Elon: FSD Beta tweets

[S4WRXTTCS]

Facts require empirical evidence and at this time there is no supporting qualitative data to prove this as a fact. While it may or may NOT turn out to be correct it is an OPINION and NOT a fact. You have no way of disproving that cameras can NEVER work and just saying so doesn't make it so.

I'm skeptical myself that the current hardware setup will ever make L4. But this just like you is my opinion.

This comes down to requirements, safety thresholds, and timeline. Whether we determine whether HW3 is capable of delivering.

Requirements - For requirements we can use Tesla's original promise for FSD. That it was capable of autonomous driving in most situations that a human was capable of. Tesla also talked about robotaxi's, and that requires a pretty significant capability for autonomous driving to really work.

HW3 Capability Limitation - HW3 has shown poor performance in typical weather in the PNW. In my own experience it frequently prevented me from doing auto lane changes even when the weather wasn't all that severe. Now people can argue that the SW will get better to handle it, but I simply don't see that happening based on what I saw from the Camera images where they were all bouncy and I couldn't see anything. Now sure L4 allows a company to limit the ODD, but they'd have to limit it so greatly that it wouldn't really be sellable.

Safety Thresholds - This one is debatable, but vehicle to pedestrian collisions is definitely not something society will accept.

HW3 Capability Limitation - The car doesn't have any down facing 360 degree cameras, and is incapable of detecting nearby pedestrians/obstacles. It has to rely on slow speed and anticipating from what it did see. The vehicle also lacks safety features like rear cross traffic alert, and a 360 degree downfacing view. So the HW limitations are limiting even human drivers. Even under L2 supervision the vehicle has never achieved Smart Summon or Autopark capabilities that put it on par with typical human drivers.

Timeline - We're already nearing the 6 year mark between when FSD was originally sold with HW2.

HW3 Timeline - From a factual standpoint I think we'll know mid 2023 where HW3 stands. Anyone in technology knows its highly unlikely that advancements are made with old HW once its been replaced. There is no reason to believe that Tesla won't proceed with HW4, and HW3 will be left in the dust. We don't know the details, but existing HW3 owners should be anticipating a moment of "will I get upgraded".

 

[thewishmaster]

There are turns that I (Beta) take that I would consider 75% in 10.13 a GREAT result and improvement.

lol painfully true. some turns in my area hover around 0% success rate on 10.12.2

 

[S4WRXTTCS]

They can see in all directions simultaneously- a human can't.--- and without ever being distracted by the radio, their phone, a screaming child, a bad medical diagnosis, etc....

So they've already got a pretty better-than-human sensor suite.




Because vision is the type of sensor the entire driving and road system was designed to work with from the start.

Plus we don't want new cars to cost $250,000.

Cost versus safety will always be a huge part of the discussion for the safety requirements for self driving vehicles. Obviously we don't want them costing $250,000, but we don't want them to be running into pedestrians at night either or having detection issues in the rain.

The entire driving and road system was designed around human capability/limitations, and not to maximize the efficiency of the road. If autonomous driving lowers the cost of transportation than that's going to lead to more vehicles on the road, and this will lead to grid lock if robots aren't allowed to leverage their strengths.

None of this says you can't have a Vision based system, You can certainly design a purely vision based that exceeds human capability for dealing with situations like direct sunlight, glare, rain, etc.

I just don't see the point of not integrating other sensors that are better for certain tasks.

That being said I am a strong believer in finding a start point to evolve from. We're still making terrible comparisons to humans who very wildly from one person to another in both skill level and sensory level. Once we can start to compare robot to robot then we can see the best way to evolve.

 

[Knightshade]

HW3 Capability Limitation - The car doesn't have any down facing 360 degree cameras

Neither do humans and they seem to drive cars.

, and is incapable of detecting nearby pedestrians/obstacles

I mean, that's not true unless the pedestrians are only like 1-2 feet tall, or laying down, and around a blind corner so cameras never see them from more than a few feet away.... the only spots the cameras can't see are ALL OF:

Very close to the car
AND
Below the hoodline of the car
AND
Near the front 1/4-1/3rd of the car (because the fender cams and rear cam can see close and below the hoodline for anything further back)

The vehicle also lacks safety features like rear cross traffic alert, and a 360 degree downfacing view. So the HW limitations are limiting even human drivers.

Again human drivers don't have those things either.

But the car has the huge advantage of being able to see in 8 directions at the same time the human doesn't- plus the ultrasonics on top.

HW3 Timeline - From a factual standpoint I think we'll know mid 2023 where HW3 stands. Anyone in technology knows its highly unlikely that advancements are made with old HW once its been replaced. There is no reason to believe that Tesla won't proceed with HW4, and HW3 will be left in the dust. We don't know the details, but existing HW3 owners should be anticipating a moment of "will I get upgraded".

FWIW I 100% agree they're not getting past L2 with HW3 for city streets (I think L3 or L4 highway might be possible- maybe)

But HW3 was specifically designed to be drop-in swappable with HW2.x, and I've seen 0 reason to think HW4 won't be the same.

 

[S4WRXTTCS]

that's not true unless the pedestrians are only like 1-2 feet tall,

I think those humans are called Toddlers.

I would expect Elon Musk to be an expert in Toddler behavior, and how unpredictable the little monsters are.

 

[2101Guy]

I love the excuses..

“FSD is superior at not running over human beings.

Unless they are CHILDREN”

Some of these guys…

 

[S4WRXTTCS]

But HW3 was specifically designed to be drop-in swappable with HW2.x, and I've seen 0 reason to think HW4 won't be the same.

Depends on what HW4 consists of.

The discussion was only about HW3 + L4 (city + highway) though.

L4 highway using rest stops to switch from robot to human would be pretty sweet. It likely won't happen in the PNW due to rain, and that we love to close rest stops.

 

[AlanSubie4Life]

Now sure L4 allows a company to limit the ODD, but they'd have to limit it so greatly that it wouldn't really be sellable.

I think there would be massive demand for an actual L3 system even with a limited ODD. So I definitely think a limited ODD (no rain) L4 system would be sellable. Even in the PNW; I am well aware it rains *literally* all the time there (gotta keep the Californians away).

All theoretically of course; I don't think any of these things are going to happen with current hardware (even though I wish it would, since it would indicate capability far in excess of what we currently have, which would be awesome).

 

[thewishmaster]

But HW3 was specifically designed to be drop-in swappable with HW2.x, and I've seen 0 reason to think HW4 won't be the same.

If the computer itself is bundled with the ryzen MCU, I understand that one is not so easy to retrofit for cars that have the atom chip. Not impossible but the different battery voltage is one such complication

 

[Knightshade]

Depends on what HW4 consists of.

The discussion was only about HW3 + L4 (city + highway) though.

Which discussion? The thread is 181 pages long, and ostensibly about Elons FSD tweets.

L4 highway using rest stops to switch from robot to human would be pretty sweet. It likely won't happen in the PNW due to rain, and that we love to close rest stops.

I think L3 would be more likely, with a requested takeover if it needed to do anything fancy in anything more than light rain... (in my experience it's perfectly fine staying in its own lane on the highway in moderate, or even somewhat heavy, rain-- it's only the fancier NoA stuff that has trouble there like lane changes)

And that'd still be pretty great, you could still be watching a movie, or reading a book, etc while the car drives. And something I think HW3 is capable of.

If the computer itself is bundled with the ryzen MCU, I understand that one is not so easy to retrofit for cars that have the atom chip. Not impossible but the different battery voltage is one such complication

AFAIK that'd be the only complication. But given all the planning that went into the HW2.x-3 conversion, and the fact HW4 had already been announced as in development when HW3 was introduced, suggests it's likely one they've planned for.

I think those humans are called Toddlers.

Do you often have toddlers appearing within a foot of a car moving at speed without being visible from further away first?

I've been driving a lot of years and can't recall that ever happening to me.... but yes I suppose if one toddled into the street from being hidden behind the bumper of a big truck just a foot or two before you reach the spot FSD might hit it.

A human 100% for sure would too though. So would an FSD car with $250,000 of LIDAR because the car STILL had no line of sight (or light) to the toddler until it was 1-2 feet away and a car at speed can't stop in 1-2 feet.... so not sure how this is relevant to anything about safety- NOTHING could avoid that accident unless the car has x-ray vision.

I love the excuses..

Physics isn't an excuse, it's physics.

“FSD is superior at not running over human beings.

Unless they are CHILDREN”

Well, no, even then.

It's only when it's really tiny children, who also magically get within a foot or so of the vehicle without having been visible in any way first.

At which point it's then ONLY -as good- as a human.... (and probably still slightly better thanks to the ultrasonics).

In every other case it's better since it can see every direction at once

 

[AlanSubie4Life]

Do you often have toddlers appearing within a foot of a car moving at speed without being visible from further away first?

Yes, usually things are visible first and put in memory, with an expected trajectory. Something FSD will need. And in the case of parked vehicles obscuring toddlers/children/pets, humans would usually see the feet first (a capability that FSD absolutely will need to have as this sort of event happens all the time). Key capability that needs to be added!

 

[2101Guy]

Why are y’all focusing on anything lower than level 5? Perhaps many of you have not seen this, starting around the statements at 1:19 and again, at 2:30. His goal is clear and his timeline is clear

 

[KArnold]

His goal is clear and his timeline is clear

Perhaps clear in his mind. He's got issues with reality. His words mean nothing.

 

[JulienW]

aWhy are y’all focusing on anything lower than level 5?.....

Because I WILL be happy with a solid L3 system. This will do almost all driving and the little it doesn't cover I will be behind the wheel to do. No need for Robotaxi and not much wants of my car driving itself to pick me up. Don't care how aspirational Elon's mouth is. For me the important thing is how aspirational the work is and I see his mouth more as a indication of the work than a set of "facts".

I went in with my eyes wide open FULLY understanding what a monumental and mammoth challenge FSD is going to be. In fact Beta has already surpassed my original expectations. I never fully embraced video games but for me "fighting with Beta" is entertaining 95% of the time. So in effect I'm getting my moneys worth now and if we get to L3 will be "tickled pink" happy.

So when I bitch and complain about Beta and/or Elon it is more form an objective position (in my subjective opinion ) and not from animosity or from a fanboy perspective.

 

[Bladerskb]

How many LIDAR sensors do humans have again?

Because vision is the type of sensor the entire driving and road system was designed to work with from the start.

I'm not sure how people haven't realized that Talking to people like KS is an exercise in wasting your time.

Notice how people like KS all have the same talking points. All from elon because they don't have an original thought.
So they regurgitate whatever elon tells them." Humans only use vision not lidar, road system was designed for vision, if map or sensors disagree what do you do. if radar/lidar and vision disagree what do you do.. and more none-sense statements along those lines.

Yet imaging radar, lidar and camera combined are the closet thing to superman's normal and x-ray vision

I bet KS is a person who would flock to all Superman movies, buy all the superman paraphernalia and gear. Talk constantly about how amazing and OP superman is that he can see in the dark, through smoke, fog, dust and in direct sunlight. How amazing that he can fly through a freeway that has zero visibility weather to save people. How he can run into smokey buildings to save people. Yet we are talking about the same freeways built for normal vision, the same buildings built for normal vision.

And KS would cheer and say how amazing it is and claim its a superpower. But when you tell him that cars should also have something a little like this superpower.

He goes...

The hilarious thing is that xray vision doesn't take away superman's normal vision but enhances it, in a world where everything is created for normal vision. Yet to these guys Elon have convinced them that its lidar vs vision instead of lidar+imaging radar+vision VS vision.

Plus we don't want new cars to cost $250,000.

So would an FSD car with $250,000 of LIDAR

When you have no idea what you are talking about.
Have no facts behind you.
Just spew BS instead and regurgitate what Elon told you and hope no one notices.

Even though high resolution lidars used for robotaxi are going for under $500.
Even though dozens of cars are coming with atleast 1x high resolution front facing lidar.
Even though the entire current Mobileye robot taxi fleet cost $15k.
Event though Baidu just unveiled an EV that they will use as robot-taxi for $37k with 8 lidars.

But no go ahead. We are debating vision car and a car with $250,000 of lidar or having lidar would make a car cost $250,000

 

[2101Guy]

I'm not sure how people haven't realized that Talking to people like KS is an exercise in wasting your time.

Notice how people like KS all have the same talking points. All from elon because they don't have an original thought.

 

[2101Guy]

Me: “Isn’t it clear that Elon has clearly stated Tesla’s FSD will reach level 5 before 2023?”

KS: “no reasonable person would draw that conclusion from Elons statements”

me: posted a video of Elon clearly and repeatedly stating he is POSITIVE Tesla FSD will reach level 5 WELL before 2023

KS: “You misunderstood Elon”

 

[Smh2210]

Any rumors on FSD Beta V13, was hoping the usual Friday. TIA

 

[diplomat33]

Plus we don't want new cars to cost $250,000.

Can we please stop this nonsense about how lidar and lots of sensors make cars too cost prohibitive. This myth needs to die. Baidu just announced their next gen robotaxi with 38 sensors, including 8 lidar, that will only cost $37k. So it is absolutely possible to put lidar and lots of sensors on a car for autonomous driving that is affordable. Heck, even the Waymo I-Pace with its 50+ sensors costs half of $250k. Putting lidar and lots of sensors do not make the car cost $250k.

Baidu reveals the next-gen robotaxi that will be deployed throughout China | TechCrunch

Baidu, the Chinese search engine giant that has plowed money into AI and autonomous vehicle technology, unveiled Wednesday a new all-electric robotaxi

techcrunch.com

 

[Knightshade]

Me: “Isn’t it clear that Elon has clearly stated Tesla’s FSD will reach level 5 before 2023?”

KS: “no reasonable person would draw that conclusion from Elons statements”

me: posted a video of Elon clearly and repeatedly stating he is POSITIVE Tesla FSD will reach level 5 WELL before 2023

KS: “You misunderstood Elon”

Naah- you just conflated different statements. Because you're incapable of honest debate.

You asked about specific statements made last week

Everyone told you no, you can NOT reasonable read those to mean Elon promised L5 by end of 2022.


Somehow you've magically, dishonestly, turned that into a discussion of OTHER statements made at a completely different time--- because you realized the ACTUAL thing you originally raised- his remarks last week- don't mean what you tried to pretend they did.


It's not even that you can't argue honestly- it's that you're not even good at doing is dishonestly.

I'm not sure how people haven't realized that Talking to people like KS is an exercise in wasting your time.

Notice how people like KS all have the same talking points.

Much like the people who know the earth isn't flat, that vaccines are effective, and that global warming is actually happening, all have the same talking points.

When you reason from first principles, you generally will say similar things to others who do so.


On the other side we have the "Don't understand how anything works and rely on magical thinking contrary to facts" folks like yourself.

Great example here where I previously mentioned NO vehicle could avoid the toddler accident unless it had x-ray vision.

And you make this crazy Superman post about how awesome x-ray vision is in reply.



Does anyone who ever actually took a science class understand why putting x-ray vision on a car is a bad idea?