Tesla replacing ultrasonic sensors with Tesla Vision

[smogne41]

Something is going on with your car. I can leave my 3 for weeks and barely lose any charge.

My car is fine, I loose a mile of range per day or less just parked. Assuming I don't wake it up and be very sure everything that needs to be is turned off. Which is still a mile more per day lost then I would like (and a mile more per day than many other EVs loose). But sentry will eat a mile of range per hour. That would make leaving the car unplugged for a week or two very inconvenient or impossible.

 

[OxBrew]

Are you concerned for recent buyers that they won't have functionality for some short amount of time or that you believe this functionality won't be available for a long time or ever?

Or maybe you think the new functionality will be too complex as Occupancy network would be able to show 3D proximity of nearby objects instead of a few colored arcs and distance numbers?

I'm joking, 1. But 2, probably not buying another Tesla.

3, jeez, if the new functionality will be so freaking mind blowingly awesome, why not actually develop it first, and then remove the sensors?

Instead of this thread, you'd have nearly everyone begging to to disable their sensors and get on the new software.

My hunch, and knowing from several friends who actually worked on the factory floor elbow to elbow with Elon pulling long overnights on the weekends to get issues solved, is that there is no other way to truely prioritize a project. If it's not a pants on fire emergency, with the scattered chaotic way Tesla project management works, it won't get the attention it needs.

 

[Schweppes]

Good luck to the new owners who aren't even going to get park assist with their new cars. I would not be happy paying current market prices and not getting something that's available on basic cars.

This is a big deal to me and the only reason I'm not cancelling my order is my work is covering my payment and insurance. If they bungle this the car will be sold the week that agreement changes.

 

[LorentzTorque]

See my post above about parking outside close to a bollard. I'll need a SKYHOOK to mount a tennis ball.

I'm on my way out to hang one from the power lines. Wish me luck!

 

[kevin1]

So much crap in this thread.
I'm attaching a photo from today of my Vision Only Model 3 after manually parking

Did you mean your Model 3 has no radar? Afaict, they haven’t actually delivered any non-USS vehicles to customers yet.

 

[m25nyc]

I find the ultrasonic sensors presently in the Tesla 2022 Model Y with visualization and number of inches distance are exceedingly helpful for me in manual parking in tight spaces, parking garages and the like, all situations where autopark does not come close to even recognizing a spot at the moment.

Tesla has quite poor visibility to the rear and front bumpers from my driver's actual view, rear view mirror is virtually useless.

It is not providing a 360 degree camera view visualization and in other vehicles I've driven that do have 360 camera view, that is not as useful to me in manual parking as front and rear ultrasonic sensors at the bumpers which are the outermost perimeter of the vehicle.

Thus, for me, the loss of ultrasonic sensors would be dreadful!

In addition, in metro areas where I drive there a great many concrete road construction barriers for miles at a time! FSD beta drives quite well between them, often only a few inches away: I find the ultrasonic sensors in FSD beta give me a comforting reassurance that the car is not steering too close to a concrete barrier.

If Tesla engineers are reading this, please take a trip out of Texas and California where parking spaces are wide, and park manually sometime in Boston, Paris, New York, London, Shanghai, Seoul!! You will find the precision of the ultrasonic sensors and visualization map objectively are essential in not hitting commercial parking garage walls and columns, or in parallel street parking other cars.

 

[Steerpike]

It strikes me that removing the USS comes from the same mindset that brought us touch screen for everything, dysfunctional automatic wipers, and removed the radar, a purist minimalist perspective that truly believes that everything can be solved by Tesla Vision.

I, on the other had, see auto wipers that are worse than any other car that I have owned in decades, and cameras that get dirty and wet and warnings of multiple cameras blocked or blinded.

I have my second Tesla but I am now seriously doubting that I will ever buy another Tesla.

And the windows won't auto close reliably.

 

[mongo]

In addition, in metro areas where I drive there a great many concrete road construction barriers for miles at a time! FSD beta drives quite well between them, often only a few inches away: I find the ultrasonic sensors in FSD beta give me a comforting reassurance that the car is not steering too close to a concrete barrier.

The USS are not active/ effective at highway speeds. Repeater cameras have a great view of vehicle to barrier spacing (at least for rear 3/4 of car).
Also, given the shape of construction barriers, the USS would give overly generous readings.

3 year old video with some of the cameras:

 

[dan031970]

This is pretty disappointing. It makes no sense to remove the sensors before the vision based features are sufficiently developed to provide the same function as the ultrasonic sensors do. There are only 2 reasons that I can see to remove the sensors now:

1. Inadequate supply of the sensors
2. Cost reduction

Simply no other valid reasons exist to remove the sensors before the vision based system is ready to function as well without the sensors. Customers have to live without functionality they thought they were paying for until the vision based system performs adequately.

Other automakers would be unlikely to do this. I own 1 Tesla, but I wouldn't buy another as I dealt with poor TACC performance for months before it got better. I want all of the features on my car to work I day I take delivery. I don't care about FSD development.

 

[dansev]

Good Lord! I simply cannot fathom trying to park without the sensors. I use them every single day. Where I live, every parking inch counts.

And the auto park is a complete joke. If we’re all going to have to rely on that, I assume Tesla will pay the bill when it screws up? Or will the feature be in beta in perpetuity like everything else? Right now, I would never, ever use it in a chaotic, parking garage situation, for example.

 

[MP3Mike]

This is pretty disappointing. It makes no sense to remove the sensors before the vision based features are sufficiently developed to provide the same function as the ultrasonic sensors do. There are only 2 reasons that I can see to remove the sensors now:

1. Inadequate supply of the sensors
2. Cost reduction

Simply no other valid reasons exist to remove the sensors before the vision based system is ready to function as well without the sensors. Customers have to live without functionality they thought they were paying for until the vision based system performs adequately.

It is probably a matter of they thought they would have the software done, Elon did mention that related changes had a end of September due date, and terminated the contract with their suppliers for the USS parts effective at the end of September. (They probably did that months ago.)

You can say they shouldn't terminate the contract until they have it working, but that means they would have to purchase hundreds of thousands of USS parts that they don't want/need. (Or pay a penalty for not buying them.)

Hopefully they are really close to completing the software, so the functionality won't be missing for long.

 

[2101Guy]

Here is how Chevy announces how the chip shortage impacts their use of parking sensors:

A Host Of New Chevy Models Won’t Offer Rear Park Assist Due To Chip Shortage​

 

[texas_star_TM3]

Here is how Chevy announces how the chip shortage impacts their use of parking sensors:

A Host Of New Chevy Models Won’t Offer Rear Park Assist Due To Chip Shortage​

at least they are honest about it... and not "Tesla vision is superior". Cutting the radar and now parking sensors is 100% about mitigating part shortages. nothing more, nothing less.

 

[Nyc_rock]

I have read this in a couple of places;

After cutting radar, Tesla now dropping ultrasonic sensors from its EVs

Most automakers use multiple redundant sensors. Tesla thinks otherwise.

arstechnica.com

this would seem like a big change and Im not really sure how the camera’s would even work. If the Tesla had a front bumper mounted camera and offered 360 degree views, then maybe. I generally give Elon the benefit of the doubt, but these seems crazy.

 

[carlos31820]

Yeah, the sides and rear of the car would probably work fine since those cameras are down low. However, it seems like the front bumper area would be vulnerable to impacting something low that doesn't extend over the car hood for the windshield camera to see it.

 

[dramsey]

The USS are not active/ effective at highway speeds...

Actually, they are, at least on my 2016 Model S.

I frequently drive along Highway 80 coming out of Reno towards the Bay Area. A large part of this is two lanes in each direction with a concrete barrier separating the opposing lanes. The driver's side ultrasonic sensors clearly detect the barrier, even at 70+ mph speeds, as indicated by the yellow radiating lines on my center display. Dunno how accurate the distance data would be, though...

I don't see how Tesla Vision could accurately gauge the distance to a featureless garage wall, but maybe Elon does. Of course, my not-terribly-good auto wipers are still a "beta" feature after many years, so maybe Elon doesn't...

 

[Z_Lynx]

I read this too. I depend on the front sensors to park my car precisely in carport so driver side doors don't hit pillar when opened. I hope cars with existing sensors will still be able to use them.

 

[2101Guy]

How Cadillac deals with it: openly, honestly, in a non deceptive manner

Cadillac Limits Park Assist Availability Due To Chip Shortage: Report​

Hey Elon: telling your customers basic truth is ok. Certain things aren’t in your control. We get it. No need to deceive

 

[nvx1977]

To your last point, the video we get to see from the cameras doesn’t look that great, and I know they do all sorts of pre-processing before sending the camera feed to the NNs. I’m really curious what the feed looks like in the various stages.

I thought Tesla recently (few months ago) moved away from training off post-processed imagery and instead are now training directly from the raw camera data. I never dug into the details, but it would seem to have similar advantages of working with photographic RAW data instead of a heavy lossy, post-processed JPG. There was also some compute efficiencies involved, since there was no need to convert the raw data to an image (a data construct needed by humans) first, and training off imagery required more compute than just dealing with the raw pixel data.

Also, I recall that even in the days when Tesla was training off imagery, those images were much more high fidelity than the crap that's shown to us via dashcam/sentry.

 

[Undecided_2]

I don't see how Tesla Vision could accurately gauge the distance to a featureless garage wall, but maybe Elon does. Of course, my not-terribly-good auto wipers are still a "beta" feature after many years, so maybe Elon doesn't...

It is possible but I don’t think with my MYLR cameras it’s possible. They’re typically obscured by sunlight, condensation or when it’s dark, they’ll all offline! With the latest software update I now have the emergency braking disabled at night on all the time - this message fights attention with the cameras being obscured.

Auto high beams and auto wipers are hysterical.