything related to safety

Yes the difference of 3 miles is virtually the same.

you're still dealing with 0.0000001. Its not safer, it just as unsafe as 1 mile per.

I mean, you're the one who appears to be posting made up figures, rather than actual facts/data, so....

GTFO. I have posted more facts and data than you have quality posts.
I have posted the link to Tesla CA report multiple times.

Weird, because every regulatory body that looks at this stuff counts them.

City driving is orders of magnitude more difficult than highway driving. This is why all SDC company make the distinction.
IF you want to talk about how good tesla is at highway driving, then be my guess look at Navigate on Autopilot.
But if you want to look at how good tesla is at city driving then look at how FSD Beta aka 'automatic driving on city streets' does on actual city streets.
But if you want to argue for the sake of arguing then have at it.

People who got rides were in an L2 vehicle- that's why none of those had to be reported to CA DMV.

That said- "someone talked out" isn't data, it's an anecdote.

You can compile most of the rides given that day and come up with statistical results. That's data. Its not the best data but it is.

The plural of anecdote is also not data.

If you want people to focus on facts instead of speculation I'd suggest you post more facts and less speculation.

I'm sorry but videos of FSD Beta are facts and data as they can be documented. The same way what JJRick has done is considered as DATA and completely admissible in court. You are misusing the term anecdote(hearsay). IF we used the word anecdote the way you are using it. WE would disqualify most of the facts and data that exist.

Waymo Rides

JJRicks Rider,License Plate,Date,Time Summoned,Start Time,End Time,Ride Duration,Distance (mi),Price,Safety Driver/ Fully Driverless,Notes or Remarks (Other),Other unique FD license plates,Wait time JJRicks,N/A,06/20/19,14:26,14:35,14:49,0:14:00,5.38,$0.00,SD,Free rides from new rider promotions

docs.google.com

JJRicks.com - Videos

Find the full playlist of videos here or at jrj.pw/waymo for quick access.

www.jjricks.com

 

[Bladerskb]

Yes.

But making up #s, for either side of the argument, helps nobody.

I didn't make up any numbers, these numbers are not exact but are 100% accurate.

 

[Knightshade]

I didn't make up any numbers, these numbers are not exact but are 100% accurate.

.... right.


Tesla sold 100,000 million cars last quarter and went 27 billion miles without interventions.

Those numbers might not be exact but are 100% accurate- because like you I don't understand how those claims directly contradict each other and I think watching a few youtube videos allows me to make up intervention rates.

Yes the difference of 3 miles is virtually the same.

you're still dealing with 0.0000001. Its not safer, it just as unsafe as 1 mile per.

It's not "3 miles", it's 3.27x less often than your claim. Or over 326% LESS OFTEN needing intervention than your claim.

You seem... really not great with math or statistics.

GTFO. I have posted more facts and data than you have quality posts.

This appears to be another non-fact-based claim of yours

 

[Bladerskb]

.... right.


Tesla sold 100,000 million cars last quarter and went 27 billion miles without interventions.

Those numbers might not be exact but are 100% accurate- because like you I don't understand how those claims directly contradict each other and I think watching a few youtube videos allows me to make up intervention rates.



It's not "3 miles", it's 3.27x less often than your claim. Or over 326% LESS OFTEN needing intervention than your claim.

You seem... really not great with math or statistics.




This appears to be another non-fact-based claim of yours

Don't get made at me cause Tesla can't crack 10 miles on avg between safety disengagement.
Get mad at Tesla.

 

[Knightshade]

Don't get made at me cause Tesla can't crack 10 miles on avg between safety disengagement.
Get mad at Tesla.

Tesla, to my knowledge, has not made any claims about miles per disengagement that turned out to be factually untrue by a factor of over 300%.

You have.

Might wanna see to that

 

[Bladerskb]

Tesla, to my knowledge, has not made any claims about miles per disengagement that turned out to be factually untrue by a factor of over 300%.

You have.

Might wanna see to that

1 mile vs 3 miles safety disengagement is virtually the same. Ask any SDC company.
Oh wait i forgot, this is Tesla. 2 more miles is a quantum leap, exponential improvement, best thing since sliced bread.
Carry on.

 

[Knightshade]

1 mile vs 3 miles safety disengagement is virtually the same. Ask any SDC company.

How bout we ask---basic math?

Do you need an introduction? Seems like it

Oh wait i forgot, this is Tesla. 2 more miles is a quantum leap, exponential improvement, best thing since sliced bread.

It is literally a 3x improvement


There's plenty of bearish stories to tell about FSD without grossly misrepresenting numbers or just making them up by watching a couple of YouTube videos.

 

[S4WRXTTCS]

What I find most interesting is that prior to fsd beta, many of us were saying the sensor suite is impossible or pointing out some hardware limitation, but now with V9, we're not talking about hardware inadequacy anymore. Now we're pointing out disengagements or dumb decisions that have nothing to do with the sensors but rather the software or perception not seeing or acting on something that is obviously there.

From the FSD videos the majority of what I've seen is path planning issues, and perception issues.

From my own driving with AP/EAP the majority of the problems I've seen are Navigation, and too much latency between the time an external event happens and the car reacts.

Those observations don't exclude Sensor limitations from being a problem at some point, but just that they don't seem to be the bottleneck today.

My prediction is the current sensor suite is perfectly adequate for an L2 ADAS system that performs fine in most cases, but won't be enough for the HW3 vehicles to get L4/L5 approval for any public road in the US let alone Europe.

 

[jeremymc7]

Hopefully HW3 is enough for L3 though. That's enough for me for now.

 

[S4WRXTTCS]

1 mile vs 3 miles safety disengagement is virtually the same. Ask any SDC company.
Oh wait i forgot, this is Tesla. 2 more miles is a quantum leap, exponential improvement, best thing since sliced bread.
Carry on.

I don't think Tesla is at a point where miles per disengagement should even be considered.

It's so premature that any results are going to be fairly random depending on whether the current SW is meant to handle the situation or not.

The only videos I even watch lately are Chucks videos because he'll purposely test it over, and over in the same situation where he's observed a problem.

I'll know FSD is starting to get pretty good when he's no longer capable of easily producing a problem. But, Tesla will have a general FSD beta release before we ever get to that point and Chuck will be some forgotten memory of a time when Tesla accidently allowed someone capable of testing their FSD beta.

 

[powertoold]

My prediction is the current sensor suite is perfectly adequate for an L2 ADAS system that performs fine in most cases, but won't be enough for the HW3 vehicles to get L4/L5 approval for any public road in the US let alone Europe.

What particular failures with V9 makes you think that? I'm mostly referring to USA.

 

[S4WRXTTCS]

Hopefully HW3 is enough for L3 though. That's enough for me for now.

The problem with L3 is it can be so situationally restricted that I don't know if it would be enough.

If its only traffic jam assist its slightly valuable. Like watching Netflex on the center screen during a traffic jam wouldn't be too bad. Not sure anyone would pay $200/month for that though.

But, if its all the way to 85MPH than it might not be too bad but I can't see that being released due to the dangers of it.

 

[S4WRXTTCS]

What particular failures with V9 makes you think that? I'm mostly referring to USA.

Failures with V9 has had no negative impact on that statement.

As we get into V10, V11, etc we'll start to see limitations of the HW Sensor Suite really start to impact what it can do.

 

[powertoold]

Failures with V9 has had no negative impact on that statement.

Interesting, so you don't think Tesla will ever be able to perform as well as Waymo in Chandler? Assuming you've watched the Waymo videos

 

[qdeathstar]

If you don’t see the improvements between fsd 9 and 9.1 you are being a negative Nancy. I’m excited about 9.2....

 

[qdeathstar]

But, if its all the way to 85MPH than it might not be too bad but I can't see that being released due to the dangers of it.

On a limited access highway what kind of dangers do you think would be posed by it going to 85? It certainly seems the cameras are good enough for that.