Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
Interesting link I found: Letter to the editor: NVIDIA Jetson severely limited during object recognition, tracking

Now I know that Jetson TX2 does not have the discrete GPU that Tesla uses, and there's no 4k camera inputs and not quite 13 cameras and the software is somewhat different too, but still interesting.

I wonder what resolution the 7fps was achieved at and with how many cams.
Not having that big discrete GPU is a big difference. This is really interesting though — I wonder if this is similar in compute power to that second Parker node on ap2.5? If so, that might support the theory that AP2.5’s secondary node is still powerful enough to do some basic image recognition in the event of a hardware failure.
 
Interesting link I found: Letter to the editor: NVIDIA Jetson severely limited during object recognition, tracking

Now I know that Jetson TX2 does not have the discrete GPU that Tesla uses, and there's no 4k camera inputs and not quite 13 cameras and the software is somewhat different too, but still interesting.

I wonder what resolution the 7fps was achieved at and with how many cams.
The TX2 does 1.5TFLOPS in FP16 (equivalent to 0.75 TFLOPS FP32). Tesla's PX2 can do 4 TFLOPS, so roughly 5x the computing power.
 
It's hard to tell from the images how much overlap the cameras have. Taking redundancy aside, since the cameras seem to have varying field of views, stitching together for a true 360 view isn't trivial.

In order to get good coverage, you either need a good 360 degree view stitched to one "large" image. Or you need non-insignificant overlap. (imagine you see something at the camera's edge, if you don't have enough overlap with the next camera, how will you know if it's a car that you need to worry about, or a shadow that you don't?)

We can compare the camera positions with those of the new Audi A8. This has a 360 degree view although it uses only 6 cameras. The postions are relatively similar except for the 2 backwards which he does not have, so we can safely assume that there is 360 degrees for the Tesla. I hope Tesla puts it together and animates it just like Audi did.

 
We can compare the camera positions with those of the new Audi A8. This has a 360 degree view although it uses only 6 cameras. The postions are relatively similar except for the 2 backwards which he does not have, so we can safely assume that there is 360 degrees for the Tesla. I hope Tesla puts it together and animates it just like Audi did.

Where is the front facing camera on the Audi A8? (There is no way it can see directly in front of the car if it is in the windshield like on the S/X/3.)
 
  • Disagree
Reactions: Troy
Source: Tesla's AI-based Autopilot tested after owner tapes cameras in experiment

Use of tape to degrade the cameras (tape isn't 100% opaque) - with 2018.10.4 running.

Front facing radar and short-range ultrasonics not obstructed.

In my opinion, discussed in another thread here, he do not cover the front cameras well enough. He does it in two tries - the first one is only covering to 2nd heat thread - no wonder AP works. The second try still has an opening on the bottom.
 
That just won't be enough.

upload_2018-3-20_10-26-5.png
 
  • Like
Reactions: J1mbo
I'm confused about your comment as he then added to more rows of tape!

LqqUoHM.jpg
Yes, but if I remember the video correctly, he concluded before adding the extra tape. And, there is still a few cm opening on the lower end.
A typical scientific failure - eureka! blown up in the news - but not yet confirmed from others, and then one finds the experiment had not control of all the variables.
 
  • Like
Reactions: croman and daud
Went out earlier today to put this to the test. Sorry for bad audio recording, video should be pretty clear, though: *debunked*. Properly and fully covering the front camera's disables Autosteer. Didn't have time for a more thorough test unfortunately. Would've liked to also test partial front-unit covering with B-pillars fully covered.

video of my test:

I did a more extensive 2018.6.1 vs. 2018.10.4 test on Monday, haven't had time to edit and put together that footage yet. My findings: doesn't give "take over immediately - hands on the wheel"-warnings anymore while taking difficult corners on local roads when it can't steer through a sharp turn itself, where it did display and give an audio warning before in 2018.6.1. Also improvements that have been noted by many others like better cornering driving on- and off-ramps - not stickering to or on the outerlane, but keeping the car center. Less availability on local roads, on locations where before in 6.1 it could navigate and activate Autosteer - without any painted lines present - on just the curb contrasting with the road surface.
 
Last edited:
Yes, but if I remember the video correctly, he concluded before adding the extra tape. And, there is still a few cm opening on the lower end.
A typical scientific failure - eureka! blown up in the news - but not yet confirmed from others, and then one finds the experiment had not control of all the variables.

Tesla is literally a breeding ground for myths so i'm not surprised.