I use the floor mounted Stop SignA little virtual tennis ball shows up on the screen and then alarms "STOP" when the ball hits the front of the Tesla, just like the ultras did when you got to 12 inches.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla replacing ultrasonic sensors with Tesla Vision
- Thread starter diplomat33
- Start date
Ah indeed, here's a less cropped version with most cameras at the 4:3 aspect ratio from AI Day 2021:
The base of the stop sign is visible from both the front wide and right pillar. And both the right pillar and right repeater see the closest tree base with adjacent rock on the ground.
Some sample clips from the new cams in this thread:
Someone posted some footage from older cams earlier in the thread too. New cams see a little more of the side of the car and less of the mirror for example. Can’t tell if they see more to the side of the car but I wouldn’t be surprised if there’s some cropping happening in the video output
Last edited:
These two items are pure speculation on your (and Elon's) part. There's simply no indication that the occupancy network renders items down to the resolution you suggest. Perhaps that is why some people are "chicken littl'ing" this. And yes, it's very similar to the removal of radar which brought back the phantom braking problem in Navigate on AutoPilot that was all but solved.
Re 4) from AI Day 2022:
Drift of 1.3 cm implies system resolution is at least that high.
Re 5)
Once the ON has a high resolution voxel cloud, representing vehicle clearance is easy.
No idea where you are quoting from but someone must have missed all grammar classes. 
It is painful to read.
Otherwise, if I understand it correctly, it should be OK with static objects in the blind spots, given the USS accuracy of >12” (30cm).
On the other hand, I am very skeptical of the .45 milli rad per meter. That is resolution of .45 millimeter from one meter distance. Not sure if it is possible with the current cameras. If anyone has the specs then we can calculate their resolution.
It is painful to read.Otherwise, if I understand it correctly, it should be OK with static objects in the blind spots, given the USS accuracy of >12” (30cm).
On the other hand, I am very skeptical of the .45 milli rad per meter. That is resolution of .45 millimeter from one meter distance. Not sure if it is possible with the current cameras. If anyone has the specs then we can calculate their resolution.
Hope someone with a new 3/Y would posts here soon. Would like to know what the display shows when parking. Does it show any warnings like Proximity Detection not Available or any distance information at all? Also any beeps or just silence?
Auto transcript from YouTube. I'd reccomend listening to verify, I was focused on the centimeter part. If milli radians per meter is the right unit, then it's a worse angular accuracy the further out something is, basically .45mm*distance ^2 giving 45mm angular accuracy at 10 meters. This could be correct for long baseline observations (visual version of SAR), not a single image.No idea where you are quoting from but someone must have missed all grammar classes.
Otherwise, if I understand it correctly, it should be OK with static objects in the blind spots, given the USS accuracy of >12” (30cm).
On the other hand, I am very skeptical of the .45 milli rad per meter. That is resolution of .45 millimeter from one meter distance. Not sure if it is possible with the current cameras. If anyone has the specs then we can calculate their resolution.
USS can detect objects closer than 12, but it doesn't give a range. There is a minimum though. Accuracy in the 12-36 range should be pretty good.
SalisburySam
Active Member
I would advise preparation to be upset.
Was there a noted requirement in the purchase agreement that they had to notify all owners of product improvements that I missed?
I saw some comments on Reddit that said that it literally shows nothing, no sounds or anything
I was in my driveway with the back of the car 3 feet from the house and it was yu first drive of the day. How does your comment explain a ghost motrocyclist behind me?
The best match the NN came up with for your house was a motorcyclist.
Now they use lumpy shapes to represent unknown things.
My HOUSE was replaced by a motorcycle? Now I know why the AI guy quit earlier this year. Elon has Ukraine peace plan, He counseled Kanye West about insulting Jews, he spoke/did not speak w Putin, he's going broke w Starlink in Ukraine, he's gonna fix Twiiter which has basically become a sewer. He's 50 which means he was born in 1972. Just checked, Apartheid ended in 1994, so I think he learned a thing or two from massa whitey. btw I'm white
stopcrazypp
Well-Known Member
As another mentioned, the current system just looks up the most similar object it has in its database (in this case a motorcyclist) and assigns that. So it's not unusual at all to falsely identify an area as a vehicle (for example I have cabinets in front of my car when I pull in and it'll show a semi truck in front of me).
The occupancy network instead only has the task of determining if the given area in a space is occupied or not (it shows blocks). At most it'll label the blocks a different color, but it won't be trying to put a predetermined object model there.
Anyways, read the links I posted. The Occupancy Network is an entirely new system that does not work like the previous one, so what you observe in the old system is irrelevant to determine how the new system would work.
Similar threads
