Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

HW2.5 capabilities

This site may earn commission on affiliate links.
Tweaked from .40? I thought .40 was a new NN?

I think that's what he said, but I was more referring to the jump from .34 to .42 in total. I didn't have .40 to compare, so I can't really speak to that. But it had been a long time before .34 that we saw any neural net changes.

Okay so which was the one that jimmy_d discovered was very similar to googlelenet?
 
Big news. First images of the 2.5 radar and Model 3's AP2.5ECU/CID (ICE):

M3 radar.jpg


M3 AP2ECU.jpg


For comparison, the drawing from electrek:

model-3-computer-4.jpg
 
Looks like the radar mount supports 2 radars. Only right side mounted, but there is a slot for a left radar?

But only one radar "cavity" on the metal part in front of it.

I wonder how two radars in the middle would help anyway? Adding corner radars would make more sense, but no such this is seen here... (they could always attach those directly to bumpers though, like the competition does, if they ever add such).
 
  • Like
Reactions: lunitiks
I guess, redundancy in case one fails... Also, I wonder if they mount it to the left/right side depending on left hand or right hand drive configuration... perhaps the slight offset helps? I'm just randomly guessing at this point!

Possible... possible...

Just for reference, the new Audi A8 has radars in each front corner (one corner has a dual-purpose one: long range and corner), while lidar is in the middle.

If I recall, my old Audi A8 (2011 model-year) had two dual-mode radars in the front corners (corners + medium/long forwards) and nothing in the middle. So quite a bit more separation.
 
"Now that the foundation of the Tesla vision neural net is right, which was an exceptionally difficult problem, as it must fit into far less computing power than is typically used, we expect a rapid rollout of additional functionality over the next several months"
Hmmm... "Far less computing power than is typically used"...

...and this would be on their supercomputer that's 40x more capable than previous generation autopilot?

GoogLeNet + DrivePX2 vs classical computer vision and a custom ASIC from 2012. Currently, progress isn't winning here! Perhaps it will... in 3 months maybe, 6 months definitely! :)
 
  • Like
Reactions: Kant.Ing
To be honest, I think Elon's reference was to the fact that the AP2 board's chips were never "rated" (in marketing specs) by Nvidia for full autonomous driving - though of course the beefier chips were meant to handle sensors fusion too which Tesla does not need as much...

It was a much discussed fact that the AP2 board ("Level 5 capable" according to Elon Musk in October 2016...) is not very powerful compared to what companies think is needed for Level 4/5.

Do I recall right that AP2 has something akin to AutoCruise and not AutoChauffeur?

Drive PX-series - Wikipedia
 
"Now that the foundation of the Tesla vision neural net is right, which was an exceptionally difficult problem, as it must fit into far less computing power than is typically used, we expect a rapid rollout of additional functionality over the next several months"
Hmmm... "Far less computing power than is typically used"...

...and this would be on their supercomputer that's 40x more capable than previous generation autopilot?

GoogLeNet + DrivePX2 vs classical computer vision and a custom ASIC from 2012. Currently, progress isn't winning here! Perhaps it will... in 3 months maybe, 6 months definitely! :)
Man, that’s cold blooded, but Elon made his bed/tweets so he can lay in them. I do appreciate the increase in transparency in the document, a welcome status update that sounds realistic.