Tesla removed Ultrasonic sensors and replaces with Tesla Vision

[hydbob]

Tesla Vision Update: Replacing Ultrasonic Sensors with Tesla Vision

Safety is at the core of our design and engineering decisions. In 2021, we began our transition to Tesla Vision by removing radar from Model 3 and Model Y, followed by Model S and Model X in 2022. Today, in most regions around the globe, these vehicles now rely on Tesla Vision, our camera-based...

www.tesla.com

 

[borski]

I think it’s wild that they’re doubling down on vision only, when we’ve seen over and over again that it fails in the edge cases. Just wild.

 

[MoniputerLM]

I think it’s wild that they’re doubling down on vision only, when we’ve seen over and over again that it fails in the edge cases. Just wild.

Time will tell. Tesla does have sensor data collected from probably millions of hours of driving and super-computers to analyze it. As an engineer, I would always like a variety of sensors but who knows?

 

[borski]

Time will tell. Tesla does have sensor data collected from probably millions of hours of driving and super-computers to analyze it. As an engineer, I would always like a variety of sensors but who knows?

Speaking as an engineer, to me this is *very clearly* a cost-cutting measure. Tesla doesn’t even have cameras on their bumper, so anything the hood blocks is effectively invisible. Also, it still struggles with phantom braking regularly, and vision-only FSD is still not up to par with the previous autopilot implementations.

Lastly, if this were driven by engineering, they would have done it when the software caught up, which it hasn’t, and won’t. Instead, it’s “we are removing features you already have, and that other competitors have had and will have, and we… promise we’ll eventually have again, trust us.”

Tesla doesn’t have a good track record on this; they cheated out on the IR rain sensor, so now the wipers are schizo. Good luck controlling the wipers when you couple that with the touchscreen. No CarPlay (and not planned), no 360 view, now no ultrasonic sensors, and yet the price keeps increasing.

Basically when you compare them to competitors (not even just Lucid, but Hyundai, Kia, Ford, Rivian, etc.) they are now far behind in terms of feature set, planned or otherwise, banking on “please trust us we know what we’re doing we promise.” Good luck with that.

I would bet money (I guess I did, sort of, by buying a Lucid) that this is cost-driven and not engineering-driven.

Damn. And I really wanted an X.

 

[AS10]

Speaking as an engineer, to me this is *very clearly* a cost-cutting measure. Tesla doesn’t even have cameras on their bumper, so anything the hood blocks is effectively invisible. Also, it still struggles with phantom braking regularly, and vision-only FSD is still not up to par with the previous autopilot implementations.

Lastly, if this were driven by engineering, they would have done it when the software caught up, which it hasn’t, and won’t. Instead, it’s “we are removing features you already have, and that other competitors have had and will have, and we… promise we’ll eventually have again, trust us.”

Tesla doesn’t have a good track record on this; they cheated out on the IR rain sensor, so now the wipers are schizo. Good luck controlling the wipers when you couple that with the touchscreen. No CarPlay (and not planned), no 360 view, now no ultrasonic sensors, and yet the price keeps increasing.

Basically when you compare them to competitors (not even just Lucid, but Hyundai, Kia, Ford, Rivian, etc.) they are now far behind in terms of feature set, planned or otherwise, banking on “please trust us we know what we’re doing we promise.” Good luck with that.

I would bet money (I guess I did, sort of, by buying a Lucid) that this is cost-driven and not engineering-driven.

Damn. And I really wanted an X.

I much prefer the multisensory (Lidar plus computer vision plus other sensors) approach that the other companies are using, but I do think Tesla's computer vision based approach is probably going to be as good as you can hope for with that kind of system.

Karpathy's talk at CVPR last year had a lot of very interesting considerations for cases in which the two sensors do not agree with each other:

As an engineer working on multisensory machine learning myself, it was definitely worth the philosophical back and forth and even though I lean towards (and am buying a car with Lidar in the AGT), I still think FSD has its merits.

 

[AS10]

A (possibly cherry picked) case of where only computer vision might be better:

Again, I'd much prefer to use all available sensor data and build a model that can be at worst the best individual sensor's quality, but that in itself is a pretty hard research problem.

 

[joec]

The whole argument that humans only have eyes, therefore a vision-only computer system can be just as good is a bit bizarre to me.

We have ears. We have noses. We have nerve endings all over our bodies that constantly report to the brain data about our environment. And those different sensors often report conflicting data that the brain needs to interpret.

When we smell gas but don’t see it, our brains are smart enough to trust the nose over the eyes.

When we run over a pothole we didn’t see, our bodies feel it and believe it.

Not so with a Tesla. That’s why it happily runs full speed over speed bumps in my neighborhood.

Not only does the car only have eyes. Its eyes only see what the programmers at Tesla have trained it to see.

If there’s a giant fire five miles ahead of you, the Tesla will gladly drive right into it. It’s only looking at the road.

The problem with FSD is that it will never be a truly intelligent system. It doesn’t take in information from sensors and then “learn” how to react. It just collects data, which then gets piled in with everyone else’s data, which then gets filtered by humans, who then “train” the model, which then gets sent out to the car as “learning” a month later.

Jeff Hawkins is right about the AI community. They keep throwing more compute power and more “data” at the problem, when our brains are nowhere near as fast and yet can figure stuff out with far less training.

Tesla’s contention that vision would be enough would be correct, if their model were actually intelligent. If it actually learned on its own. As it stands, the only way to make it better is to give it more data, not less.

 

[RichMalden]

It makes no sense because it is about enhancing the ability, making it better and safer than we are not just as good. Using their logic they will replace the wheels with legs at some point

 

[zkhan2]

Tesla doesn’t even have cameras on their bumper, so anything the hood blocks is effectively invisible. Also, it still struggles with phantom braking regularly, and vision-only FSD is still not up to par with the previous autopilot implementations.

This is the most reoccurring issue with my coworkers Model 3 and beta FSD. No matter what updates they do to improve drivability, there are a number of turning and merging situations where the front camera is just out of reach and makes the car behave erratically right before it alarms for a take over. They will need a lot more cameras and probably two or more GPUs to keep up with all the data to make it work right.

 

[Drendino]

Wow! I don’t think Elon is just winging it, but I do think that his ego is driving this decision. He rejected LiDAR, which is some of the best technology on the market. If he chooses Lidar now, then he will be copying Lucid and would have to retrofit countless cars. I think he has backed himself into a corner and it will be interesting to see how he gets out.

In a few years, I believe that Lucid will be the champion of FSD, as our cars are loaded with the best hardware available. We just need the software to match. Let’s go software team!!

 

[joec]

As far as cost cutting goes, I mostly agree with @borski that this is a big motivator for the decision. But I do think it's more complicated than that.

At the end of the day, the bulk of Elon's fortune is tied to Tesla's stock price. And that stock price has been grossly overinflated for years, all because he has successfully sold Wall Street on the notion that FSD will be coming "any day now", and once it's perfected, Tesla will not only be able to get all their customers to buy it, it will enable all sorts of other billion-dollar businesses, such as robo taxis, and so on.

There's no math that gets Tesla even close to being worth its stock price by selling cars alone. Especially now that other EV makers are making inroads, and Tesla is no longer the only viable game in town. That's only going to get worse over time.

Cutting costs and increasing the profits on cars is essential to keeping Tesla growing. But they won't get there, no matter how many sensors they take out.

As Wall Street starts waking up to the fact that FSD is not coming anytime soon, the danger of a stock tank becomes very real. In the short term, they can get by with more revenue per sale, but that won't last long.

Selling Wall Street on Tesla's "unique" approach to the FSD problem will keep some entertained for a bit longer, however.

"Look, we can remove sensors and still outperform our competition. We're obviously the only company that is approaching this correctly." Sure makes it sound like no other automaker is going to be able to catch up anytime soon, though that's pure nonsense. And all the YouTubers out there proclaiming every update is a "huge improvement" when in reality these updates are incremental at best drives the point home further. It plays into the narrative of the genius founder perfectly.

And since that won't last forever, either, becasue the competition absolutely will catch up and surpass Tesla eventually, pivoting to C3PO-type robots that will revolutionize the factory is the next move. (In reality, these robots will likely do little more than what Amazon's robotics already do in its warehouses.)

But that's Elon. He can make a dumb robot look like a sci-fi character and convince Wall Street that it's just as important an innovation as FSD would have been. It's all smoke and mirrors to keep the stock overinflated.

When the robots fail, it'll be something else.

He's not Steve Jobs. He's P.T. Barnum.

 

[DJ@SJ]

Speaking as an engineer, to me this is *very clearly* a cost-cutting measure. Tesla doesn’t even have cameras on their bumper, so anything the hood blocks is effectively invisible. Also, it still struggles with phantom braking regularly, and vision-only FSD is still not up to par with the previous autopilot implementations.

Speaking as an engineer, I completely agree.

 

[hmp10]

. . . no 360 view . . .

The Lucid is the first car I've owned with the 360º bird's eye view. I never thought twice about not having it, but now that I do, I get almost angry every time I pull our Model S Plaid into a parking space while trying to guess where the lines and the concrete bump stop are.

And I really wanted an X.

Even with the update, Tesla didn't bother to put a center armrest or any kind of storage in the second row of the Model X, even though they finally did with the Model S update. And they call the Model X a family vehicle. Go figure.

 

[Drendino]

As far as cost cutting goes, I mostly agree with @borski that this is a big motivator for the decision. But I do think it's more complicated than that.

At the end of the day, the bulk of Elon's fortune is tied to Tesla's stock price. And that stock price has been grossly overinflated for years, all because he has successfully sold Wall Street on the notion that FSD will be coming "any day now", and once it's perfected, Tesla will not only be able to get all their customers to buy it, it will enable all sorts of other billion-dollar businesses, such as robo taxis, and so on.

There's no math that gets Tesla even close to being worth its stock price by selling cars alone. Especially now that other EV makers are making inroads, and Tesla is no longer the only viable game in town. That's only going to get worse over time.

Cutting costs and increasing the profits on cars is essential to keeping Tesla growing. But they won't get there, no matter how many sensors they take out.

As Wall Street starts waking up to the fact that FSD is not coming anytime soon, the danger of a stock tank becomes very real. In the short term, they can get by with more revenue per sale, but that won't last long.

Selling Wall Street on Tesla's "unique" approach to the FSD problem will keep some entertained for a bit longer, however.

"Look, we can remove sensors and still outperform our competition. We're obviously the only company that is approaching this correctly." Sure makes it sound like no other automaker is going to be able to catch up anytime soon, though that's pure nonsense. And all the YouTubers out there proclaiming every update is a "huge improvement" when in reality these updates are incremental at best drives the point home further. It plays into the narrative of the genius founder perfectly.

And since that won't last forever, either, becasue the competition absolutely will catch up and surpass Tesla eventually, pivoting to C3PO-type robots that will revolutionize the factory is the next move. (In reality, these robots will likely do little more than what Amazon's robotics already do in its warehouses.)

But that's Elon. He can make a dumb robot look like a sci-fi character and convince Wall Street that it's just as important an innovation as FSD would have been. It's all smoke and mirrors to keep the stock overinflated.

When the robots fail, it'll be something else.

He's not Steve Jobs. He's P.T. Barnum.

I’m with you until the PT Barnum part. That infers that he is a sort of snake oil salesman, selling illusion and tricks. I know that you are anti-Elon, but let’s not forget what he created, on earth and in space AND what he has done to change the war in Ukraine with Starkink. He has succeeded where governments and many other private companies have failed. I own zero Tesla products and don't agree with Elon on many things but I respect his accomplishments and don’t think he is PT Barnum. I also think he is making a bad decision removing radar, but what do any of us really know about the data that he is working with?

 

[borski]

I’m with you until the PT Barnum part. That infers that he is a sort of snake oil salesman, selling illusion and tricks. I know that you are anti-Elon, but let’s not forget what he created, on earth and in space AND what he has done to change the war in Ukraine with Starkink. He has succeeded where governments and many other private companies have failed. I own zero Tesla products and don't agree with Elon on many things but I respect his accomplishments and don’t think he is PT Barnum. I also think he is making a bad decision removing radar, but what do any of us really know about the data that he is working with?

PT Barnum wasn’t a snake oil salesman either; he was a well-respected politician, built a massive and well-liked circus business, spoke positively for the ratification of the 13th amendment, and was a well-respected author and debunker as well as changing advertising as we know it.

Elon has done amazing things, but largely through showmanship. People forget he didn’t found Tesla; he invested in it and eventually replaced the founders. SpaceX was a huge win, and a major bet, but the thing Musk has always been amazing at is PR and convincing people he is “an engineer’s engineer,” similar to the Steve Jobs aura. Both were/are amazing founders, but so was Barnum.

I don’t hate Musk or even dislike him; I think he’s a fantastic founder, even if I don’t think he’s the “world’s best engineer.” I quite like what SpaceX is doing. I disagree with Musk’s politics and theatrics, but that’s okay too.

Those are all distinctly irrelevant from thinking Tesla, and Musk, have lost their way here and made a major mistake that is focused on shoring up short-term profit margin at the expense of long-term technology advantage.

 

[joec]

I’m with you until the PT Barnum part. That infers that he is a sort of snake oil salesman, selling illusion and tricks. I know that you are anti-Elon, but let’s not forget what he created, on earth and in space AND what he has done to change the war in Ukraine with Starkink. He has succeeded where governments and many other private companies have failed. I own zero Tesla products and don't agree with Elon on many things but I respect his accomplishments and don’t think he is PT Barnum. I also think he is making a bad decision removing radar, but what do any of us really know about the data that he is working with?

There was a time when I found Elon fascinating and even somewhat respectable. I've met the man. I played music at a couple of Tesla events. (Including the IPO. Man, I was an idiot for not investing that day.) I know a few former Tesla employees who still have a lot of respect for him.

That time has long since passed for me.

Do good deeds in one's past excuse current unforgivable behavior? No. But that doesn't mean I begrudge him some of those good deeds.

Like most rich people who have had nothing but praise heaped on them for decades, whether that be from the media or his followers, Elon has let his ego grow far too big to be healthy. The results have been frightening.

Overinflated egos and ungodly sums of money seldom lead to anything good. The existence of SpaceX has led many to believe that he's somehow an exception to that rule. I am skeptical.

 

[Drendino]

PT Barnum wasn’t a snake oil salesman either; he was a well-respected politician, built a massive and well-liked circus business, spoke positively for the ratification of the 13th amendment, and was a well-respected author and debunker as well as changing advertising as we know it.

Elon has done amazing things, but largely through showmanship. People forget he didn’t found Tesla; he invested in it and eventually replaced the founders. SpaceX was a huge win, and a major bet, but the thing Musk has always been amazing at is PR and convincing people he is “an engineer’s engineer,” similar to the Steve Jobs aura. Both were/are amazing founders, but so was Barnum.

I don’t hate Musk or even dislike him; I think he’s a fantastic founder, even if I don’t think he’s the “world’s best engineer.” I quite like what SpaceX is doing. I disagree with Musk’s politics and theatrics, but that’s okay too.

Those are all distinctly irrelevant from thinking Tesla, and Musk, have lost their way here and made a major mistake that is focused on shoring up short-term profit margin at the expense of long-term technology advantage.

This is why I still enjoy this place. Both you and @joec always lay out a thoughtful point of view and sometimes disagree without being disagreeable. Very rare for an online forum.

I do agree that Tesla seems to be losing its way due to unreasonable valuations and the pressure to produce in mass quantities. The reason I didn’t buy a Tesla was because they didn’t offer a real luxury car, with amazing range, great performance and superior fit and finish. If they had, I would have bought one regardless of what Elon said on Twitter last week.

 

[hmp10]

. . . what he has done to change the war in Ukraine with Starkink.

Typo?

If you have any influence with Musk, I would appreciate your help in getting my deposit back from Starl[k]ink. I paid to sit on the waiting list for over a year during which I heard nothing from them. Then the press reported they were refocusing on commercial applications instead of residential applications. So I canceled my reservation and got an email saying my refund would be processed. That was over four months ago and, based on all the reports of difficulties in getting car deposits back as promised, I haven't bothered. Fooling with their Comcast-like customer service is just not worth my time.

 

[Drendino]

Typo?

If you have any influence with Musk, I would appreciate your help in getting my deposit back from Starl[k]ink. I paid to sit on the waiting list for over a year during which I heard nothing from them. Then the press reported they were refocusing on commercial applications instead of residential applications. So I canceled my reservation and got an email saying my refund would be processed. That was over four months ago and, based on all the reports of difficulties in getting car deposits back as promised, I haven't bothered. Fooling with their Comcast-like customer service is just not worth my time.

I have clumsy fingers, what can I say? Let’s go with Starkink though. Since we are on the subject of Tesla poor customer service, I’m going with a solar company who is charging 8k more than Tesla. I’m going with them so I don’t have to deal with horrible Tesla “customer service”, which is an oxymoron.

 

[hydbob]

Typo?

If you have any influence with Musk, I would appreciate your help in getting my deposit back from Starl[k]ink. I paid to sit on the waiting list for over a year during which I heard nothing from them. Then the press reported they were refocusing on commercial applications instead of residential applications. So I canceled my reservation and got an email saying my refund would be processed. That was over four months ago and, based on all the reports of difficulties in getting car deposits back as promised, I haven't bothered. Fooling with their Comcast-like customer service is just not worth my time.

Based on your post, I think Starkink is accurate. Because they are really giving you something....