Tesla replacing ultrasonic sensors with Tesla Vision

[stopcrazypp]

Tesla says it is adding radar in its cars next month amid self-driving suite concerns | Electrek

Headline is extremely misleading (as always for a lot of media posts). Tesla never said they were adding radar to cars. They just requested to withhold documents from their FCC filing from being released to the public on the 12/7/2022 deadline, as they won't use the radar device until mid-January 2023. That doesn't necessarily mean it'll be for cars, it can be for the Semi also.

Other threads that discussed this:
Tesla Radar Speculation
Tesla bringing back RADAR?

 

[emmz0r]

Headline is extremely misleading (as always for a lot of media posts). Tesla never said they were adding radar to cars. They just requested to withhold documents from their FCC filing from being released to the public on the 12/7/2022 deadline, as they won't use the radar device until mid-January 2023. That doesn't necessarily mean it'll be for cars, it can be for the Semi also.

Other threads that discussed this:
Tesla Radar Speculation
Tesla bringing back RADAR?

why would you need it for semi if full self driving is good enough with cameras ? Also hasn't deliveries already started ?

 

[SalisburySam]

I’m one of the 0.5% of people (according to Teslafi) who are still holding strong on 2022.20.8. Kind of surprised it’s not a higher number with all of the drama around disabling radar.

I am another. V2022.20.8 is working quite well for me on my 2018 LR RWD Model 3, and I expect to remain on that version until subsequent ones reach operational parity (AP speed max, following distance, mandatory high beams/wipers, etc.). None of the newer releases deliver much that I care about, to be honest.

 

[stopcrazypp]

why would you need it for semi if full self driving is good enough with cameras ?

Semi has more potential blind spots, it is much larger and heavier and has significantly less margin for error. Plenty of reasons why you would likely want more sensors for it.

Also hasn't deliveries already started ?

Tesla has already started deliveries of the 2023 model year too for their other cars. That's not an indicator whether Tesla will add it in the future, as Tesla doesn't follow traditional release cycles. They may also install the hardware deactivated or they may offer retrofits (both of which they have done in the past for various features).

None of this however is particularly relevant to the point. The point is Tesla never said it was being used for cars. They may do so anyways in the future, but not at the moment. It's misleading reporting like this that leads to people saying Tesla "promised" something and broke their promise when they never made the promise in the first place.

 

[Boza]

Semi has more potential blind spots, it is much larger and heavier and has significantly less margin for error. Plenty of reasons why you would likely want more sensors for it.

Wasn’t the party line “More sensors -> more noise”? Therefore, the semi must have Tesla Vision - to reduce the errors.

 

[Eto Demerzel]

why would you need it for semi if full self driving is good enough with cameras ? Also hasn't deliveries already started ?

Wasn’t the party line “More sensors -> more noise”? Therefore, the semi must have Tesla Vision - to reduce the errors.

3/4 the length and overall dimensions of a loaded semi is not made by Tesla... which may or may not be relevant since it's entirely speculation

The technology could also be for HW4.

I don't understand why Tesla owners are so antagonistic and not willing to have an open discussion, especially when it's pretty much all conjecture.

 

[Battpower]

Tesla owners are so antagonistic and not willing to have an open discussion

I agree that reporting of issues like this potential new radar isn't always as straightforward as it could be. Media needs to make stuff interesting! So take it with a pinch of salt. I count myself as an owner who feels Tesla is too quick to distance itself from inconvenient commitments it made when the realities of delivering them maybe turn out to be harder than expected. I also think Tesla has made a rod for its back by selling ahead of stuff being available. How many people paid for FSD on finance and saw nothing for it?

I don't feel antagonistic but I do feel there is more to complain about than to be excited about. I want it to be different from that!

it's pretty much all conjecture.

It is most of the time especially when guessing what Tesla is planning. It is what it is.

But it is not conjecture that Radar was removed (disabled) from cars without a feature identical replacement. It isn't conjecture that people ordered cars with USS and got cars without. Likewise, fogging b-pillar cameras, problematic auto high beam and wipers. No sign of progress meeting regulatory requirements as a step towards delivering FSD in certain markets.

These do not look like positives for Tesla and if owners feel let down by some of this, I think that's understandable.

 

[WyoDude]

This is my biggest concern. Cameras fog up and get dirty, how will they prevent this from happening?

New acid sprayer system

 

[WyoDude]

I am another. V2022.20.8 is working quite well for me on my 2018 LR RWD Model 3, and I expect to remain on that version until subsequent ones reach operational parity (AP speed max, following distance, mandatory high beams/wipers, etc.). None of the newer releases deliver much that I care about, to be honest.

You don’t want the ability to make your car fart????

 

[stopcrazypp]

Wasn’t the party line “More sensors -> more noise”? Therefore, the semi must have Tesla Vision - to reduce the errors.

Article mentioned Elon's response to radar removal question that is relevant to your question:

Tesla CEO Elon Musk told Electrek in June of last year after Tesla stopped using its radar:

The probability of safety will be higher with pure vision than vision+radar, not lower. Vision has become so good that radar actually reduces signal/noise.

However, the CEO also added that Tesla might still use radar if it had a “very high-resolution radar”:

A very high resolution radar would be better than pure vision, but such a radar does not exist. I mean vision with high res radar would be better than pure vision.

 

[TarmacSurfer]

Why do aircraft use radio signals and ground radar altimetery during precision approaches? Because the visual cues are worthless, or aren’t there at all. Tesla at some point will realize this happens to them too. The visual data from the cameras alone simply won’t provide enough information in all situations. Low light, fog, snow, rain… these happen. Even something so simple as an old road with faded (or non-existent) pavement edge striping can pose challenges. Add darkness or some snow on the surfaces and it’s a very formidable challenge to simply figure out where the road edge is by using visual cues as a human being.

So how to make it better and allow the technology to exceed human limitations? Allow the machine to “see” more than a human can. That’s where things like USS or radar can help - a lot.

Tesla Vision isn’t bad per se. But it’s not the end game. Focusing on making it as good as possible is a good goal. Then add in supplemental information sources / sensory devices (like USS) to deal with the remaining gaps (like the above situations) where TV alone struggles. That will indeed be a robust and reliable system, but is also one that’s probably 3-5 years away. By then, at current rates of development, I suspect Tesla may very well have been left in the dust, particularly by Mercedes, who is very aggressively pursuing autonomous driving technology and having a lot of success at it. My money is on them to be the eventual leader, not Tesla. We shall see. I do still love my Tesla and what it does, and how the company has shaken up an industry that very badly needed it, but they are allowing themselves to fall behind because they overpromise rather than quietly just getting stuff done and then springing it on an unsuspecting industry (their model for success with such things as the original roadster, the model S and the supercharger network). My advice would be for them to stop talking about what they’re planning to do and instead just do it. Bam.

 

[TarmacSurfer]

You don’t want the ability to make your car fart????

EVERYONE wants the ability to make their car fart! My kids would be devastated if that feature ever went away!

 

[SalisburySam]

You don’t want the ability to make your car fart????

Oh my version has fart certainly…that’s been a prized “feature” in the car for many, many software versions prior to 2022.20.8. Perhaps the most useful of the totally not useful features in the vehicle.

 

[Mike7564]

Seems wrong they’re willing to sell you EAP for $6k on a new delivery now, yet most of the features are currently disabled until the “near future”. Shouldn’t EAP be discounted for the time being?

 

[JulienW]

Seems wrong they’re willing to sell you EAP for $6k on a new delivery now, yet most of the features are currently disabled until the “near future”. Shouldn’t EAP be discounted for the time being?

Well they are selling ALL standard 3/Y with features currently disabled. As I've stated the USS should only been removed POST software.

 

[sleepydoc]

I don't understand why Tesla owners are so antagonistic and not willing to have an open discussion, especially when it's pretty much all conjecture.

Except every one of your arguments has been conjecture and you haven’t bothered to actually refute anything beyond saying ’after listening to Andrei I have faith’ and ‘Tesla has done great things.’

Tesla has done great things and Andrei is a brilliant engineer but Tesla has also failed in the past and Andrei has also been wrong before so one can also use Tesla and Andrei’s history as an argument against success.

I’m more than willing to listen to a logical argument about why removing the sensors is a good idea but you need to have more of an argument than faith.

 

[Eto Demerzel]

Except every one of your arguments has been conjecture and you haven’t bothered to actually refute anything beyond saying ’after listening to Andrei I have faith’ and ‘Tesla has done great things.’

Tesla has done great things and Andrei is a brilliant engineer but Tesla has also failed in the past and Andrei has also been wrong before so one can also use Tesla and Andrei’s history as an argument against success.

I’m more than willing to listen to a logical argument about why removing the sensors is a good idea but you need to have more of an argument than faith.

Actually you said "i have faith" and you weirdly continue to say that.

Eto Demerzel said:
I don't understand why Tesla owners are so antagonistic and not willing to have an open discussion, especially when it's pretty much all conjecture.

Yeah, I said it's all conjecture- that would be including me. Which is why I don't understand the antagonistic stance.

I’m curious as to why you have so much faith that Tesla Vision will even come close to replacing ultrasonic sensors in the near future if ever.

"So much faith" is a mischaracterization of what I said.

The reason I keep an open mind about Tesla Vision is because I listened to the reasoning behind the decision by Karpathy.


What I did say is this:
However, I'm in a wait and see mindset with what they will come out with.
At this point, if I had the choice to permanently stay with uss or wait and go with tv, i'll just wait.
Well, honestly, I have no choice. LOL

 

[cantchoosename]

Why do aircraft use radio signals and ground radar altimetery during precision approaches? Because the visual cues are worthless, or aren’t there at all. Tesla at some point will realize this happens to them too. The visual data from the cameras alone simply won’t provide enough information in all situations. Low light, fog, snow, rain… these happen. Even something so simple as an old road with faded (or non-existent) pavement edge striping can pose challenges. Add darkness or some snow on the surfaces and it’s a very formidable challenge to simply figure out where the road edge is by using visual cues as a human being.

I don't think an autonomous car can really be compared to an aircraft in this sense. Airplanes line up the runway with things like VORs. This only gives them a straight line to the VOR and then they have a cleared runway to land on. No cars, pedestrians, or anything else to worry about other than getting the plane on the ground in a straight line. This can be done in low visibility. They even have air traffic controllers directing other planes away from each other and helping the planes taxi around the runways. Driving a car in very low visibility conditions really isn't possible (whether it's FSD or you driving). Radar can't follow a line in the road or detect traffic lights or stop signs. Just like if you can't see any of these things with your eyes, you can't (or shouldn't) be out driving. The cameras are required to make this work (with or without radar). Tesla's approach is that the cameras will work for FSD when cars are capable of being on the road safely.

 

[stopcrazypp]

Why do aircraft use radio signals and ground radar altimetery during precision approaches? Because the visual cues are worthless, or aren’t there at all. Tesla at some point will realize this happens to them too. The visual data from the cameras alone simply won’t provide enough information in all situations. Low light, fog, snow, rain… these happen. Even something so simple as an old road with faded (or non-existent) pavement edge striping can pose challenges. Add darkness or some snow on the surfaces and it’s a very formidable challenge to simply figure out where the road edge is by using visual cues as a human being.

This point was brought up a billion times. The issue you point out means the system can continue to run TACC in those conditions when equipped with a radar, but the system still relies on the camera to determine the road lines, so AP wouldn't be safe to remain active anyways. Basically using vision is unavoidable for AP, given so far none of the other sensors offer an alternative.

The roads today are not equipped for non-visual "instrumented" base driving when there is little or no visibility. There's ideas thrown about with having every vehicle having transponders (and even pedestrians) and also the roads and traffic signals being able to communicate, but so far that remains pie in the sky. Airplanes and airports however have all that, so the analogy doesn't work.

A lot of people wish for a more conventional ACC implementation (or even a dumb CC option), but it's clear Tesla doesn't prioritize that.

So how to make it better and allow the technology to exceed human limitations? Allow the machine to “see” more than a human can. That’s where things like USS or radar can help - a lot.

Tesla Vision isn’t bad per se. But it’s not the end game. Focusing on making it as good as possible is a good goal. Then add in supplemental information sources / sensory devices (like USS) to deal with the remaining gaps (like the above situations) where TV alone struggles. That will indeed be a robust and reliable system, but is also one that’s probably 3-5 years away. By then, at current rates of development, I suspect Tesla may very well have been left in the dust, particularly by Mercedes, who is very aggressively pursuing autonomous driving technology and having a lot of success at it. My money is on them to be the eventual leader, not Tesla. We shall see. I do still love my Tesla and what it does, and how the company has shaken up an industry that very badly needed it, but they are allowing themselves to fall behind because they overpromise rather than quietly just getting stuff done and then springing it on an unsuspecting industry (their model for success with such things as the original roadster, the model S and the supercharger network). My advice would be for them to stop talking about what they’re planning to do and instead just do it. Bam.

The idea Tesla had was that it is inevitable that Vision must be "solved" if they want to achieve general driving everywhere. That's why they are so focused on it. If it is ultimately unsolvable, then they may be going down a dead end road. Their goal wasn't like the traditional automakers, to release something like limited heavily geofenced L2/L3. If that was the goal, then loading up the car with other sensors is obviously the better approach.

 

[sleepydoc]

The reason I keep an open mind about Tesla Vision is because I listened to the reasoning behind the decision by Karpathy.

You listened to what Karpathy said and believe (i.e. have faith) it will work but have no evidence beyond his words.

I use the word ’faith’ because it’s accurate.

I have less of an open mind based on the reasons I listed above. Like I said, if you care to refute those points, please do so.