On Monday, Tesla-investor-turned-outspoken-critic Ross Gerber posted an image taken inside his Tesla showing it driving straight toward the Sun, with an alert on the car’s infotainment screen yelling at him to “Take Over Immediately.” The reason given: “Front camera blocked or blinded.” “Got the new tesla FSD [14.2.1.25](http://14.2.1.25) and within two minutes of driving in the Sun. This,” Gerber wrote. “These cars can’t drive into Sun…”
Another day, another Gerber doom video. Stock up 5% rn
At this point Gerber just seems bitter that he sold his Tesla stock too early and lost a bunch of money
Or in the snow. Or in the fog. Or in anything more than moderate rain.
bUt hUmAnS oNlY uSe vISiOn!1!1
So it’s perfect as long as you are a vampire.
FSD or any autonomous system is defined by what it does not do "good enough", not what it does do "good enough". This is the problem that all the Tesla fanboys have. As an FSD owner who paid for it in early 2017, I assure you: FSD has relentlessly improved from "not safe enough to drive while you read" to "not safe enough to drive while you read." There is every possibility it is asymptotically approaching "not good enough."
Sunny day? Another tricky edge case.
It seems like a well known limitation. You see the same thing in Tesla’s Robotaxis. It matters for maybe ten minutes a day (0.7% of the time) when you’re facing within maybe 1.5° of the sun (0.8% of possible headings), so I think it’s just a numbers game. Like if they don’t actively avoid driving certain times, or certain directions at certain times, then there’s a small chance of Robotaxis suddenly hard breaking due to the sun, and there’s a small chance that someone will rear-end them due to the braking, and a small chance someone will die from that rear-ending. But on the whole, they’ve consider those acceptable deaths, hopefully offset by overall lower rates of fatalities per mile driven. And legally, they can argue that it’s the other driver’s fault, so the deaths might not cost them much financially.
That’s the complaint? Please. That’s only a problem for something like a third to half of the year…
Or as long as your work place isn't easy of your home in the morning. Then your f**d going to work and home.....for summer at least. 😬
Driving into the sun never actually happens, 100% ready for robotaxi. Tesla is perfect for places where the sun don't shine.
But they do great in the rain. /s
In those conditions, a human would wear Polaroid lenses, to reduce optical glare.
He only got this issue now!!!? It’s been going on with FSD and Autopilot since they disabled radar.
Stock at ATH. Absurd, really.
Or at night...
If I tried to sell you any other stock at a PE ratio of 338 and told you sales are going down YoY and that the CEO was part time, recently gave not one, but two “salutes” to a large crowd on camera, and was pushing 10 years late on delivering a key product that many customers paid in advance for you’d laugh in my face. Tesla is a meme stock that at some point will succumb to gravity. Maybe soon, maybe in a few years. Books will be written about the willful negligence of institutional investors who kept spending endowment and pension money despite rampant signs that this thing was destined for a Hindenberg level return to earth.
And lidar laughs and continues to see everything just fine ... Also I think you mean polarized not Polaroid
Or where there is no snow or inclement weather.... This must be why Elmo wanted to build those dumb tunnels
We tell passengers to aim *away* from Sun—they should know that.
I mean, it’s not wrong - humans have fairly lousy optical sensors (the human eye only sees about 1 degree of our field of view in high-res focus) augmented by a super-duper-computer-level perception system. Elon’s problem is he’s bad at math and basic reasoning. LiDAR is cheap and energy efficient. Supercomputers are expensive and consume massive energy. So the solution for cars is obviously to use superhuman-level sensors with subhuman-level computing.
Fucking yikes
Shitty cars made with shitty materials gunna do shitty things. Version 14 "fsd" still out there running red lights and driving onto oncoming traffic.
Lol cynical. If it wasn‘t for TSLA, I would consider you a psycho. But what you write sounds actually legit.
None of what you said disproves what I wrote . Gerber has lost a lot of money by getting rid of his Tesla stock too early . Meme stock or whatever stock , money is money
Wait, didn't also had problems at night?
Do not forget at night.
My Tesla can NEVER handle simple cruise control. A speck of dust on the camera is enough for it to fail.
Why does this look like a headline from The Onion?
...who doesn't drive at night either
Yes, Robotaxis don’t drive into the sun, they drive to uranus.
And the fog /ss
>This,” Gerber wrote. “These cars can’t drive into Sun…” Maybe Ross can sell T shirts that read "I'm going to drive my next Tesla into the Sun"?
It can drive in the snow. It can drive in heavy rain, forget moderate rain. It can drive in the fog to the extent I can. I have a Model S and been using FSD for the past 8 month with no interventions.
Ok then. You’re dead, but relax because you’re just an edge case.
Humans can squirt and/or put down the sun visor. FSD can do neither.
Only drive west in the morning, and east in the evening.
Safety driver-less Robotaxis already in Austin!
It works fine in the sun direct sun. I use v14 at these sun angles daily going home. Bros camera glass area was probably dirty (which happens twice a year ish, needs cleaning).
Not sure how squirting would help, but I’m not ruling it out
lol
Lidar is a fools errand... because it would kill my profit margins
ACME FSD will take us to Mars
Why can't Tesla just sell a $5,000 option for sunglasses for the camera?
We have two Tesla Y's, a '21 and a '23. Love them both. But the reality is that autopilot and FSD are both seriously flawed and an only be used safely on divided highways.
bring a towel!
and not scared of leaves at all
it's a new tech using a chemical process to transfer images to paper where a camera (not pointing at sun) can capture the image. just a bit of latency, but no more than pulling down a sun visor and applying mascara
Or around school buses.
Honestly that would probably be safer than whatever the hell Tesla is doing
Remember when Elon “debunked” the FsD glare issues by claiming they use “photon counting” that wasnt affected by glare LMFAO
Or add $200 Lidar
Using high quality sensors can allow you to use a raspberry pi. Not that i would use one of those in a production environment where human life is at risk. Om just giving you an idea of the computing power required with good data. With vision only you literally need a multi gpu server in the car trying to interpret what it is seeing and then simulate an entire 3d world based on that data. A raspberry pi level computer with high end sensors is told there is an obstruction in front of the car A tesla with vision only has to guess if there is an obstruction in front of the car.
My Model Y drives just fine for me in the bright FL sun, and in rain and in pretty thick fog and at night when the idiots driving towards me have their bright lights on so bright I can't see anything.
But his cars also cant work whefe there's no GPS! Which is why those cars in the tunnels need drivers. Imagine having offline sensors like LiDAR, Radar, ultrasonic, proximity that can map out the environment you are in like a fucking robot vacuum that we have had for decades. L
Those are edge cases. A gently lit indoor track it’s almost flawless assuming the ambient sound is just right.
You joke about the pi but plenty of L2 systems out there with less compute.
How often do people really drive during the day though.
Or they could just add the lidar and not do that.
Or if there are shadows on the road. But at least they'll drive slightly better in perfect conditions than a drunk teenager texting while driving.
So moderate psoriasis?
Jesus Christ Elons going to buy Kodak isn’t he?
Great for overcast weather without rain, taking over the taxi markets right now!
Yeah but that’s cheaper. Everyone knows cheap is bad every other statement.
So does everyone else Elon. My god damn flip phone counted photons, probably worse than a potato though.
If only there was a technology that did not rely on optical vision for measuring distance. Maybe it will be invented in the year 3025.
I’ll consider FSD _might_ not be a death trap scam when the auto wipers on my model Y start properly identifying when it is and isn’t raining.
I think drizzles cause most issues with cameras outside cars Also. Leaves are a big issue. The way the move when there is wind Best to remove all these before you turn on FSD
Leon picking camera only is peak fucking idiocy.
I had continuous warnings about disabled cameras driving on I5 from Olympia to Tacoma. 2020 Model Y. Your S must be special.
They said radars and lidars made their cars less safe. Sounds like an excuse to be cheap, but if your engineers don’t know what they’re doing, or don’t have the time or budget to integrate it properly, that’s possible.
each sensor itself would need its own processor/AI to determine obstructions and inform the car appropriately, moving the processing requirements from the central processor to the satellite locations of each senors (power and cooling becomes a bigger issue)... you also have an even bigger issue of each sensor's AI deciding what it sees has precedence over another senor's interpretation of reality... (even if they are of the same type). "Here's what I see" (point cloud) is very different than "I see a person within this point cloud that is walking into my driving path with a probability of X that I might hit them, and I should do Y action to prevent it" (object classification, rational thinking) PS, I didn't write this to defend Tesla's approach at all... they need an Imaging radar at minimum to safely get above L2 in all weather applications. and HW4 does not offer enough processing power to make sense of a dense point cloud generated by an Imaging Radar with doppler 4D rich data of all points moving/stationary.
And exiting tunnels where the shift in light has the same potential... Or I dunno... an angry person with a very bright LED flashlight that doesn't like Tesla, and knows that it can blind the camera with a $10 item...
I've been in situations where I couldn't drive safely either.
It's great, except when it's not.
Installing Lidar would be admitting he was wrong. He had doubled or tripled down on cameras only being the best solution.
It’s fine so long as you don’t expect it to deal with aberrant conditions such as sunlight or dirt. Or fog Or smoke Or optical illusions.
Dude, during the day is when your car turns into a Robotaxi and makes you millions!
Bright sunny afternoon in Chicago last Monday 12/8. Sitting at a red light when I got rear ended by a 2026 Tesla Model Y in (you guessed it) Full Self Drive mode. Conditions were perfect otherwise, but the bright sun combined with the reflection from the snowfall from a few days earlier apparently scrambled the FSD.
Causing accidents and plowing pedestrians in no time.
One would think the Vegas loop would have have exclusively roboTeslas from the start
But it’s great for driving into motorcycles at night!
Dude the world is an onion headline right now
And the part where it tries to murder people.
With today’s cost I don’t even think it will. I’m no AI expert but I think felon is going all in on vision because his training data is all vision based. He will lose a lot of his advantage (the one he had) if he back tracks and adds lidar info. Please any AI expert can chime in and clarify my basic understanding.
I wasn’t mentioning the Austin situation out of respect, but out of a mixture of disgust and irony.
Except when humans sit inside the car
Fuzzy_Inevitable9748 • 9h ago In Doctor who they have quantum angels, they are statues that can only move when no one looks at them, but they move damn fast so you can’t even blink. This post has me picturing a world where teslas are hunting us down quantum angel style. << That 100%
Sprouting potatoes prove: they are fucking good detecting certain wavelengths.
Imagine being a back passenger to the robotaxi when the car is crossing a bridge that crosses the ocean like the San Francisco bridge or driving through high mountains that have no guard rail with a 5,000 feet drop.
Thats not really fair though. Our foveas achieve something like 3\~6x the effective pixels-per-degree of a car mounted camera. So really we have the ability to resolve much finer detail. That combined with our supercomputer brain that can point it exactly where detail is needed most is where it all comes together.
They validate models using lidar as seen by the kit installed on ADAS vehicles often spotted and posted online.
Camera sensor technology is developing and it was recently announced that Sony are producing a new sensor for Tesla that may utilise infra red which can see through fog and have low lux 'starlight' capabilities. Still does not solve rain, snow and other physical obstacles that can only really be overcome by radar. The resolution of synthetic aperture radar is getting as good as high resolution camera sensors and may actually become a single technology replacement for cameras. You could say cameras are the fools errand
Hey i thought Musks dorky robots can do everything so why not just have them drive cars? You won’t have to make self driving cars then.
Or a snowstorm.
Or at night? Or in the rain? So it’s great except for the part where it sucks night and day?
to be fair the sane conclusion would have been: public transit + bicycle infrastructure. well and once we ignore that the solution would be: all the possible sensors + the best compute, because that is what you do if the goal would be to make it as safe as possible and not just "safe enough". the idea to not use lidar wouldn't even remotely come up, if elon ever cared about human lives. if tesla wouldn't have run on scamming people, the cars would have gotten sold with a place for the hardware ready to insert it and a place for cheap enough lidar and other sensors ready to put in, once all of those are cheap enough to be sold to people. but then he couldn't have done all the lying of "hey the hardware is there, next year the software will be delivered for sure, full self driving is coming..."
My model Y windshield wipers were similarly great. Except they often times would wipe furiously in the bright sunshine for no reason. And in the rain, they were suddenly exhausted and didn’t want to wipe at all. But man, when it wasn’t super sunny or super rainy, those wipers were awesome.
Yes, best possible sensor suite + best possible compute is the rational approach - and that’s what Waymo did. Now that it works, Waymo is in cost optimization mode as they gradually scale. Cost optimization should only happen AFTER the product is working at scale. That’s how you build cutting edge tech. Trying to cost optimize BEFORE it works is literal suicide (both for the company and the humans involved in testing). If NASA had tried to cost optimize in the 1960s before getting the first space shuttle to the moon - it would have taken a decade longer and a LOT of astronauts would have died in the process. But Elon sees his customers as expendables.
>before getting the first space shuttle to the moon now that's a theory i haven't even heard :) \_\_\_\_ shuttles never went outside of earth's orbit and there is no dispute about that either so funny typo by you or sth creating a completely new idea of what might have happened in regards to moon landings ;) also i believe, that the claimed manned moon landings were a terrible example, because regardless of what you believe happened, one thing all agree is, that there was no real cost optimizations at all. it was an endless money funnel running things with tons of companies all around the country and that even being desired in some regards, because jobs and shit. and lots of nasa contracts back then and today still were actually decided against cost optimizations. you went over the cost? well you get paid for the extra, so you literally get paid to not reach the price you claimed for example. "cost-plus" contracts is the term used for this insanity. now the spaceshuttle was supposed to be a cheap and decently quickly reusable vehicle. IT DIDN'T TURN OUT TO BE THOUGH. \_\_\_ so i think you could have chosen a far far better comparison. and you are absolutely correct of course, that you first figure things out and THEN think about cost optimizations. does it work? is it safe? can we make it even safer? alright it is ok then let's cost optimize.
A “space shuttle” accurately describes spacecraft that shuttled humans into space during the first moon landing. Orbital shuttles like STS-1 being called “Space Shuttles” by NASA for marketing purposes is a misnomer.
I also think Musk having to admit he is wrong about camera only also negates Tesla's scalability and every customer turning their Tesla into a money printing machine. I don't think that many Tesla owners would want to have random strangers in their cars without supervision anyways. There is also the aesthetics of lidar and radar that he probably doesnt like.
It is no match for a thick cloud of high speed insect spatter.
Cleary, all the evangelists for FSD have never driven across the Great Plains on a summer evening.
LiDaR iS oBsOLeTe
Login is required to comment.
Login with Google