FSD rewrite will go out on Oct 20 to limited beta

[AlanSubie4Life]

That's actually not the worst idea. That guy definitely failed the test and should have his beta revoked! haha.

You can rest assured that they are keeping a watchful eye on the driver responses to these alertness checks, and will be revoking privileges shortly for this driver, as these events are automatically uploaded to Tesla for review and grading by the FSD team. It's all designed in.

 

[helvio]

I don’t think his beta should be revoked, he reacted in a timely fashion. Did he drive the wrong way for a bit? Yea. I wouldn’t have done it differently. There was no oncoming traffic, so no need to get out ASAP. It’s ok to have some comfort while getting back to normal. You guys keep saying “what if” and then the “if” happens and you give the guy a hard time lol cut him some slack. It was safe!

 

[drtimhill]

You fail to appreciate that they are building these failures into the beta, which are designed to only occur in safe situations, to help keep the beta testers on their toes. Why else would they have included the "It may do the wrong thing at the worst time, so you must always keep your hands on the wheel and pay extra attention to the road" message, for the beta? They designed it that way. You are not giving Tesla enough credit here.

Are you serious???

Do you have any evidence to bsck this?

 

[Daniel in SD]

I don’t think his beta should be revoked, he reacted in a timely fashion. Did he drive the wrong way for a bit? Yea. I wouldn’t have done it differently. There was no oncoming traffic, so no need to get out ASAP. It’s ok to have some comfort while getting back to normal. You guys keep saying “what if” and then the “if” happens and you give the guy a hard time lol cut him some slack. It was safe!

He was not holding the wheel. Look at the picture, he allowed it to turn nearly 90 degrees!

 

[helvio]

He was not holding the wheel. Look at the picture, he allowed it to turn nearly 90 degrees!

Aw you gave me my first thumbs down
But seriously, i don’t think it was that bad. Was he perfect? No. Wrong enough to be publicly shamed and revoked? Also no. It’s just fine and we get more examples. I’m good with that!

 

[drtimhill]

Aw you gave me my first thumbs down
But seriously, i don’t think it was that bad. Was he perfect? No. Wrong enough to be publicly shamed and revoked? Also no. It’s just fine and we get more examples. I’m good with that!

no no NO ... if you revoke people who dont respond right you have it backwards. The goal is NOT to choose drivers who correct the cars errors .. the goal is to find where the car has to ask ordinary drivers for help.

 

[Daniel in SD]

no no NO ... if you revoke people who dont respond right you have it backwards. The goal is NOT to choose drivers who correct the cars errors .. the goal is to find where the car has to ask ordinary drivers for help.

The goal is to find the bugs and fix them. That does not require swerving into the oncoming lane. This guy didn't even report the bug and then starts using his phone while driving at the end of the video! Hardly an "expert and careful" driver. Not everyone is qualified to test autonomous vehicles on public roads. I think what Tesla is doing is against the CA DMV rules and if there are incidents it seems likely that testing by customers will be banned. Tesla has plenty of "expert and careful" drivers to test the system, there is no need to take this risk.

 

[drtimhill]

The goal is to find the bugs and fix them. That does not require swerving into the oncoming lane. This guy didn't even report the bug and then starts using his phone while driving at the end of the video!

The driver didn't swerve into the oncoming lane .. the car did. Sure, he didn't react the way I would have liked, but he bought the car under control pretty quickly. Would you have done any better? And as I've noted, the issue here is not to find "skilled" drivers who can handle a car that misbehaves. The goal is to show that the car is safe in the hands of drivers like this.

Consider this if you will: Near the end of world war 2, the US was bombing Germany, but was losing a lot of bombers to anti-aircraft fire. To fix this, they decided to increase the armor-plating on the bombers. But they quickly realized that if they did that all over the bombers, the poor aircraft would be too heavy to fly. So, they decided on a practical approach: take a tally of which bits of the airplane were hit most often, and only armor-plate those parts.

Was this a good solution?

 

[Daniel in SD]

The driver didn't swerve into the oncoming lane .. the car did. Sure, he didn't react the way I would have liked, but he bought the car under control pretty quickly. Would you have done any better? And as I've noted, the issue here is not to find "skilled" drivers who can handle a car that misbehaves. The goal is to show that the car is safe in the hands of drivers like this.

"It may do the wrong thing at the worst time, so you must always keep your hands on the wheel and pay extra attention to the road"
The car did the wrong thing but fortunately not at the worst time. I don't see how you can be so confident that there wouldn't have been a collision if it was at the worst time? It should be obvious that a car that does the wrong thing at the worst time is not safe when driven by someone not holding the steering wheel correctly.

Consider this if you will: Near the end of world war 2, the US was bombing Germany, but was losing a lot of bombers to anti-aircraft fire. To fix this, they decided to increase the armor-plating on the bombers. But they quickly realized that if they did that all over the bombers, the poor aircraft would be too heavy to fly. So, they decided on a practical approach: take a tally of which bits of the airplane were hit most often, and only armor-plate those parts.

Was this a good solution?

No, the bullets hit the plane randomly, you should add armor to the places that were not hit on the planes that returned. I've heard this before but I'm not sure how it applies here?

 

[mhan00]

Obviously that guys beta privileges should be revoked. Keep your hands on the wheel!
What's odd is that the perception looks fine, I wonder what went wrong?
View attachment 604619
Everyone who posts videos with bad camera angles like this guy should also have their beta privileges revoked.

Funny how I got downvoted for being upset that another beta tester posting videos clearly wasn’t holding the wheel the entire time he was using the system. As I said in that post: I don’t need to see hands off the wheel; I can see the blue wheel signifying the system is on, so keep your damn hands on the wheel so you can not only take over ASAP if it swerved, but also so you can feel the torque of the wheel if it does start to do something unexpected or wrong.

 

[AlanSubie4Life]

So, they decided on a practical approach: take a tally of which bits of the airplane were hit most often, and only armor-plate those parts.

That's the opposite of what they did.

 

[MadOverlord]

We are in a curious world where we need to make self driving cars drive WORSE to reassure the human occupants.

Not quite correct. The car should never drive worse than it can, but it should not attempt maneuvers that are outside the comfort zone of a typical experienced but conservative driver. Personally, it occurs to me that given the inherent randomness of the environment (aka "weird *sugar* happens"), this approach is going to be required to get those last few 9's of reliability, because that builds in a buffer to react to the truly strange stuff.

 

[P3dStealth]

The driver didn't swerve into the oncoming lane .. the car did. Sure, he didn't react the way I would have liked, but he bought the car under control pretty quickly. Would you have done any better? And as I've noted, the issue here is not to find "skilled" drivers who can handle a car that misbehaves. The goal is to show that the car is safe in the hands of drivers like this.

Consider this if you will: Near the end of world war 2, the US was bombing Germany, but was losing a lot of bombers to anti-aircraft fire. To fix this, they decided to increase the armor-plating on the bombers. But they quickly realized that if they did that all over the bombers, the poor aircraft would be too heavy to fly. So, they decided on a practical approach: take a tally of which bits of the airplane were hit most often, and only armor-plate those parts.

Was this a good solution?

The truth of that is opposite of what you are saying. They armored the un-hit portions of the planes that made it back. The planes that came back with damage survived so looking at their damage and talking to their crews was the wrong approach. It's called survivor bias.

It doesn't apply here. Unless we benefit from sacrificing some of the beta driver's somehow. I'd prob start with the ones who just scream on youtube.

 

[byeLT4]

When taking the exit to the Victoria, TX supercharger (also my exit for work), my car always chooses the wrong lane of the feeder road. Sometimes when there is no traffic I let it stay briefly to see if it will correct itself but I have never seen it happen. Full disclaimer- I am still on crappy pre-Beta FSD

 

[hamoneaster]

Any word on when the rest of us might get the FSD? lol

 

[quadroplexor]

Any word on when the rest of us might get the FSD? lol

Guess my 4 YouTube subscribers means I’m last in line.

 

[brkaus]

Guess my 4 YouTube subscribers means I’m last in line.

Your 3 ahead of me!

 

[Saghost]

Any word on when the rest of us might get the FSD? lol

"End of the year."




Again.

 

[drtimhill]

The truth of that is opposite of what you are saying. They armored the un-hit portions of the planes that made it back. The planes that came back with damage survived so looking at their damage and talking to their crews was the wrong approach. It's called survivor bias.

It doesn't apply here. Unless we benefit from sacrificing some of the beta driver's somehow. I'd prob start with the ones who just scream on youtube.

You are correct. But my point wasn't about risk per se, it was about using a biased sample. The planes that came back were not representative, and so the conclusions were incorrect (in fact, you need to armor the part that do not have bullet holes). A lot of the posts here talk about "trained" or "skilled" drivers, and say "such and such should not be allowed in the beta". But then you are creating a biased sample. You will end up with a car that can drive only with skilled drivers, which isnt the goal here.

Of course, this depends a lot on what phase Tesla feel they are at in the testing cycle. If they are really at alpha (dont even know f it will work), then handing the car over to anyone would indeed be foolhardy. But the claim is they are not .. and in this case they really do need to use everyday drivers for testing.

 

[drtimhill]

No, the bullets hit the plane randomly, you should add armor to the places that were not hit on the planes that returned. I've heard this before but I'm not sure how it applies here?

It applies because in the case of the airplanes they were looking at a biased sample. All the calls here for "trained drivers" etc are, in effect, creating a biased sample of drivers. That's useless when training the car (and NNs in general).