FSDBeta Neural Network/Machine Learning Thoughts




[ad_1]

FSDBeta 6 version 2020.44.15.4 video showing visualizations on the Model Y. These are my own thoughts on what the FSD Beta is doing with the settings behind the scenes in light of the new information shared by @greentheonly on Twitter. It is obvious in the conversation on this video that I am not a Machine Learning or Neural Network expert. Just an engineer trying to understand what is going on. Let me know if you find this useful.

Please see Green’s post here: https://twitter.com/greentheonly/status/1336467014727110656

Follow me on twitter @Chazman

Tesla Referral Code for free Supercharger miles or Solar discounts.
https://ts.la/charles46473

#FSDBeta #FSD #ModelY @Tesla @elonmusk

Source


[ad_2]

Comment List

  • Chuck Cook
    December 9, 2020

    Valuable insight; thank you for sharing it.

  • Chuck Cook
    December 9, 2020

    That's exactly what I though when I started noticing that different Teslas react differently in similar situations on the same version. It makes a lot of sense and I'm highly confident that that's precicely what the Tesla team is doing

  • Chuck Cook
    December 9, 2020

    Now that you know this information will it change the way you Beta test? I’m curious to understand when will the car learn from the network. It seems it is only getting better via a software update.

  • Chuck Cook
    December 9, 2020

    FSD should definently detect shops, malls, schools , libraries, street markets, restaurants in working hours etc pedestrian top priority locations and to take more precautions around them – to slow down before each blind zone, to prevent pedestrians from panic & unexpected moves, to consider if road & rules are slipy because of weather etc, to assume higher probability of pedestrians to be drunk…

  • Chuck Cook
    December 9, 2020

    Thats what i thought because people Had different Problems with the System. BUT I didnt see any Beta Tester understanding More than me when it comes to AI in the beta. Thank you for the Insight smart dude

  • Chuck Cook
    December 9, 2020

    Got it. Interesting.

  • Chuck Cook
    December 9, 2020

    Poor Brandon has always gotten the absolute worst version of FSD. His experiences have been way worse than the rest of you.

  • Chuck Cook
    December 9, 2020

    Is there a slider for the 'Trolley Problem' yet?

    Based on the FSD's lack of collision avoidance in your previous video with the other car nearly running the red into you, I actually doubt they have yet programmed in any ethical behaviour. Any dangerous moves by FSD are hopefully just accidental and not intentional. For now….

  • Chuck Cook
    December 9, 2020

    This could also be a potential explanation for why cars seem to learn even without software updates. The weights could be adjusted based on disengagements though we probably will never know.

  • Chuck Cook
    December 9, 2020

    Good video. It may explain the different behaviors of the vehicles at different locations. 🤔

  • Chuck Cook
    December 9, 2020

    Thanks Chuck, it is great that one of you made a video on the subject. Exactly my thought when I saw green's post earlier today : all FSDBeta testers are not given the same set of weights, thanks for your thoughts as to why it may be the case and why it may make sense.

    So it wasn't really Brandon's car the problem in early versions where it misbehaved, it was his own sets of weights/parameters 🙂

  • Chuck Cook
    December 9, 2020

    Good video/insight…I’ve read his tweets and I am almost 100% sure “California Stop” is going to be a future option for the end user lol.

  • Chuck Cook
    December 9, 2020

    Great insight!

Write a comment