Autonomous Car Progress

[KLam]

Actually, the whole thing was set up to fail at the Stop sign if you view the clip carefully.

0.21 FSD engaged with requested speed at 36 mph and car going at 39 mph
0.23 Car slowing for curve, Ross up speed requested to 40 mph. ????
0.26 Stop sign ahead warning shows. Should Ross have move his foot to the brake paddle ????
0.31 Car speed at 32 mph due to slowing on curve. But moving up due to 40 mph setting.
0.33 Car speed at 37 mph. Ross slow down by scroll wheel to 36 mph?? Still no foot on the brake!
0.34 Stop sign
0.36 Car speed at 34 mph. Disengage.

I have to assume that Ross have traveled this road before and knew about the short blind stop distance and that Tesla do not detect and act on Stop ahead warnings which Traffic engineers put up to warn travelers. Also, the curves on the road should have slowed the car down a bit from the warning sign to the actual Stop sign. But Ross speed up the car by scroll wheel to 40 mph to negate the slow down.

 

[diplomat33]

Actually, the whole thing was set up to fail at the Stop sign if you view the clip carefully.

0.21 FSD engaged with requested speed at 36 mph and car going at 39 mph
0.23 Car slowing for curve, Ross up speed requested to 40 mph. ????
0.26 Stop sign ahead warning shows. Should Ross have move his foot to the brake paddle ????
0.31 Car speed at 32 mph due to slowing on curve. But moving up due to 40 mph setting.
0.33 Car speed at 37 mph. Ross slow down by scroll wheel to 36 mph?? Still no foot on the brake!
0.34 Stop sign
0.36 Car speed at 34 mph. Disengage.

I have to assume that Ross have traveled this road before and knew about the short blind stop distance and that Tesla do not detect and act on Stop ahead warnings which Traffic engineers put up to warn travelers. Also, the curves on the road should have slowed the car down a bit from the warning sign to the actual Stop sign. But Ross speed up the car by scroll wheel to 40 mph to negate the slow down.

Why would Ross set it up to fail? He's a huge pro-Tesla investor.

 

[MP3Mike]

Here's the screen shot again- You can see steering wheel is still blue (FSD on), speed is 34, the car is AT the sign it should be stopped at, and Ross does not have his foot on the accelerator.

The steering wheel color, and the chime, are lagging indicators. In that same screen shot you can see that the path tentacle is already grey and thin. FSDb has been disengaged at that point. (It seems that all of the disengagement animations take about 2 seconds, starting with the path and ending with the steering wheel.)

 

[Knightshade]

The steering wheel color, and the chime, are lagging indicators. In that same screen shot you can see that the path tentacle is already grey and thin. FSDb has been disengaged at that point. (It seems that all of the disengagement animations take about 2 seconds, starting with the path and ending with the steering wheel.)

Here's a screen shot half a second earlier (it still is at the 6 second mark on the timer)

The car is going 36, the tentacle AND steering wheel are blue, and Ross has his foot clearly OFF the accelerator. The car is maybe a car length from the sign and would be impossible to stop in time for it at this point too.

So again his claim he was already disengaged while accelerating is directly debunked by the actual footage.

As is his claim he accelerated BECAUSE of the white car cutting him off- since he's already moved OFF the accelerator here and the white car isn't even visible yet

 

[KLam]

Why would Ross set it up to fail? He's a huge pro-Tesla investor.

Good question! May be to make Tesla improve its AI a bit on planning like recognizing other road signs and reduce wrong lane changes before critical route redirections.

 

[MP3Mike]

Here's a screen shot half a second earlier (it still is at the 6 second mark on the timer)

The car is going 36, the tentacle AND steering wheel are blue, and Ross has his foot clearly OFF the accelerator. The car is maybe a car length from the sign and would be impossible to stop in time for it at this point too.

So again his claim he was already disengaged while accelerating is directly debunked by the actual footage.

@Knightshade can you explain this:

At 35 seconds the path tentacle is clearly blue and says "Stopping for stop sign":

at 36 seconds it has clearly changed to grey, and the animation has started to show the disengagement and the "Stopping for stop sign" message is gone because FSDb is no longer in control:

Here later, still at the 36 second mark, the path has gone to the thin grey line, while the steering wheel indicator has still not changed from blue because that animation lags the actual disengagement:

edit: for reference I am using the high-res video from here: Tesla Full Self-Driving Beta 11.4.4 ignores stop sign in live test with Ross Gerber and Dan O’Dowd

 

[Knightshade]

@Knightshade can you explain this:

Not sure what you think there is to explain?

I agree the blue tentacle goes away first before the wheel does- but I pointed out the tentacle was ALSO still blue AFTER he'd moved his foot away from the accelerator-that's why I posted the second screen cap with the tentacle still blue half a sec earlier and clearly too close to the sign to stop in time....


So his claim he was disengaged while accelerating is visibly false - heck your own first screen shot in this post shows the same.

As is his claim he accelerated due to the white vehicle that isn't even visible until after he'd moved away from the accelerator- again your own shot shows this to be a false claim by Ross.

Good question! May be to make Tesla improve its AI a bit on planning like recognizing other road signs and reduce wrong lane changes before critical route redirections.

Or he could just have been nervous and been pressing the accelerator without realizing it, and now refuses to admit it so keeps making up obviously false claims instead.

 

[Knightshade]

Thanks for the hi res link BTW, let's see if we actually disagree on anything here:

This is at 35 seconds. Tentacle is blue, steering wheel is blue, Ross has foot over (or on?) accelerator- stop sign is maybe 25-30 feet away and clearly visible, white car Ross claimed he was reacting to is clearly NOT visible (let alone cutting them off).
System has not been disengaged at this point (how could it be? what disengagement method?)

Are we still on the same page here at 35 seconds? If so let's move on!



This is ALSO at 35 seconds, so less than 1 second after the first shot... blue tentacle and wheel still are blue... stop sign is now maybe 10-15 feet away, white car still not visible let alone "cutting off" anyone. Only change is Ross has moved his foot from accelerator to brake and car slowed from 37 to 36 (with a 35 max)- we can not see if he's "pressing" or hovering that brake- but this is the absolute earliest he could have disengaged FSD even if we allow there's a visibile lag of ANY indiciation of disengagement, since the only way to disengage we are shown is brake pedal.
So system either is still engaged, or JUST was disengaged, at 36 mph, 10 feet from the stop sign.

Are we still on the same page later in second 35? If so let's move on!



I think you'll like this one! This is at 36 seconds. You can't entirely tell from screen shot but it's obvious in the video Dan has pressed the brake pedal (you can see his leg/foot move to do this between 35 and 36 seconds). Blue tentacle and wheel are still there- confirming that there's at least fractions of a second lag (but not in a way that changes anythign else that has been discussed)-- because the car is at or at best 5 feet from the stop sign at this point, still doing 35 mph.
So it clearly would not have stopped at the stop sign at the point disengagement actually happens. We DO see the stopping for stop sign message here- but far too late to actually stop....if you watch for that message to appear it does not appear until the car is MAYBE 10 feet from the sign, and doing 36 mph.

Also that white car that "cut him off" is still not even visible. Even if we allowed he has a slightly better angle to see it than the camea, it's clearly not "cutting him off" at this point, or even in the intersection yet.

So- just as I originally said--- All the claims Ross made are directly debunked by the actual footage, and we are left with 2 options:


1) FSD just blew the stop sign. Which many others have reported it doing once in a while (and I've had it do once too)
or
2) Ross was (perhaps unintentionally) holding down the accelerator right up until about 35 seconds into this video, leaving FSD unable to slow down and be able to stop for the sign and by the time he moved away from the accelerator it was far too late to stop in time.

 

[rbt123]

Why would Ross set it up to fail? He's a huge pro-Tesla investor.

A short-term dip helps the cash-rich buyer.

I don't believe this is the case, but the pump-and-dump has also has the bash-and-buy to go with it.

 

[stopcrazypp]

Why would Ross set it up to fail? He's a huge pro-Tesla investor.

I think he already set it up for failure even doing this demo because he bought into Dowd's argument that the system is expected to be flawless at this stage when it's squarely a L2 system. Dowd is coming from the perspective that FSD is hugely unsafe, treating it like an L4 system (that's why all the demos about running over children). But it's a squarely a L2 system (for example the accelerator override being discussed in this thread would not exist if it was a L4 system). As I posted before, we have long had L2 systems that would happily run the car into oncoming traffic left to its own devices, but generally that is accepted as fine because the driver is the one in control.

I don't follow Ross Gerber, but I would guess he is not a Youtuber or social media person who is used to doing recorded drives, so he probably was under a lot more pressure and anxiety than perhaps others that have been doing these drives all along.

I don't believe for a second even if the drive went flawlessly, Dowd would feel the system was safe, so it's largely a pointless exercise. He can always dig up counterexamples (which he himself has demonstrated).

 

[spacecoin]

A short-term dip helps the cash-rich buyer.

I don't believe this is the case, but the pump-and-dump has also has the bash-and-buy to go with it.

I think Goober knows he shat the bed with this stunning failure to kill off O'Dowd and came out swinging for FSDb afterwards so he wouldn't get the Gali cancel-treatment from the Tesla fan mob on Twitter.

 

[Knightshade]

But it's a squarely a L2 system (for example the accelerator override being discussed in this thread would not exist if it was a L4 system).

Quick point of order- as I agree with everything else in your post.... but L4 does not, at all, prohibit human control, override, or intervention.

It just can't ever require a human.

So "system will accelerate if human presses accelerator" could absolutely exist in an L4 version of FSD- though the system would not be operating at L4 while the human was doing that (but could drop right back to it once the human stopped).

 

[stopcrazypp]

Quick point of order- as I agree with everything else in your post.... but L4 does not, at all, prohibit human control, override, or intervention.

It just can't ever require a human.

So "system will accelerate if human presses accelerator" could absolutely exist in an L4 version of FSD- though the system would not be operating at L4 while the human was doing that (but could drop right back to it once the human stopped).

That's what I mean. As soon as the human overrides, it'll be out of "engaged" mode. The fact FSD Beta allows you to do that while still "engaged" is because it's a L2 system. I'm not saying humans can't do any form of intervention in L4 (for example safety drivers obviously can slam on the brakes), just that it won't remain engaged when it happens.

 

[Doggydogworld]

So it clearly would not have stopped at the stop sign at the point disengagement actually happens.

I agree this was a FSD fail and Ross is full of it. But the car is not supposed to stop at the stop sign here. The stop line is 40-45 feet past the stop sign, as measured on Google Maps. The intersection is in Montecito, near Santa Barbara. Ross was driving north on Sycamore Canyon Road.

34°26'01.4"N 119°38'45.4"W

goo.gl

 

[Knightshade]

I agree this was a FSD fail and Ross is full of it. But the car is not supposed to stop at the stop sign here. The stop line is 40-45 feet past the stop sign, as measured on Google Maps. The intersection is in Montecito, near Santa Barbara. Ross was driving north on Sycamore Canyon Road.

34°26'01.4"N 119°38'45.4"W

goo.gl

That's a crazy distance from sign to line I have to say.... But we can do some quick math here on this

Car is going 36 or 37 when Ross takes his foot away from accelerator.... Even if we assume the car has zero reaction time to begin HARD FULL braking, it'd take between 62 and 65 feet to stop. Taking it roughly 20 feet past the stop line.


EDIT- I went back and grabbed a screen shot the instant the "Stopping for stop sign" message appears--- car is going 36 mph, and is clearly less than the 62 feet it'd take to stop with MAX braking force in the remaining distance to the stop line (and FSD does not generally brake with max force in normal use either).... Ross could well have screwed things up with the accelerator, but the "stopping for stop sign" message would've had to appear earlier to have made even the stop line-- unless we assume there's also a multi-second delay from actual FSD braking start until this message appears?

 

[MP3Mike]

You can't entirely tell from screen shot but it's obvious in the video Dan has pressed the brake pedal

No, Dan did not press the brake pedal.

So it clearly would not have stopped at the stop sign at the point disengagement actually happens.

As it shouldn't have, that isn't where you are supposed to stop.

That's a crazy distance from sign to line I have to say....

Not really given the hidden nature of the stop line with the speed limit being 35 MPH, you need advanced warning.

Car is going 36 or 37 when Ross takes his foot away from accelerator.... Even if we assume the car has zero reaction time to begin HARD FULL braking, it'd take between 62 and 65 feet to stop. Taking it roughly 20 feet past the stop line.

It isn't unheard for FSDb to slam the brakes on. Did Ross stop 20 foot past the line? (Hard to tell exactly where he stopped given the angle of the camera view we have.) It looks to me like he stopped right about at the line, certainly not 20 feet past it, or he would have been in the middle of the intersection and collided with the white vehicle; so your calculations can't be correct. Ot maybe your stopping distance calculation is correct, likely, but your estimate to the distance to the stop line is wrong.

 

[zoomer0056]

No, Dan did not press the brake pedal.


As it shouldn't have, that isn't where you are supposed to stop.


Not really given the hidden nature of the stop line with the speed limit being 35 MPH, you need advanced warning.



It isn't unheard for FSDb to slam the brakes on. Did Ross stop 20 foot past the line? (Hard to tell exactly where he stopped given the angle of the camera view we have.) It looks to me like he stopped right about at the line, certainly not 20 feet past it, or he would have been in the middle of the intersection and collided with the white vehicle; so your calculations can't be correct.

There's a whole thread, just on this. Ross Gerber shows Dan O'Dowd that FSD is much safer than a human

 

[Knightshade]

No, Dan did not press the brake pedal.

Fair! I mean Ross of course.

As it shouldn't have, that isn't where you are supposed to stop.

Ok, but it also couldn't have stopped at the stop line either because physics.

It isn't unheard for FSDb to slam the brakes on. Did Ross stop 20 foot past the line? (Hard to tell exactly where he stopped given the angle of the camera view we have.) It looks to me like he stopped right about at the line, certainly not 20 feet past it

It does?

Here's the last moment you can see the stop line in frame and the car is still doing 32 mph.

Here's the street view of the stop line, and the tall poles on the right and the yellow sign on the right.

Here's the video at the last moment we can see that pole- which in the previous shot is just a few feet before the stop line-- the pole is just ahead of the A-pilar so the nose of the car should be at, or at MOST a couple feet, before the stop line. The car is still going 24 mph.

It's physically impossible for the car to have stopped at the stop line give it's going over 20 mph roughly AT THE STOP LINE.

Stopping Distance Calculator

Stopping distance calculator finds the distance your car travels before it comes to a stop.

www.omnicalculator.com

From 24 mph, with 0 reaction time, we get a stopping distance of 27.53 feet.

Now it's possible the pole to the stop line in pic 2 is maybe 5 feet and maybe he's another 5 from the pole even in pic 3.

Both look like less to me, but sure.... and heck add another 20% to both distances.

And he still ends up at least 10-15 feet past the stop line because physics.

, or he would have been in the middle of the intersection and collided with the white vehicle;

He WAS in the intersection- that's why the white car is honking at him. There's no good way to get screen caps for this bit but watch the above-yoke visual for the second or two from about 10 mph to 0, you can see the intersection visualization with the curve to the right of the adjacent road and you can see that right curve vanish as the S stops beyond it.... (the GPS visualization also suggests he's partly in the intersection but it's not nearly zoomed/accurate enough for that to mean much beyond agreeing with all the other data.)

 

[JB47394]

Criminy. Does anyone live in that area and can just path through that intersection the same way?

 

[spacecoin]

Not really given the hidden nature of the stop line with the speed limit being 35 MPH, you need advanced warning.

You mean like this sign that Ross and the car passed 5 seconds earlier and completely ignored?