Holiday weekends are great for some fun are not they?

Worked with @rice_fry on the Tesla depth perception NN.

Remember that stuff in ?

Well, the cars actually have a depth perceiving net inside indeed. though unlike what's shown it's a lot lower res
A whopping 160x120 grid (so 1/8th of the native camera resolution) but hey those are actual 3D points in space. I don't know how to make a 3D video, so just visualizing "distance" as "brighter = closer" on the scale of 5 to 62 meters (the actual limit of the output):
Here's the corresponding time-synced video to better understand what you are looking at.

Pretty cool actually despite all the limitations.
you might also notice there's a strange white line in the middle.
What I think is going on is Tesla trained against main + fisheye cams and on the older cars there's a heater element (shown from old screenshot) introducing garbage into the stream.
That's why some think they are retraining with lidar now - because pure optical flow is too hard.

Here's also an example of merged objects shown as one that then corrects into two separate ones later that might be helped with the lidar "groundtruth" as well.
The sky rarely changing might be also contributing to the artifacts at the top.

the above segment with pedestrians in dynamic
Some short city driving
And this one is visualizing another axle of points.

The effect is super nice, reminds me of the "Hedgehog in the fog" somewhat:
visualized single frames of point clouds (do NOT load this in firefox if you can avoid it), scrolable/zoommable/rotating with mouse

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with green

green Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @greentheonly

4 Jul
So in regards to the recent Chinese crash with video where AP exits roadway in sharp turns:
I long planned to do a demo of a similar condition but always forgot and I guess better late than never.
It's actually really easy to reproduce
m.weibo.cn/status/4655186…
Find an (empty!) road with a sharp curve and point autopilot at it. For most safety ensure there's an extra space outside the curve so you can catch the car before it crashes into stuff and boom!
(this is 2021.4.18.3 btw)

Remember to always pay attention when on AP!
If you enable debug mode you see the car actually outputs various alerts before the eventual "take over we are giving up" (weird there was no diseng chime btw)
and it knows it cannot take the turn safely at those speeds, in fact it tries to lower speed but definitely not enough
Read 7 tweets
31 May
Some more fun with radar-less Model Y (thanks @spleck)

First we laced a brown box with foil (for great radar response) and tested my model X on it (FW 2021.4.18)

It rammed it like nothing, but we did get an FCW.
Internal telemetry view (no radar overlaid because it makes the picture too busy)
One frame from radar view.
From it we learn the box is reported as a vehicle, but existence probability is only 30% and obstacle probability is 60%
Read 6 tweets
5 Apr
If a picture is worth a thousand words, a video must be good for a couple gazillions.
With all the debates about how suitable @Tesla cabin cam for driver monitoring, I will let you form your own opinion.

It does look like "full color" RGB based on the red elements

@rice_fry
I was a bit surprised that in a city at night there's enough environment lighting to make DM seem very workable.
All these videos are full resolution, 36 fps as Tesla uses them.
Simulated distracted driver part 1
Read 11 tweets
6 Mar
With all the brouhaha about Ford Mach-E, I decided to test-drive one.
Local Ford dealer had one in stock even (Ford >> Audi)
I got a salesperson that was first time in the car so we had to use manual a lot (Tesla >> Ford)
Manual depiction of UI does not match the actual UI 🤷‍♂️
The good:
I feel like it's ready to replace model 3/Y If you don't need to ever drive outside the city you live in.
Not as fast as a Tesla but plenty enough for light-to-light hopping.
Has wifi hotspot built-in ($$$)
Bird-eye view that actually works and turns on as you approach
Ford ADAS is on par with Tesla BAP. Better for lane-changes where it lets you to take over without disengagement, but a bit flip-floppy in the lane in turns as if steering angle is severely limited.
10 seconds between steering nags at 80mph

The UI is a bit confusing. "READY"?
Read 21 tweets
9 Dec 20
If Tesla was a REAL software company, after you press that "YES" dialog button about "This is beta, blah blah, I understand what I am doing and will keep both pieces if it breaks" setting windows become accessible ImageImage
A lot of settings...
So many settings it's take you several minutes to even read them
Read 8 tweets
3 Dec 20
Some seriously cool and very practical information Esp. for the tinkering types.

@Tesla now gives access to repair manuals, service information, vehicle diagnostic (remote in EU) and such.
For free!
I was told it's not a bug.
Seriously step in the right direction!
go to service.tesla.cn and then you create account with country set to China and access level: diagnostic + software.
in second screen fill in some extra details like this:
you'll get to the registration screen in likely your local region that looks like this:

YOU ARE DONE, don't press the register button, press the sign-in and that brings you to the service site with credentials you have created.
Read 11 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!

:(