-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
They just need to reboot the MCU's. I'll give them 10 minutes moreWe've passed the 15 minute late mark. Signed $1 bill for whomever guesses the right start time?
.
DurandalAI
Member
The auto CAN network is a great example of how automotive engineers haven't typically designed with a security mindset. V2V comms scare the *sugar* out of me. Having a Tesla is a bit of a stretch in my comfort zone, as I've never been a fan of anything safety related such as a vehicle's control system being tied to any form of wireless communications. I realize it's necessary for a Tesla, and I'm a tech-head more than I am a tin-foil hat head, so the tech wins out.
Ulmo
Active Member
Ulmo
Active Member
This is real: they're finally not using non-local RAM for in-pipe processing.
All the other decisions are looking good. E.g., offloading DMU to SW forces SW to be streamlined anyway.
21x performance multiple makes understanding reality more real.
All the other decisions are looking good. E.g., offloading DMU to SW forces SW to be streamlined anyway.
21x performance multiple makes understanding reality more real.
Last edited:
So far, I am thoroughly impressed with the chip numbers.
I studied EE in college but went directly into software instead, and with the numbers displayed I'd be shocked if when HW3 is deployed we don't see some big leaps and bounds.
Really is a software problem at this point. Obviously this is an understatement, but I don't doubt the hype now.
I studied EE in college but went directly into software instead, and with the numbers displayed I'd be shocked if when HW3 is deployed we don't see some big leaps and bounds.
Really is a software problem at this point. Obviously this is an understatement, but I don't doubt the hype now.
Ulmo
Active Member
He's leaving out shape in his example. You need to see shape to understand it in his example. I think on roads that's "considered" less important, but I disagree. However, these people are fairly smart. Let's see.
I'm a novice at pixelated rather than shape analysis, but from what I inferred, the neural nets figure out the shapes' meaning somehow; that became evident by the end of the software presentation, especially with the 3D and cue examples. Everything else in the presentation sounded great. I think he addressed most my questions.
His answer to the lane change human training is great, for example.
I'm a novice at pixelated rather than shape analysis, but from what I inferred, the neural nets figure out the shapes' meaning somehow; that became evident by the end of the software presentation, especially with the 3D and cue examples. Everything else in the presentation sounded great. I think he addressed most my questions.
His answer to the lane change human training is great, for example.
Last edited:
Having the request for similar images for edge cases across the fleet was something I never even thought of, and is a pretty brilliant point in illustrating the necessity for real world data.
Ulmo
Active Member
The audience LIDAR question is dumb given the presentation. The presentation explained how vision solves that better than LIDAR can ever do.
Similar threads
