Tesla nefarious behavior?

[majorfriend]

I'll stick with NHTSA instead of some random dude with a website. No wonder you don't understand how dangerous FSD is, you don't understand what a good source is.

I mean you do you, but the crowdsourced data during Covid was much more accurate than the CDC. Quite a few times the CDC was caught making basic data errors by folks working on crowdsourced data like Covid tracking project and they had to make corrections to published material.

Maybe that same incompetence doesn't extend to the NHTSA, but to me it's enough to call it into question.

Sponsored

 

[majorfriend]

I just watched the video, and to me the takeaway was that a self-driving system with Lidar would performance much better than a human in situations with poor visibility.

I'm pretty sure a person would have failed the same tests the Tesla did (except for the Wile E. Coyote wall).

It actually increased my confidence in FSD, just don't use it if visibility is poor (of if there are a lot of Wile E. Coyote walls in your area), otherwise it is pretty comparable to Lidar systems.

 

[captainjp]

I just watched the video, and to me the takeaway was that a self-driving system with Lidar would performance much better than a human in situations with poor visibility.

I'm pretty sure a person would have failed the same tests the Tesla did (except for the Wile E. Coyote wall).

It actually increased my confidence in FSD, just don't use it if visibility is poor (of if there are a lot of Wile E. Coyote walls in your area), otherwise it is pretty comparable to Lidar systems.

The problem with all self driving platforms is that they react rather than anticipate. Like someone mentioned before, most good drivers take notice of conditions 1/4 mile ahead.

 

[pamalabama]

Actual proof. We can definitively say that FSD (HW3 12.5.4.2) does NOT see the wall but HW4 and FSD 13.2+ can clearly detect the fake wall

That means if mark rober would have used a HW4 tesla with FSD enabled it would have passed 6/6 tests

Worth mentioning that HW3 FSD build 12.5.4.2 is more than 6 months old and there has been a HW3 update since then

 

[pamalabama]

The problem with all self driving platforms is that they react rather than anticipate. Like someone mentioned before, most good drivers take notice of conditions 1/4 mile ahead.

That's incorrect. Self driving platforms that are in commercial release can see further than a human using lidar. They can anticipate much further than a human can at night.

Secondly, these self driving systems anticipate similar to a human using similar visual clues. Big difference is that they are putting 100% effort at all times into anticipating from 360 degrees whereas a human can see only one direction at a time

The reason most car ADAS systems (other than tesla) appear to be reactive is because they physically cannot see stationary objects with the radar systems they use. Self driving cars do not have this limitation because they sense depth through cameras and/or lidar which are more robust.

 

[captainjp]

That's incorrect. Self driving platforms that are in commercial release can see further than a human using lidar. They can anticipate much further than a human can at night.

Secondly, these self driving systems anticipate similar to a human using similar visual clues. Big difference is that they are putting 100% effort at all times into anticipating from 360 degrees whereas a human can see only one direction at a time

The reason most car ADAS systems (other than tesla) appear to be reactive is because they physically cannot see stationary objects with the radar systems they use. Self driving cars do not have this limitation because they sense depth through cameras and/or lidar which are more robust.

Aren’t you the person saying lidar is inferior?

 

[pamalabama]

Aren’t you the person saying lidar is inferior?

no. Lidar is not inferior. It never was. RADAR which is what rivian uses is inferior is less safe if you are using it for as a backbone sensor as rivian claims

Tesla got rid of radar because it is useless for self driving.

If you have really good imaging radar then that provides some amount of benefit, but it is no backbone sensor

 

[s4wrxttcs]

So people die with Tesla's system because it is too good?

I think that's exactly it.

My predication is that the grows exponentially as an L2 system gets better and better.

In fact I predict that a Gen2 Rivian is less safe than a Gen 1 despite being more capable. That capability brings additional danger as its used more often and likely trusted more.

Where Tesla FSD is orders of magnitudes more dangerous than a Gen 2 Rivian for exactly the same reasons.

Once Tesla FSD gets to 99.5% its going to crash all the time because the drivers simply won't be paying attention and some edge case will happen.

My prediction is that Tesla's will begin to be uninsurable OUTSIDE of Tesla insurance due to crash rates being so high.

Buying the government was Elons only option not to face the FSD music.

 

[captainjp]

no. Lidar is not inferior. It never was. RADAR which is what rivian uses is inferior is less safe if you are using it for as a backbone sensor as rivian claims

Tesla got rid of radar because it is useless for self driving.

If you have really good imaging radar then that provides some amount of benefit, but it is no backbone sensor

These types of radar sensors do not work like that. They only provide range, bearing and speed of an object. The radar has no ability to identify the object it is detecting

 

[pamalabama]

These types of radar sensors do not work like that. They only provide range, bearing and speed of an object. The radar has no ability to identify the object it is detecting

With high enough resolution you can classify objects

Cruise control radars (which is what rivian has for corner radars) are fundamentally dangerous and vision is better

Self driving cars independently classify objects from lidar and they can see pedestrians at night from 100s of meters away

 

[pamalabama]

I think that's exactly it.

My predication is that the grows exponentially as an L2 system gets better and better.

In fact I predict that a Gen2 Rivian is less safe than a Gen 1 despite being more capable. That capability brings additional danger as its used more often and likely trusted more.

Where Tesla FSD is orders of magnitudes more dangerous than a Gen 2 Rivian for exactly the same reasons.

Once Tesla FSD gets to 99.5% its going to crash all the time because the drivers simply won't be paying attention and some edge case will happen.

My prediction is that Tesla's will begin to be uninsurable OUTSIDE of Tesla insurance due to crash rates being so high.

Buying the government was Elons only option not to face the FSD music.

your logic is that fsd will get to 99.5% and then people will start crashing. What qualifies a system as 99% anyways? If I go for 3 weeks without touching steering wheel or pedals and then I have a 10 second intervention, is that not more than 99% already? That's what FSD is like for me now.

And your mistake is assuming all "edge" cases will be dangerous or cause fsd to crash all the time.

It's pretty easy to avoid hitting something.

That's why Elon mentioned one time to pay attention to how many FSD interventions are absolutely critical. They are much less than you think and if you ignore any issues which are mapping limitations (which can easily be solved without any change to FSD), then the FSD interventions are already pretty close to the 99.5% you are talking about.

There are some caveats. FSD has a blind spot without a bumper camera. So while you can get in your car and safely start FSD, the car can't just start itself without you telling it to go.

 

[s4wrxttcs]

The problem with all self driving platforms is that they react rather than anticipate. Like someone mentioned before, most good drivers take notice of conditions 1/4 mile ahead.

This is something I noticed when I first had AP back when I had a 2015 Model S with AP1.

I felt like I lost situational awareness so I started to limit my usage of AP on long road trips.

Then I started see all the autopilot save videos where most of them simply wouldn't have happened as a good driver would have identified the threat well before hand.

I'm not really anti-AP or anti-FSD.

I'm definitely Anti-Elon. Screw that guy!!!
But, Elon is not Tesla.

He's barely doing anything for Tesla, and mostly he just digs them holes to climb out of like he's been doing for awhile.

Anyways all L2 systems can cause drivers to lose situational awareness. We're starting to see crashes pile up with Blue Cruise for example. Where they seem similar crashes with AP/FSD.

So I'm not particularly interested in L2 systems. Sure I like them from a standpoint of being able to open up a bag of cookies while I'm driving down a lonely stretch of freeway. but, I think they're all a bit dangerous if the driver has any trust in them.

I love riding in Waymo's and I look forwards to Waymo being available in the PNW.

I'm also looking forwards to anyone who introduces a L3/L4 capable system that can operate in the rain on my long drives to my Oregon Coast home. That would be awesome.

 

[pamalabama]

This is something I noticed when I first had AP back when I had a 2015 Model S with AP1.

I felt like I lost situational awareness so I started to limit my usage of AP on long road trips.

Then I started see all the autopilot save videos where most of them simply wouldn't have happened as a good driver would have identified the threat well before hand.

I'm not really anti-AP or anti-FSD.

I'm definitely Anti-Elon. Screw that guy!!!
But, Elon is not Tesla.

He's barely doing anything for Tesla, and mostly he just digs them holes to climb out of like he's been doing for awhile.

Anyways all L2 systems can cause drivers to lose situational awareness. We're starting to see crashes pile up with Blue Cruise for example. Where they seem similar crashes with AP/FSD.

So I'm not particularly interested in L2 systems. Sure I like them from a standpoint of being able to open up a bag of cookies while I'm driving down a lonely stretch of freeway. but, I think they're all a bit dangerous if the driver has any trust in them.

I love riding in Waymo's and I look forwards to Waymo being available in the PNW.

I'm also looking forwards to anyone who introduces a L3/L4 capable system that can operate in the rain on my long drives to my Oregon Coast home. That would be awesome.

The fundamental difference is that FSD is doing more than it ever did back in AP1 days.

The car sees better. It predicts better. It classifies hard to see objects better.

FSD right now could drive down the wrong side of a divided highway and it could dodge every car that it comes up against

Dangerous, sure, but this is an extreme edge event that would probably never happen.

The mistake is calling FSD a level 2 system. It is "technically" classified as level 4 for a reason. That's the intent and the system does not have any hard ceiling in terms of performance.

 

[s4wrxttcs]

your logic is that fsd will get to 99.5% and then people will start crashing. What qualifies a system as 99% anyways? If I go for 3 weeks without touching steering wheel or pedals and then I have a 10 second intervention, is that not more than 99% already? That's what FSD is like for me now.

And your mistake is assuming all "edge" cases will be dangerous or cause fsd to crash all the time.

It's pretty easy to avoid hitting something.

That's why Elon mentioned one time to pay attention to how many FSD interventions are absolutely critical. They are much less than you think and if you ignore any issues which are mapping limitations (which can easily be solved without any change to FSD), then the FSD interventions are already pretty close to the 99.5% you are talking about.

There are some caveats. FSD has a blind spot without a bumper camera. So while you can get in your car and safely start FSD, the car can't just start itself without you telling it to go.

My prediction is the crashes will go up significantly as FSD improves beyond a certain trust point. All the numbers I threw out were just guessing at where it would happen.

There is point where most human beings will begin to trust a system.

As soon as that point is hit then crashes will increase significantly.

I don't think Tesla is anywhere close to 99% let alone 99.5%.

I think it needs to be 99.99% before its even considered for unsupervised.

I do plan on renting a FSD capable HW4+ Tesla about once a year and trying it out on my long drive during the middle of winter.

 

[pamalabama]

My prediction is the crashes will go up significantly as FSD improves beyond a certain trust point. All the numbers I threw out were just guessing at where it would happen.

There is point where most human beings will begin to trust a system.

As soon as that point is hit then crashes will increase significantly.

I don't think Tesla is anywhere close to 99% let alone 99.5%.

I do plan on renting a FSD capable HW4+ Tesla about once a year and trying it out on my long drive during the middle of winter.

your assumption is that FSD interventions are directly related to safety limitations of the system.

Right now, FSD does not slow down for school zones. Every time it's an intervention. Assuming this data is integrated into mapping this can be solved without changing FSD.

HOWEVER, FSD can continue to improve with interventions being identical because interventions like this are not impending accidents if we do not intervene.

FSD is actually punished right now because their interventions are due to poor mapping. Something like 80% of tesla interventions could be fixed overnight with a map update

FSD is hard to react when it does something wrong. It's not as easy as it looks. But FSD does not crash like you think it does.

FSD might do something something stupid like not see a chain blocking a parking space. No issue because you can easily react.

Sponsored