Another of my predictive polls: FSD will or will not be achieved with "vision only". VOTE NOW!

[diplomat33]

Oh yes lidar is super important did you know Waymo had 60 incidents in the latest report of NTSHA and Cruise with lidars and radar stop working 4 times in 3 months because all this is connected to a center and stop working if no connection the goal of Tesla is the car continue to drive to safest place not stuck in middle of an intersection in case of no map working or connexion lost from Tesla. No for Today but for the future.

Yes, I know about the 60 Waymo accidents. But you need to provide some context. Many of them were not in autonomous mode. They were all minor. And those accidents had nothing to do with lidar. And those accidents were over a long period of time. Does not mean that we should ditch lidar. How many accidents did lidar prevent? Lidar does not magically prevent all accidents but it does help reduce accidents.

And the lost of connection is a Cruise problem. Waymo does not have that problem because Waymo does not require a connection to drive. And Tesla does not continue to drive to safest place. There are numerous examples of FSD Beta still trying to crash into objects, if not for the quick intervention of the driver.

 

[diplomat33]

Not better to have lidars and radars all around the cars

Pour l'instant, les robots taxis autonomes font un peu n'importe quoi

Les robots taxis à conduite 100% autonome, ça existe déjà. Mais pour l'instant, il y a de gros ratés. Certains de ces robots taxis décident

www.turbo.fr

That Cruise problem has nothing to do lidar or radar. Waymo has a lot of lidar and radar and does not have this problem.

 

[cwerdna]

Yeah, I forgot about charging. However Smart Summon might be able to get you into a spot with enough attempts!

But yeah I forgot that the NY to LA prediction was actually summoning across the country.

On this note, some of his claims related to this are at the below:

Elon Musk Today

elonmusk.today

Elon Musk Today

elonmusk.today

Elon Musk Today

elonmusk.today

 

[diplomat33]

I voted for Vision only, but it will be achieved by George Hotz before Elon...

https://twitter.com/x/status/1494915073394966532

Waymo can do 50,000 miles without an intervention and they have not not solved autonomous driving yet. But I am sure Hotz promising to drive 3.4 miles without an intervention by the end of the year means he is totally close to solving autonomous driving. (sarcasm). LOL.

 

[diplomat33]

That should be one "not". I accidentally put two "not" and I can't edit my post anymore.

 

[CyberGus]

Does “not not” self-negate, or does it doubly-emphasize the negation?

“I am not NOT going with you into that abandoned house”

 

[diplomat33]

Does “not not” self-negate, or does it doubly-emphasize the negation?

“I am not NOT going with you into that abandoned house”

I guess it could depend how you emphasize the words. But in written form, I think "not not" would normally self-negate. So saying Waymo has not not solved autonomous driving would mean that they have solved autonomous driving. That is why I posted the correction. I wanted to be clear that I meant to write that Waymo has not solved autonomous driving yet.

 

[emmz0r]

Not so hung up on sensors. I am more worried about how they will solve the problem with undetected objects.
Seems like they have to explicitly train for all types of objects instead of having some reasoning in classic "Software 1.0" (I hate that jargon) to detect what's not supposed to be in the lane.

 

[Olle]

I voted yes for vision solving FSD. but you also didn’t ask by when...

I believe that the type of sensors have little to do with the problem of universal self driving. How the system interprets and responds to the sensor output has everything to do with it.

 

[DJT1]

Waymo can do 50,000 miles without an intervention and they have not not solved autonomous driving yet. But I am sure Hotz promising to drive 3.4 miles without an intervention by the end of the year means he is totally close to solving autonomous driving. (sarcasm). LOL.

Problem with Comma, is Hotz just posted on Discord that Comma is burning $200K/month and is no longer profitable. The Comma-3 is double as expensive as the Comma-2 and the functionality, at the moment, is about the same for many cars and it just doesn't make sense to buy their next version. There is also a $300 clone of the Comma-2 by a Chinese company.

VC financing is drying up for non-profitable companies right now. They potentially have a serious problem.

 

[Goose66]

Oh yes lidar is super important did you know Waymo had 60 incidents in the latest report of NTSHA and Cruise with lidars and radar stop working 4 times in 3 months because all this is connected to a center and stop working if no connection...

I don't know if you have FSD beta, but if you do, can you honestly say that Tesla's vision-only FSD beta wouldn't have far and away exceeded Waymo's 60 incidents but for having a driver behind the wheel? Comparing Waymo's safety record vs Tesla's is meaningless, (similarly to comparing their respective hardware platforms) because Waymo has cars driving around without drivers and Tesla doesn't. There are no fully-automated Teslas. There isn't a driverless Tesla vehicle anywhere in the World.

 

[Ramphex]

I don't know if you have FSD beta, but if you do, can you honestly say that Tesla's vision-only FSD beta wouldn't have far and away exceeded Waymo's 60 incidents but for having a driver behind the wheel? Comparing Waymo's safety record vs Tesla's is meaningless, (similarly to comparing their respective hardware platforms) because Waymo has cars driving around without drivers and Tesla doesn't. There are no fully-automated Teslas. There isn't a driverless Tesla vehicle anywhere in the World.

Tesla doesn't trust FSD so much that it's forcing cabin camera to watch you, making sure you are 100% of the time watching FSD drive

 

[qdeathstar]

Yes it will be accomplished. Within 5 years or 50 years who knows.

 

[Ramphex]

Yes it will be accomplished. Within 5 years or 50 years who knows.

But will it? At which point will Elon see it as a waste of money with too little progress, too slow, to either transition to alternative methods or drop the project altogether? I mean honestly, this whole thing hangs on a thread of dumb people who keep buying their 2nd, 3rd, 4th Tesla and dumping multiple thousands of dollars on unfinished product that has been overpromised for years. If people wake up and stop buying it.... then what? Think he's going to waste his money on this project?

 

[Khashoggi]

But will it? At which point will Elon see it as a waste of money with too little progress, too slow, to either transition to alternative methods or drop the project altogether? I mean honestly, this whole thing hangs on a thread of dumb people who keep buying their 2nd, 3rd, 4th Tesla and dumping multiple thousands of dollars on unfinished product that has been overpromised for years. If people wake up and stop buying it.... then what? Think he's going to waste his money on this project?

Well, dumb or just want to be part of the future solution. it's a race and if you haven't noticed the cars are a serious blast to drive in the meantime!

And... TSLA is not his money. He is an investor like many and he is obligated to do what is best for all investors.

 

[Mr. Michael]

So maybe the question should be will we get a person to Mars before we get actual (non-beta bs) FSD? I vote Mars - less traffic.

 

[Ramphex]

Well, dumb or just want to be part of the future solution. it's a race and if you haven't noticed the cars are a serious blast to drive in the meantime!

And... TSLA is not his money. He is an investor like many and he is obligated to do what is best for all investors.

It makes sense to pay for FSD for the first car when you get promised X and Y, then be disappointed when X and Y don’t happen in years and it’s all just broken promises. But to then pay for it second and third car as the promises contine to be broken… well you know what they say fool me once shame on you, fool me twice…

 

[Goose66]

Not to be repetitive, but IMO there are two equally likely outcomes here: either Autosteer on City Streets is released to owners on the current HW3 "vision only" platform, but it is no more than the current ADAS features in the beta (with single vision stack for all features and perhaps some additional improvements), or Elon tweets that pressure from regulators has caused them to stop development of autonomous features, even though and early version of beta 12 in his car was mind-blowingly awesome. In this scenario, they will probably offer owners some credit for future purchase of EAP in your next Tesla.

 

[qdeathstar]

But will it? At which point will Elon see it as a waste of money with too little progress, too slow, to either transition to alternative methods or drop the project altogether? I mean honestly, this whole thing hangs on a thread of dumb people who keep buying their 2nd, 3rd, 4th Tesla and dumping multiple thousands of dollars on unfinished product that has been overpromised for years. If people wake up and stop buying it.... then what? Think he's going to waste his money on this project?

Vision only fsd will always be the most cost effective solution to autonomous driving. Even if Tesla capitulates and adds sensors I see know reason to stop working on a more efficient solution.

Also, I think we are a ways away from being sensor-restrained with cameras. Current fsd beta doesn’t react correctly to many things that the cameras should be able to perceive.

Current limitations appear to be software and processing related. But really I’m not in this field at all, so just a layman’s perspective.

 

[Olle]

Exactly, vision only is the most cost effective, at least from what we can think of today.

As you say the limitations appear to be processing related. If that is the case, a capitulation to use another input channel seems far fetched. As anyone who have used a camera understands, a person could drive a car based on the output from a camera as long as the image is wide and detailed enough and updated often enough. If you replaced the channel with something else, sonar for example and got it presented through a picture or through Braille, you could still drive safely, albeit slower because the resolution of the world would probably be worse and take longer to understand. The point is, as long as the shape of the world is transmitted and there is an intelligence to make decisions based on receiving the transmission, it could drive a car through it.

Artificial intelligence, able to process the input and decide on the correct output, is the unknown piece that Tesla is relentlessly working on. Once that AI is in good enough and it gets information about the world around it, the input channel appears trivial.
As anyone who has driven FSD beta has probably noticed by now, most of the hard to solve mistakes that the program makes is how to predict and respond to to the world, not the exact distance that for example LIDAR has an inherent ability to directly measure. Sure, ghost braking because a bridge looks like truck for a moment can be very annoying but its trivial compared to the huge problem of true autonomy.

This focus on the input channel reminds me of the 80s when we were all buying gold plated cables to get better sound in our HiFi systems