The catastrophe of FSD and erosion of trust in Tesla

[Barklikeadog]

Capability. And while Im nowhere NEAR an expert on autonomous driving (or even regular driving..LOL)...I have sense enough to know. That whats in place TODAY on a 2022 Tesla? Is NOT going to be enough to do what the CEO of Tesla has promised, what FSD will be "capable" of doing.

I've always thought that a couple other items are required... Lidar, and the ability for cars to communicate position and velocity with each other. If you -really- wanted to make it work, cure congestion, etc,...

Just a quick sanity check here.
The cameras in my MS can see a lot more than I can see and can do so simultaneously. The image processing core digests all those different camera angles continuously without feeling the need to check its email. In short, my MS vision dramatically exceeds my own and yet I can drive from NY to LA without intervention from my wife although she would likely dispute the need for intervention.

So, what is missing here is not a sensor suite that can do the job but a brain that and work with the existing MS sensor suite that far exceeds mine.

The question for me is if the processing power is sufficient to out perform the average human to a significant enough degree that we will accept the machine's carnage over that of the average person.

It seems to me that Elon is correct when he says things like Lidar is nothing more than an expensive crutch. Do the job right and you do not need more than vision.

The job might get done right, but it won't be for any autos currently on the road. There's your sanity check.

 

[enemji]

So let me get this straight..we are supposed to believe that with the "current hardware" installed on todays teslas, these cars are going to hit level 5 capabilities to include "NYC to LA with no human input". With JUST the cameras/sensors installed today? A literal robotaxi service capability with JUST todays hardware?

But yet other companies have to use this much cameras sensors?

Elon is a wild dude.

View attachment 805737

Don’t forget he built rockets that all of NASA said he couldn’t and shouldn’t be doing it.

 

[enemji]

That could be backed up if they had done the job right. But when my MS hits the brakes hard for a shadow, I no longer have confidence in the system or the company that released a system that will brake for shadows.

Slowing down (for a shadow) is a problem? In what world? I slow down everyday when I come in and out of a tunnel or am under visually suspect conditions. Never had any issues.

 

[TS.42]

That could be backed up if they had done the job right. But when my MS hits the brakes hard for a shadow, I no longer have confidence in the system or the company that released a system that will brake for shadows.

This is my experience as well and am a new owner who experienced it at 80mph.

 

[2101Guy]

Slowing down (for a shadow) is a problem? In what world? I slow down everyday when I come in and out of a tunnel or am under visually suspect conditions. Never had any issues.

Did you just attempt to justify slamming on brakes at 80mph on a bright sunny day on the interstate solely due to trees casting a shadow on the road? Is that what you just attempted to justify?

 

[COS Blue]

Did you just attempt to justify slamming on brakes at 80mph on a bright sunny day on the interstate solely due to trees casting a shadow on the road? Is that what you just attempted to justify?

Yep, exactly what happened to me, except in my case it was the shadow of an overpass. Slammed on the brakes at 80mph on the Interstate. Yes, I consider that a problem.

Just one example of many many times the car has hit the brakes for no valid reason. It's embarrassing to me when I have passengers in the car, and it's an embarrassment to Tesla. Honestly, if they can't get that right or even figure out when to turn on/off the high-beams, no way should they be trusted to "fully self drive" the car.

 

[aronth5]

Just a quick sanity check here.
The cameras in my MS can see a lot more than I can see and can do so simultaneously. The image processing core digests all those different camera angles continuously without feeling the need to check its email. In short, my MS vision dramatically exceeds my own and yet I can drive from NY to LA without intervention from my wife although she would likely dispute the need for intervention.

So, what is missing here is not a sensor suite that can do the job but a brain that and work with the existing MS sensor suite that far exceeds mine.

The question for me is if the processing power is sufficient to out perform the average human to a significant enough degree that we will accept the machine's carnage over that of the average person.

It seems to me that Elon is correct when he says things like Lidar is nothing more than an expensive crutch. Do the job right and you do not need more than vision.

While I agree with your basic premise that the cameras on the whole can see a lot more than I can and do it simultaneously the one place the actually see less than I do is the B-pillar camera. When approaching a blind intersection by leaning forward (yes, just over the steering wheel in really bad intersections). I can see the intersection sooner than the B-pillar and make safety decisions sooner. ` Since the front camera's don't see 90 degrees they have limitations in this scenario. I have had to manually stop the car many times because I saw cars approaching on a blind corner sooner than FSD.

I don't see how the current camera locations will ever solve this. Many others have stated the same. And Lidar isn't needed either.

 

[daktari]

Don’t forget he built rockets that all of NASA said he couldn’t and shouldn’t be doing it.

I think this is an interesting statement.
You should not be shy about Nasa's achievements. The space shuttle, developed from the 60ies and early 70ies was built to be reusable. They reused the solid boosters, they reused the shuttle and replace worn protection on it very easily. Of course, Space X, 40 years later in the 2010s, took the concept even further with Falcon rockets. I would claim SpaceX is more of an natural next step incremental progress, not revolutionary.

 

[aronth5]

I think this is an interesting statement.
You should not be shy about Nasa's achievements. The space shuttle, developed from the 60ies and early 70ies was built to be reusable. They reused the solid boosters, they reused the shuttle and replace worn protection on it very easily. Of course, Space X, 40 years later in the 2010s, took the concept even further with Falcon rockets. I would claim SpaceX is more of an natural next step incremental progress, not revolutionary.

Have you been following Starship? If this isn't revolutionary I don't know what is. Same for Starlink.
SpaceX Super Heavy/Starship (BFR/BFS) - Earth to Deep Space
SpaceX's Massive Starship Could Soon Transform Spaceflight

 

[nvx1977]

I think this is an interesting statement.
You should not be shy about Nasa's achievements. The space shuttle, developed from the 60ies and early 70ies was built to be reusable. They reused the solid boosters, they reused the shuttle and replace worn protection on it very easily. Of course, Space X, 40 years later in the 2010s, took the concept even further with Falcon rockets. I would claim SpaceX is more of an natural next step incremental progress, not revolutionary.

If it were the "natural next step" of "incremental progress," reuse should be ubiquitous by now, given how old the shuttle program is.

People said the same thing about the iPhone (incremental, not revolutionary) because capacitive touch screen tech existed already. And yet that modality shift upended an entire industry. Revolutionary progress is not solely judged by the sum of technological components.

Shuttle was reusable but still cost a sh*tton of money AND time to prep for next flight. What SpaceX revolutionized was reuse WHILE significantly decreasing cost and time to refurbish. If all this were just incremental improvements, why don't we see any other launch systems with comparable capability? How is it that SpaceX has sent 7 crewed orbital missions, while the veteran ULA just managed to finally get its uncrewed demo flight off yesterday?

And all of this doesn't even consider Starship's capabilities.

 

[bradtem]

The SpaceX team are great innovators, and Elon's leadership has produced great results.
All that tells you is that maybe another team he leads can be a contender, it doesn't provide any assurances, and more to the point, when they keep doing it wrong, keep making no progress compared to the other teams, that they will somehow end up leading.

The space industry was (and is) a hidebound industry driven by a massive bureaucracy feeding stagnant companies on cost plus contracts. It is not surprising that one could outdo it. He brought thinking from silicon valley into the space business and as is often the case when you do that, you can get great results.


In self-driving, Tesla competes with Google, Apple, Amazon, Intel/MobilEye, and a raft of super well funded high tech startups. Tesla won EVs by bringing silicon valley thinking to the hidebound car industry. But he can't outdo Google and the rest by bringing silicon valley thinking to the self-driving space. He's competing against the companies that are already better at it than he is.

 

[lolachampcar]

"He's competing against the companies that are already better at it than he is." (bradtem)

I guess we will see

 

[bradtem]

"He's competing against the companies that are already better at it than he is." (bradtem)

I guess we will see

Well, we have already seen. Tesla FSD is, by the standards of the leading self-drive companies, absolutely terrible. I mean really dreadful. It's where Waymo was more than a decade ago, though it does it with simpler sensors. And while Waymo was able to drive in its mapped territories, Tesla FSD can't do the job *anywhere*. (It's funny how people sometimes think Tesla is better because the other companies drive their mapped areas and Tesla can't drive any area.) Remember, "can drive" means "can drive fully reliably every single time" and not "can drive once" or even "can drive 99.9% of the time."

Tesla is competing against Intel/MobilEye and Nvidia who are better at processors and mapping.
They are competing against Google which is vastly better at machine learning and AI and mapping.
Almost every competitor is better at sensors.
Telsa has a larger fleet to gather training than most, though their fleet is tiny compared to MobilEye and gathers vastly less data per car than the other companies with smaller fleets.

It is useful that Tesla has a large fleet they can control, and an army of people willing to test the prototype. But beyond that, it's not clear what advantages they have. On the plus side, Tesla is secretly doing some mapping and can switch to doing more without hardware cost. They will have a much harder time updating their sensors the way everybody else does.

 

[drtimhill]

Capability. And while Im nowhere NEAR an expert on autonomous driving (or even regular driving..LOL)...I have sense enough to know. That whats in place TODAY on a 2022 Tesla? Is NOT going to be enough to do what the CEO of Tesla has promised, what FSD will be "capable" of doing.

Pure speculation. If you are going to accuse Elon of speculating, dont do so by speculating yourself. it's pointless.

 

[drtimhill]

It's where Waymo was more than a decade ago, though it does it with simpler sensors. And while Waymo was able to drive in its mapped territories, Tesla FSD can't do the job *anywhere*. (It's funny how people sometimes think Tesla is better because the other companies drive their mapped areas and Tesla can't drive any area.) Remember, "can drive" means "can drive fully reliably every single time" and not "can drive once" or even "can drive 99.9% of the time."

it's funny how people sometimes think Waymo is better because it can drive well in its tiny geofenced areas.

It's also pointless comparing an aspirational L2 system with an aspirational L5 system.

 

[Bouba]

In self-driving, Tesla competes with Google, Apple, Amazon, Intel/MobilEye, and a raft of super well funded high tech startups. Tesla won EVs by bringing silicon valley thinking to the hidebound car industry. But he can't outdo Google and the rest by bringing silicon valley thinking to the self-driving space. He's competing against the companies that are already better at it than he is.

So where can I buy a Google/Apple/Amazon et al car? I can park it next to my FSD Tesla

 

[voyager]

https://twitter.com/x/status/1527700685055307776

 

[Terminator857]

...
In self-driving, Tesla competes with Google, Apple, Amazon, Intel/MobilEye, and a raft of super well funded high tech startups. Tesla won EVs by bringing silicon valley thinking to the hidebound car industry. But he can't outdo Google and the rest by bringing silicon valley thinking to the self-driving space. He's competing against the companies that are already better at it than he is.

Elon has a big advantage over those companies. Elon is willing to take risk. Google won't take much risk. You can't drive a car without risk. The decision to enable Level 3 is going to be about risk. Google will have an expensive over engineered product that minimizes risk. Elon will release the first thing that kind of works.

 

[Bouba]

Right now you have a choice of a very very expensive Mercedes with a pretend L3 that you can drive at 30 something miles per hour on an Autobahn....or a reasonably priced Tesla that can do L2+ anywhere in the world....or stick to adaptive cruise control (and there is absolutely nothing wrong with adaptive cruise control, it’s a great way to drive)

 

[daktari]

So where can I buy a Google/Apple/Amazon et al car? I can park it next to my FSD Tesla

Just buy a white Jaguar I-pace. (You can put some plastic boxes on the corners and on a roof rack to look like the real thing). Both the I-pace and your fsd Tesla will be similarly unable to drive autonomously.