Elon Musk admits full self-driving likely won't be possible on older Tesla vehicles

Skye Jacobs

Posts: 280   +7
Staff
Why it matters: Tesla's journey towards full self-driving capabilities continues, but the road ahead may be longer and more complex than initially anticipated. Issues with older vehicles and the continuous hardware upgrading required complicate Tesla's FSD roadmap.

During a conference call following Tesla's Q3 2024 financial results, CEO Elon Musk acknowledged that the company's older self-driving hardware may not achieve full autonomy. The admission comes after years of promises that all Tesla vehicles would soon be able to drive themselves.

We are not 100 percent sure. HW4[Tesla's latest iteration of self-driving computer hardware for its vehicles introduced last year] has several times the capability of HW3. It's easier to get things to work on HW4 and it takes a lot of efforts to squeeze that into HW3. There is some chance that HW3 does not achieve the safety level that allows for unsupervised FSD [Full Self Driving]

Some owners may see this statement as a slap in the face. For five years, Musk has consistently claimed that full self-driving was just around the corner for existing Tesla vehicles. Musk has floated the idea of free upgrades to owners who purchased the Full Self-Driving package with HW3-equipped vehicles to stem potential outrage. It is unclear how much this will cost the company.

However, implementing these upgrades may not be as straightforward as it sounds. The HW4 system uses different power and camera harnesses, and the computer has a different form factor. Retrofitting existing vehicles with HW4 could be complex and costly.

Additionally, while Musk describes the cameras in HW3 vehicles as "capable," the HW4 cameras offer significantly higher resolution and improved low-light performance. It's unclear whether these camera upgrades would also be necessary to achieve full self-driving capabilities. Tesla's cars are already under an NHTSA investigation over a fatal crash involving low visibility.

Even if Tesla manages to overcome these technical hurdles, questions remain about the viability of full self-driving technology. Musk admitted that he is unsure whether HW3 can eventually achieve FSD functionality. Although he remains confident that HW4 can attain FSD capability, the future is uncertain due to issues that extend beyond the system's advancement.

The computational demands of FSD systems are immense. Tesla has had to upgrade hardware to meet the software's requirements continuously. Couple that with the fact that the vehicles must achieve a safety level that satisfies regulators and the public. Tesla has faced numerous investigations and recalls relating to its Autopilot and FSD systems.

Many experts have doubts about FSD as well. William Stein, a technology analyst at Truist Securities, recently test-drove a Tesla equipped with an FSD system. Each time Stein was in the car, he reported it made unsafe or illegal maneuvers. Once, the vehicle turned left from a through lane while running a red light.

Approximately 500,000 Tesla owners have activated the FSD feature on their vehicles for use on public roads, or just over 20 percent of all Teslas currently in operation. Tesla has been clear that, despite its name, the FSD system does not make vehicles fully autonomous. The company emphasizes that drivers must remain alert and prepared to take control at any moment. To enforce this, Tesla has implemented a monitoring system that tracks driver behavior and can revoke access to FSD if users fail to maintain adequate supervision.

Permalink to story:

 
"Full Self Driving (Human Driver Required)"
"All Cars Supported (HW4 Required)"

But maybe HW3 owners shouldn't fret - HW4 may not get there either.

Tesla has often had great products. Maybe it'd be better to stick with the facts than embellish further with the marketing.
 
You kinda have to look at Tesla the way you look at Apple.

The initial product is interesting but probably suffers from serious flaws and is missing key features.

Wait till the 2nd Gen: you get what you should have gotten initially.

Wait till the 3rd gen and it's close to perfect.

Full Self Driving is problematic as there are countless situations the computer can't provide for - and that's ultimately the reason Cybercabs aren't gonna work off closed courses - especially when there's no steering wheel or pedals to intervene. Even Johnny Cab had a steering wheel and pedals.

Trying to run FSD and Autopilot on just cameras is a huge mistake. A better move would have been cameras, ultrasound and even LiDAR layered atop each other.
 
I dont blame him. Everything evolved so far since he started making those.
Chips, detectors and sensors.
I think, 5 years is a reasonable time to expect greatly improved updates for a self driving car.
Then just bug fixes and security patches.
 
Some owners may see this statement as a slap in the face. For five years, Musk has consistently claimed that full self-driving was just around the corner for existing Tesla vehicles.
Listen, you can no longer cite opinions from Electrek as fact regarding Tesla. They are a very biased news source concerning Tesla and they've even admitted this. They have made almost every recent Tesla headline sound like bad news, and they make uncited claims in their reporting (ie. that Tesla opened up their supercharger network because they would be subsidized if they did so).
Electrek even previously reported something similar to this news, but their immediate conclusion was that Tesla will not retrofit HW4 into HW3 vehicles (obviously wrong). Electrek headlines the article with accusations that the company could be committing fraud. The crazy thing about Electrek's reporting is they specifically mention that Tesla has previously paid to retrofit HW2 vehicles with newer hardware free of cost for those who bought FSD in advance. Apparently to Electrek, Tesla committing fraud is a more realistic situation: https://electrek.co/2024/10/15/tesla-needs-to-come-clean-about-hw3-before-the-word-fraud-comes-out/

Here is a source that would be incredibly useful to cite, an actual industry teardown expert in automobiles. He made comments regarding the feasibility of a HW4 retrofit in this interview. He owns no Tesla stock (this is mentioned in the interview) and is paid by competitors for accurate analyses:
 
Trying to run FSD and Autopilot on just cameras is a huge mistake. A better move would have been cameras, ultrasound and even LiDAR layered atop each other.
Despite what you say, I can easily drive without laser, radar, and ultrasonic transmitters/receivers. It may seem like a huge mistake, but Tesla is trying to take the cake here. Competitors have been trying to do exactly what you suggest and they've been stuck for years. They're completely unprofitable because they rely on detailed maps of roads in order to drive on them: https://waymo.com/waymo-driver/
 
Maybe it'd be a good idea to think about upgrading the HW3 just for a small fee only to cover the manufacturing cost. But instead of the HW4, it might be worth looking into a new, even better HW5 with better and bigger camera sensors (because there is a lack of radar and lidar sensors), which could also have an auto self-cleaning mechanism for the camera. That could be a useful addition, especially since the lens tends to catch dust. Safety is important because if the car has safety issues, those traveling with their families won't be comfortable and they will not buy it. It’s not a big deal to change few cameras and a processing unit. It would be reasonable to assume that other car companies without that advantage would be keen to maximize even the slightest issue. The system at it's ideal state should be fast, responsive and safe under all conditions, and the driver should be able to take a nap without worrying.
 
Last edited:
As a Tesla HW3 owner I think the cameras would need a built in Lidar or similar to achieve the level of 360 degree sensor array you’ll need for FSD - which means HW4 won’t be good enough - and most likely not HW5 either.
There’s alot of talk around FSD - as if that is the holy grail, but personally that is not really what I’m looking for in a car. I love what I have - with an easy OS filled with features, great charging infrastructure, camping mode etc. …The actual car I’m capable of driving and parking myself
 
Other than using "cruise control" on my vehicle, I wouldn't want full self drive. I'm one of those old farts that ENJOYS driving a car. Yeah, having it drive down an interstate highway is one thing, but I'm one of those that love driving the back roads, hills & curves.
 
I dont blame him. Everything evolved so far since he started making those.
Chips, detectors and sensors.
I think, 5 years is a reasonable time to expect greatly improved updates for a self driving car.
Then just bug fixes and security patches.
The one thing that hasn't evolved is computers ability to emulate a human and drive a car, so where do you stand on Turings halting problem?
 
Listen, you can no longer cite opinions from Electrek as fact regarding Tesla. They are a very biased news source concerning Tesla and they've even admitted this. They have made almost every recent Tesla headline sound like bad news, and they make uncited claims in their reporting (ie. that Tesla opened up their supercharger network because they would be subsidized if they did so).
Electrek even previously reported something similar to this news, but their immediate conclusion was that Tesla will not retrofit HW4 into HW3 vehicles (obviously wrong). Electrek headlines the article with accusations that the company could be committing fraud. The crazy thing about Electrek's reporting is they specifically mention that Tesla has previously paid to retrofit HW2 vehicles with newer hardware free of cost for those who bought FSD in advance. Apparently to Electrek, Tesla committing fraud is a more realistic situation:

Here is a source that would be incredibly useful to cite, an actual industry teardown expert in automobiles. He made comments regarding the feasibility of a HW4 retrofit in this interview. He owns no Tesla stock (this is mentioned in the interview) and is paid by competitors for accurate analyses:
Elon has been promising FSD will be ready in 12-18 months for years. This isn't from some source you don't like it's just what Elon has been doing: over-promising and under-delivering. If no one has signed a contract where Tesla said they will upgrade cars to the new hardware standard the promises Elon is making doesn't mean anything. Elon's words don't mean anything.

"And my personal guess is that we’ll achieve Full Self-Driving this year, yes, with data safety level significantly greater than present."

Do you know when Elon said the above quote?
 
Honestly they should just go with thermal sensors in addition to the optical cameras. It would have likely prevented all the FSD crashes that have happened so far. And thermal doesn't cost as much as LIDAR.

That said Elon has always made big promises and simply never delivers. Take it with a grain of salt like always.
 
Honestly they should just go with thermal sensors in addition to the optical cameras. It would have likely prevented all the FSD crashes that have happened so far. And thermal doesn't cost as much as LIDAR.

That said Elon has always made big promises and simply never delivers. Take it with a grain of salt like always.
You probably don't know how cameras work, but most cameras CAN pick up infrared wavelengths (which you are probably referring to by "thermal sensors"). But most cameras include a color filter (RGGB) that only allows wavelengths of light pass through that are typically seen by the human eye (the human eye is more sensitive to green, so there are two pixel sensors that measure light through green filters).

Tesla's cameras use RCCB which is the standard for automotive purposes. It increases low light sensitivity dramatically and will pick up light from infrared wavelengths because there is no color filter for 2/4 of the pixel sensors (hence C for clear).

The problem for Tesla FSD is not low light hardware, it is software/AI training. This is why FSD keeps improving significantly.
 
Elon has been promising FSD will be ready in 12-18 months for years. This isn't from some source you don't like it's just what Elon has been doing: over-promising and under-delivering. If no one has signed a contract where Tesla said they will upgrade cars to the new hardware standard the promises Elon is making doesn't mean anything. Elon's words don't mean anything.

"And my personal guess is that we’ll achieve Full Self-Driving this year, yes, with data safety level significantly greater than present."

Do you know when Elon said the above quote?
Guess vs. Promise. Maybe you should look up the difference between those? Please show me a "promise" that FSD would come. You're never going to find it.

And Tesla hasn't been under-delivering. They've just been late to Elon's predictions. The vast majority of his predictions have come true eventually, and he has admitted that his timelines are optimistic. If you listen to quarterly calls, you will hear him answer plenty of questions to when he "thinks" Tesla will release/achieve something. If you listen to really old quarterly calls, basically everything he's predicted eventually came true. The main exception is Tesla Solar.
 
You probably don't know how cameras work, but most cameras CAN pick up infrared wavelengths (which you are probably referring to by "thermal sensors"). But most cameras include a color filter (RGGB) that only allows wavelengths of light pass through that are typically seen by the human eye (the human eye is more sensitive to green, so there are two pixel sensors that measure light through green filters).

Tesla's cameras use RCCB which is the standard for automotive purposes. It increases low light sensitivity dramatically and will pick up light from infrared wavelengths because there is no color filter for 2/4 of the pixel sensors (hence C for clear).

The problem for Tesla FSD is not low light hardware, it is software/AI training. This is why FSD keeps improving significantly.
You didn't fall for the AI training scam did you. People who don't understand what a neural network is should know that Musk doesn't understand them either. If you think a singularity is a real thing you should stop reading the comic books, and learn a bit of science.
The rest of the world is coming to the unfair conclusion that all Americans are *****s.
 
Guess vs. Promise. Maybe you should look up the difference between those?
A person who has no knowledge of a company's progress on FSD wouldn't have any idea how far away it was, but the CEO would know exactly where they are on FSD progress. So when the CEO says he thinks FSD will be this year and it still hasn't happened 4 years later did he really think it was going to be that year or, as a CEO who goes to meeting with PMs to discuss project progression, did he just lie about his guess since they were nowhere close to it and haven't been for the 3 years before that.

He's said the price of the cyber truck was going to be a specific number, have a specific range and be ready by a specific month. None of those things happened. The cyber truck was late, much higher priced than it was announced and had a much shorter range. That was a under delivery. Why do you keep saying "predictions" He's the CEO he should know how long engineers say it will take and if an engineer says "we have no idea how long it will take to fix this issue" I wouldn't estimate a date, price, or performance until the vehicle designed and production has been figured out.

So just like I said you can't trust anything Elon says. He's shown he makes outlandish claims about their ability to make cares that hit specific milestones of price, availability and performance.
 
A person who has no knowledge of a company's progress on FSD wouldn't have any idea how far away it was, but the CEO would know exactly where they are on FSD progress. So when the CEO says he thinks FSD will be this year and it still hasn't happened 4 years later did he really think it was going to be that year or, as a CEO who goes to meeting with PMs to discuss project progression, did he just lie about his guess since they were nowhere close to it and haven't been for the 3 years before that.

He's said the price of the cyber truck was going to be a specific number, have a specific range and be ready by a specific month. None of those things happened. The cyber truck was late, much higher priced than it was announced and had a much shorter range. That was a under delivery. Why do you keep saying "predictions" He's the CEO he should know how long engineers say it will take and if an engineer says "we have no idea how long it will take to fix this issue" I wouldn't estimate a date, price, or performance until the vehicle designed and production has been figured out.

So just like I said you can't trust anything Elon says. He's shown he makes outlandish claims about their ability to make cares that hit specific milestones of price, availability and performance.
You’ve probably never worked a day in your life in R&D, huh?

No one has completed FSD. So no one actually knows how long it takes. The Cybertruck has been out less than a year, and just like all Teslas they’ll come out with more trims, and make it far more cheaply. So they’re not done yet. It doesn’t mean it’s a lie.

But you seem to only point out the negative. Instead, Tesla is the most productive and innovative car company to exist. All of its competitors have been scrambling to copy and get ahead of it, but because they cannot keep up they’re losing massive amounts of money in the equivalent ventures.

Here are some examples:
- The Chevy Bolt - “Ahead” of the Model 3 by a year but it had serious battery issues, a cheap design that no one wanted, and ended up going extinct
- Hummer EV - Also “ahead” in release date, but overweight to try meeting specs, far overpriced, ugliest Hummer design, and it can barely sell
- All other EVs - Extremely unprofitable, unable to sell much at discounted prices, not very safe, no mass market appeal

You complain about Elon Musk leading Tesla, but it’s going to make all its competitors go out of business. Elon is even trying to avoid this by creating standards from its tech to spoon feed its competitors.

In terms of manufacturing, design, costs, capability, efficiency, and software, no carmaker is innovative in any of these except maybe Rivian. But you’re concerned about Elon Musk’s timelines when everything done by his companies is uncharted?

Anyways, I’ve been investing in Elon’s “bad” timelines for years. I care about the results. Maybe you should short the stock waiting for it to go bankrupt and see how badly that works for you.
 
You’ve probably never worked a day in your life in R&D, huh?

No one has completed FSD. So no one actually knows how long it takes. The Cybertruck has been out less than a year, and just like all Teslas they’ll come out with more trims, and make it far more cheaply. So they’re not done yet. It doesn’t mean it’s a lie.

But you seem to only point out the negative. Instead, Tesla is the most productive and innovative car company to exist. All of its competitors have been scrambling to copy and get ahead of it, but because they cannot keep up they’re losing massive amounts of money in the equivalent ventures.

Here are some examples:
- The Chevy Bolt - “Ahead” of the Model 3 by a year but it had serious battery issues, a cheap design that no one wanted, and ended up going extinct
- Hummer EV - Also “ahead” in release date, but overweight to try meeting specs, far overpriced, ugliest Hummer design, and it can barely sell
- All other EVs - Extremely unprofitable, unable to sell much at discounted prices, not very safe, no mass market appeal

You complain about Elon Musk leading Tesla, but it’s going to make all its competitors go out of business. Elon is even trying to avoid this by creating standards from its tech to spoon feed its competitors.

In terms of manufacturing, design, costs, capability, efficiency, and software, no carmaker is innovative in any of these except maybe Rivian. But you’re concerned about Elon Musk’s timelines when everything done by his companies is uncharted?

Anyways, I’ve been investing in Elon’s “bad” timelines for years. I care about the results. Maybe you should short the stock waiting for it to go bankrupt and see how badly that works for you.
You would be better off investing in my time machine. All I need is a trillion dollars and 20 years.
 
You kinda have to look at Tesla the way you look at Apple.

The initial product is interesting but probably suffers from serious flaws and is missing key features.

Wait till the 2nd Gen: you get what you should have gotten initially.

Wait till the 3rd gen and it's close to perfect.

Full Self Driving is problematic as there are countless situations the computer can't provide for - and that's ultimately the reason Cybercabs aren't gonna work off closed courses - especially when there's no steering wheel or pedals to intervene. Even Johnny Cab had a steering wheel and pedals.

Trying to run FSD and Autopilot on just cameras is a huge mistake. A better move would have been cameras, ultrasound and even LiDAR layered atop each other.
Have you considered that the engineers at Tesla have considered your concerns tho?
 
Back