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

FSD V12 Elon Demo

This site may earn commission on affiliate links.
imagine the impression on Elon for "mind blowing" as he has Asperger's and on the brilliant spectrum
vs us ordinary folk
will it be mind exploding for us?

I think his perspective comes from an appreciation of the cumulative engineering work that's inherently behind the scenes.

Back in the 1990s when I was getting my EE degree, some friends and I took an ancient Oldsmobile POS and turned it into an EV with a bunch of scrap pb acid batteries, a forklift motor and a 486 motherboard. For most, pressing the accelerator would result in a somewhat feeble experience for the few miles the tired batteries would last. For us though, who breathed life in that kluge of a system, it was nothing short of a "mind blowing" miracle every time we drove it successfully.
 
So far the route seems pretty simple. I am not seeing any super complex edge cases. It is basically suburban driving. Having said that, the driving looks very smooth. It is handling roundabouts, pedestrians etc really well. Elon keeps reiterating that they did no hardcoding, V12 is inferring how to drive this new route based on the video training.
 
So far the route seems pretty simple. I am not seeing any super complex edge cases. It is basically suburban driving. Having said that, the driving looks very smooth. It is handling roundabouts, pedestrians etc really well. Elon keeps reiterating that they did no hardcoding, V12 is inferring how to drive this new route based on the video training.
There have been a few spots where I noticed the car was driving over the left lane for a few seconds on turns, but no cars were present, it corrected, and I assume would be easily smoothed out by public release. The roundabouts were really impressive!
 
  • Like
Reactions: diplomat33
First intervention at a red light. I think the car thought the green light for the turn only lane was it's light so it started to go when it actually had a red. Elon says they will feed it more video of that scenario to train it.

So basically first intervention after only 20 minutes of driving. So it looks like intervention rate is very poor. I think the key will be how quickly the intervention rate improves with video training.
 
The control network seems to have learned appropriate driving speed for the current road ignoring the set speed (seemed like it was at 85mph the whole time). It also happened to switch into the lane with fewer cars waiting at a red light, so maybe it learned that people switch lanes when approaching red lights?

I didn't quite catch the full discussion around NHTSA recall requiring complete stops where Elon Musk and Ashok Elluswamy were talking about under 5% of drivers behave that way. Were they saying they needed to oversample the training data with examples of full stops to comply with NHTSA before wide release?

Overall indeed pretty impressive that all of it's handled by neural networks without explicit control logic for lanes and objects. Hopefully that will address many issues related to lane selection and merges, etc.
 
First intervention at a red light. I think the car thought the green light for the turn only lane was it's light so it started to go when it actually had a red. Elon says they will feed it more video of that scenario to train it.

So basically first intervention after only 20 minutes of driving. So it looks like intervention rate is very poor. I think the key will be how quickly the intervention rate improves with video training.
It does something like that for me now. There is a road where the cross traffic comes at a slight angle and when their light turns green, FSDb starts to run the red light that is in front of it into cross traffic. it seems if it’s in the left lane, the light to the left is the one it uses to decide to go, in this case, it’s the light signaling cross traffic. IMG_0172.jpeg