2023 Model Y no USS discussion, incl. Discussion of Tesla Vision Firmware

[ewoodrick]

Re-pairing did not work.
I deleted the entire phone connection on the touchscreen. Make sure you have the car key when doing this to avoid being locked out.

There are two Bluetooth connections from the phone to the car. Only the classic connection appears in the Bluetooth settings. The car key is relatively invisible, it uses BLE.
When repairing a phone, it should be expected that all existing references to the existing pairing is removed. Otherwise it's not repairing, it's reconnecting.
Because it is just the audio connection that you should be impacting (not the key) then the card key really shouldn't be needed. They are two very different things.

General example is that when the Delivery Advisor first got you in the car, they should have setup the phone key for you. Dependent on the advisor, they probably didn't setup the normal phone audio connection,

Yes, if repairing is recommended, you should always remove all vestiges of the other device from each device. This is important in ANY Bluetooth pairing. (Headphones, computers, etc)

 

[L1TR]

I assume you address #1 in my post. The car is moving to the garage with 3 cameras in front and 4 cameras on the side. It's not a single camera. You think engineers are stupid? And politicians or influencers are smart?

Didn't know about triple camera in front - ok, theoretically it may predict a distance in the front, if at least two cameras are looking in the same direction (you know, like Nintendo 3DS worked with "3D photos" , still out-dated technology in comparing with LiDAR in iPhone, but at least something). Back cameras are located too far from each other to be used in correct distance prediction. No chance this "USS replacement" will work as someone may expect - I recommend to forget about this model Y generation if you care about USS and park assist till double cameras added to the back, or another modern alternative to calculate a distance. Take a look on 2022 models or wait a proper hardware solution installed. This park assist will never fixed by the air.

 

[rlsd]

There are two Bluetooth connections from the phone to the car. Only the classic connection appears in the Bluetooth settings. The car key is relatively invisible, it uses BLE.
When repairing a phone, it should be expected that all existing references to the existing pairing is removed. Otherwise it's not repairing, it's reconnecting.
Because it is just the audio connection that you should be impacting (not the key) then the card key really shouldn't be needed. They are two very different things.

General example is that when the Delivery Advisor first got you in the car, they should have setup the phone key for you. Dependent on the advisor, they probably didn't setup the normal phone audio connection,

Yes, if repairing is recommended, you should always remove all vestiges of the other device from each device. This is important in ANY Bluetooth pairing. (Headphones, computers, etc)

Interesting to know about 2 bluetooth connections.
The card key is needed in case you remove the phone but have not re-added yet and you get out of the car. Also, if I remember correctly, when I removed the phone and re-added, it asked me to swipe the card key.

Delivery advisor is not really dependable. My phone did not work the first time. The advisor removed the phone and re-added then it worked. Also when you have home delivery there is no advisor there.

 

[rlsd]

Didn't know about triple camera in front - ok, theoretically it may predict a distance in the front, if at least two cameras are looking in the same direction (you know, like Nintendo 3DS worked with "3D photos" , still out-dated technology in comparing with LiDAR in iPhone, but at least something). Back cameras are located too far from each other to be used in correct distance prediction. No chance this "USS replacement" will work as someone may expect - I recommend to forget about this model Y generation if you care about USS and park assist till double cameras added to the back, or another modern alternative to calculate a distance. Take a look on 2022 models or wait a proper hardware solution installed. This park assist will never fixed by the air.

You gave opinion without knowing about the car. The side cameras can measure the distance too (distance from front and rear of the car).

 

[EVRob78]

If it’s anything like what happens when entering my garage - it starts beeping/warning like crazy and telling you to stop for no good reason because either it thinks there’s an object when there isn’t, the tolerances aren't honed enough (I only have about 4-6 inches on either side of the mirrors when entering the garage bay) - or they did no QA testing at all for garage entry and sloped entrances or exits with this version of the software.

Just like my USS-equipped non-Tesla. I ignore all the beeping and lights, and use cameras/mirrors. Tesla, with more cameras, is better at this. After a few tries backing a new Tesla into the garage, without USS or the vision system software, its easy to judge and get the right result.

We humans are losing our ability to do things for ourselves. I love all the tech, but at least periodically go manual, and don't lose your hand/eye/machine control skills. No inference to anyone in this forum - but in general, there seems to be a direct correlation between those that rely excessively on technology and a general lack of self-confidence.

 

[ewoodrick]

Interesting to know about 2 bluetooth connections.
The card key is needed in case you remove the phone but have not re-added yet and you get out of the car. Also, if I remember correctly, when I removed the phone and re-added, it asked me to swipe the card key.

Delivery advisor is not really dependable. My phone did not work the first time. The advisor removed the phone and re-added then it worked. Also when you have home delivery there is no advisor there.

Assuming that you didn't go into the Tesla app and have to add the car as a key afterwards, the phone was still a key.
Keys show up under settings - keys, not Bluetooth

 

[stopcrazypp]

You gave opinion without knowing about the car. The side cameras can measure the distance too (distance from front and rear of the car).

If you want to understand how it works, it is based on the occupancy network, which stitches all the cameras:

A Look at Tesla's Occupancy Networks

In 2022, Tesla announced a brand new algorithm about to be released in their vehicles. This algorithm is named occupancy networks, and it is supposed to improve the HydraNets from Tesla. But how does it work? Why is it needed? Let's find out... Why a new kind of algorithm is

www.thinkautonomous.ai

Given the blind spot in the front below the front bumper, the front is based on persistence of the occupancy network results as you pull in. However, accurate front detection is still dependent on your wall in front having some features, so the AI has cues to estimate the depth. If it's a completely featureless white wall (and there is no clear line that separates floor from wall either) I think it'll be fairly difficult for the AI to figure things out. The side cameras won't really help that much.

 

[rlsd]

If you want to understand how it works, it is based on the occupancy network, which stitches all the cameras:

A Look at Tesla's Occupancy Networks

In 2022, Tesla announced a brand new algorithm about to be released in their vehicles. This algorithm is named occupancy networks, and it is supposed to improve the HydraNets from Tesla. But how does it work? Why is it needed? Let's find out... Why a new kind of algorithm is

www.thinkautonomous.ai

Given the blind spot in the front below the front bumper, the front is based on persistence of the occupancy network results as you pull in. However, accurate front detection is still dependent on your wall in front having some features, so the AI has cues to estimate the depth. If it's a completely featureless white wall (and there is no clear line that separates floor from wall either) I think it'll be fairly difficult for the AI to figure things out. The side cameras won't really help that much.

If you want to understand how it works, it is based on the occupancy network, which stitches all the cameras:

A Look at Tesla's Occupancy Networks

In 2022, Tesla announced a brand new algorithm about to be released in their vehicles. This algorithm is named occupancy networks, and it is supposed to improve the HydraNets from Tesla. But how does it work? Why is it needed? Let's find out... Why a new kind of algorithm is

www.thinkautonomous.ai

Given the blind spot in the front below the front bumper, the front is based on persistence of the occupancy network results as you pull in. However, accurate front detection is still dependent on your wall in front having some features, so the AI has cues to estimate the depth. If it's a completely featureless white wall (and there is no clear line that separates floor from wall either) I think it'll be fairly difficult for the AI to figure things out. The side cameras won't really help that much.

When I move the car in the garage, I see it can detect the opening wall of the garage door. When I park tail in, the car can detect the painted line of the parking lot. The grove on the surface below the garage door (similar to the painted line in a parking lot) and the walls of both sides of the garage door could give some information.

If the car can detect the distance T of the tail of the car to the garage door wall and the depth D of the garage then the distance of the wall to the front of the car = D - T - car length. Inference may give result. I don't know for sure they can do that. Everything here is just speculation and discussion. Nobody knows for sure.

 

[L1TR]

When I move the car in the garage, I see it can detect the opening wall of the garage door. When I park tail in, the car can detect the painted line of the parking lot. The grove on the surface below the garage door and the walls of both sides of the garage door could give some information.

If the car can detect the distance T of the tail of the car to the garage door wall and the depth D of the garage then the distance of the wall to the front of the car = D - T - car length. Inference may give result. I don't know for sure they can do that. Everything here is just speculation and discussion. Nobody knows for sure.

Car may know only "possible" sizes of tagged objects on the road and "predict" a distance. It can't measure a distance. No way.

 

[stopcrazypp]

When I move the car in the garage, I see it can detect the opening wall of the garage door. When I park tail in, the car can detect the painted line of the parking lot. The grove on the surface below the garage door and the walls of both sides of the garage door could give some information.

If the car can detect the distance T of the tail of the car to the garage door wall and the depth D of the garage then the distance of the wall to the front of the car = D - T - car length. Inference may give result. I don't know for sure they can do that. Everything here is just speculation and discussion. Nobody knows for sure.

The question is how can it detect D if it can't figure out where the wall begins? That is why I mentioned the horizon line. It's fairly easy to test this by using some black electrical tape to create something high contrast on the edge between the wall and ground (or white tape if wall and ground is dark). Can also do vertical lines for reference points. Then vary the speeds at which you pull in to see how it affects the persistence.

If you read the link, it doesn't work based on subtraction of car depth, it just builds a 3D model of the environment around it.

 

[rlsd]

Car may know only "possible" sizes of tagged objects on the road and "predict" a distance. It can't measure a distance. No way.

The question is how can it detect D if it can't figure out where the wall begins? That is why I mentioned the horizon line. It's fairly easy to test this by using some black electrical tape to create something high contrast on the edge between the wall and ground (or white tape if wall and ground is dark). Can also do vertical lines for reference points. Then vary the speeds at which you pull in to see how it affects the persistence.

If you read the link, it doesn't work based on subtraction of car depth, it just builds a 3D model of the environment around it.

The link you quoted is just an opinion and may not be used for park assist. What we know is up to now the car can drive smoothly on the freeways and stop and go at every intersection. Let's wait and see.

 

[stopcrazypp]

The link you quoted is just an opinion and may not be used for park assist. What we know is up to now the car can drive smoothly on the freeways and stop and go at every intersection. Let's wait and see.

No, it's not just opinion, it's exactly what Tesla says they are using to replace the USS!

Read the FAQ from Tesla!
"Along with the removal of USS, we have simultaneously launched our vision-based occupancy network – currently used in Full Self-Driving (FSD) Beta – to replace the inputs generated by USS."

Tesla Vision Update: Replacing Ultrasonic Sensors with Tesla Vision | Tesla Support

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

 

[ewoodrick]

If you want to understand how it works, it is based on the occupancy network, which stitches all the cameras:

A Look at Tesla's Occupancy Networks

In 2022, Tesla announced a brand new algorithm about to be released in their vehicles. This algorithm is named occupancy networks, and it is supposed to improve the HydraNets from Tesla. But how does it work? Why is it needed? Let's find out... Why a new kind of algorithm is

www.thinkautonomous.ai

Given the blind spot in the front below the front bumper, the front is based on persistence of the occupancy network results as you pull in. However, accurate front detection is still dependent on your wall in front having some features, so the AI has cues to estimate the depth. If it's a completely featureless white wall (and there is no clear line that separates floor from wall either) I think it'll be fairly difficult for the AI to figure things out. The side cameras won't really help that much.

But I suspect that if the forward wall is featureless, and you pull to the point where the camera can't see the wall transition line, the side cameras may be used to determine distance to the wall by calculating the distance moved forward.

 

[ewoodrick]

Car may know only "possible" sizes of tagged objects on the road and "predict" a distance. It can't measure a distance. No way.

How do you do it with your eyes?

And don't forget that there are multiple forward looking cameras that CAN measure distance.

 

[rlsd]

No, it's not just opinion, it's exactly what Tesla says they are using to replace the USS!

Read the FAQ from Tesla!
"Along with the removal of USS, we have simultaneously launched our vision-based occupancy network – currently used in Full Self-Driving (FSD) Beta – to replace the inputs generated by USS."

Tesla Vision Update: Replacing Ultrasonic Sensors with Tesla Vision | Tesla Support

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

It's the opinion of the guy who wrote the article. Tesla did not say NN has 5 basic flaws like the author claimed. If the claim were true then we would not be able to drive the car at FSD 11.3.x now.

 

[stopcrazypp]

It's the opinion of the guy who wrote the article. Tesla did not say NN has 5 basic flaws like the author claimed. If the claim were true then we would not be able to drive the car at FSD 11.3.x now.

No it's not, the flaws pointed out were presented by Ashok Elluswamy of Tesla, the article just cites that presentation. The flaws in the previous NN were what the occupancy network (which was long in FSD before 11.3.x) aimed to address.

For example it links where Tesla compared the occupancy network with the previous NNs:

 

[L1TR]

How do you do it with your eyes?

And don't forget that there are multiple forward looking cameras that CAN measure distance.

"Binocular vision" - I mentioned it in previous posts. But it may work in front only. And don't forget, it's easy to orient on a road if we have cars around (we know what a car size in general to "predict" how far it is), lines, etc. Tesla doesn't know anything about lawn mover, for example, so next step - tag all possible staff in garage - is any advantage in migration to Tesla vision, I don't think so.

 

[ewoodrick]

"Binocular vision" - I mentioned it in previous posts. But it may work in front only. And don't forget, it's easy to orient on a road if we have cars around (we know what a car size in general to "predict" how far it is), lines, etc. Tesla doesn't know anything about lawn mover, for example, so next step - tag all possible staff in garage - is any advantage in migration to Tesla vision, I don't think so.

I don't have to recognize an object to determine it's distance. I just need to understand it's an object.
And doesn't the car's movement help provide the stereo picture needed to measure distance?

I don't know how the car is doing it, but there seems to be a number of relatively easy solutions, using algorithms already developed to measure distances.
Stereo vision on the front and the forward movement to emulate stereo vision on the side.
The rear doesn't have the stereo vision, but it does have movement to help determine distance.

 

[rlsd]

Regarding distance measurement: Tesla has a patent for that.

Interesting Patents: Tesla's Distance Detection through Image Data

This week, Tesla was granted a patent for a system that utilizes image data from a vehicle's camera along with a trained machine learning model to identify the distance of an object from a vehicle.

founderslegal.com

One thing we don't know is which method Tesla uses for park assist.

 

[rlsd]

No, it's not just opinion, it's exactly what Tesla says they are using to replace the USS!

Read the FAQ from Tesla!
"Along with the removal of USS, we have simultaneously launched our vision-based occupancy network – currently used in Full Self-Driving (FSD) Beta – to replace the inputs generated by USS."

Tesla Vision Update: Replacing Ultrasonic Sensors with Tesla Vision | Tesla Support

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

My bad for not reading the whole thing.