TMC is an independent, primarily volunteer organization that relies on ad revenue to cover its operating costs. Please consider whitelisting TMC on your ad blocker and becoming a Supporting Member. For more info: Support TMC
  1. TMC is currently READ ONLY.
    Click here for more info.

Andrej Karpathy - AI for Full-Self Driving (2020)

Discussion in 'Autopilot & Autonomous/FSD' started by Bladerskb, Apr 21, 2020.

  1. mtndrew1

    mtndrew1 Active Member

    Joined:
    May 12, 2015
    Messages:
    1,311
    Location:
    Gardena, CA
    Right. This thread is becoming insufferable. A LIDAR system alone can only make it up to its first intersection safely. Then it’s time to find safe harbor.
     
    • Like x 2
    • Love x 2
    • Funny x 1
  2. diplomat33

    diplomat33 Well-Known Member

    Joined:
    Aug 3, 2017
    Messages:
    6,832
    Location:
    Terre Haute, IN USA
    "An autonomous car can safely drive on vision alone."

    How safely though? Mobileye says that camera-only cannot achieve the desired safety level they need.

    The big reason why I am pushing camera + lidar is because I believe it is the ONLY way to achieve safe, reliable FSD. I don't believe that camera-only can achieve the desired safety level for FSD.
     
  3. mtndrew1

    mtndrew1 Active Member

    Joined:
    May 12, 2015
    Messages:
    1,311
    Location:
    Gardena, CA
    Great. NOTED.
     
  4. diplomat33

    diplomat33 Well-Known Member

    Joined:
    Aug 3, 2017
    Messages:
    6,832
    Location:
    Terre Haute, IN USA
    I agree on the insufferable part.

    This thread is basically this:

    "Lidar can supplement vision for added accuracy"

    "you need to solve vision"

    "vision is not good at X"

    "you need to solve vision"

    "Solving vision takes too long"

    "you need to solve vision"

    "You can't solve vision to 99.999999% without lidar"

    "You need to solve vision."

    etc
     
    • Disagree x 4
  5. mtndrew1

    mtndrew1 Active Member

    Joined:
    May 12, 2015
    Messages:
    1,311
    Location:
    Gardena, CA
    If that’s what you’re taking away from this then I believe you’re on a different wavelength than the rest of the contributors, myself included.
     
    • Like x 3
    • Love x 1
  6. diplomat33

    diplomat33 Well-Known Member

    Joined:
    Aug 3, 2017
    Messages:
    6,832
    Location:
    Terre Haute, IN USA
    #446 diplomat33, May 13, 2020
    Last edited: May 13, 2020
    Honestly, I do feel like we are talking past each other. :(

    Perhaps the reason why we are not communicating properly is because we disagree on what "solving vision" means. I think "solving vision" only takes you so far, maybe 99.99% and you need lidar to get the last 9's. But you all seem to define "solving vision" as actually getting to 99.999999% and therefore vision is so accurate and reliable that you don't need other sensors. That is why you don't think lidar is needed after "solving vision" but I do.
     
  7. diplomat33

    diplomat33 Well-Known Member

    Joined:
    Aug 3, 2017
    Messages:
    6,832
    Location:
    Terre Haute, IN USA
    Yes. That is what I mean. You could rely on lidar for some things so your camera vision would not need to be solved to the same 9's for some tasks.
     
  8. mtndrew1

    mtndrew1 Active Member

    Joined:
    May 12, 2015
    Messages:
    1,311
    Location:
    Gardena, CA
    I’ll make one final post here and I’ll honor that promise. While I’m absolutely certain it will fall on deaf ears here we go.

    Imagine a scale of 1-100. One represents a car with no driver aids whatsoever, just a steering wheel and pedals. One hundred represents a car capable of safely driving itself in all circumstances to the degree that it’s so much safer and more convenient than manual driving that there is no question about its abilities.

    Ten nines, four million nines, whatever. A 100 represents a compelling product to use on public roads which customers can opt to configure their car with where the benefits outweigh the drawbacks. This means safety, affordability, and desirability as a package. Nailing those means 100.

    Everything on this scale slides up towards the end goal of 100, of course.

    Vision alone, today, can get to say 60 on this scale. It can do a lot, it can sort of drive the car in the city, it can effectively drive the car on the highway. Useful, but not 100 on our scale.

    Adding LIDAR to this setup, today, immediately bumps that to 85 on this scale. You get to leapfrog vision overnight and rely on this incredible dot cloud to navigate the world. But you can’t get past 85 with (year 2020’s) vision capabilities + LIDAR.

    LIDAR’s additive abilities END where they are today, more or less, because it can’t see signs or markings or lights or all the things we need to navigate the world safely, and it never will. LIDAR is fundamentally incapable of these tasks and always will be. LIDAR let’s you cut in line but not to the front, only to 85. Not good enough for the end goal but some immediate and impressive progress.

    Now what? LIDAR plus 2020-vision got you to the 85 mark and you’re stuck. You cannot opt out of lane markings and signs and speed limits and light colors and all this stuff. That will never be optional. It must be extremely reliably solved and LIDAR can’t help. You’ll never get to 100 with this setup.

    So then what? Without solving vision to the degree that it can do all this on its own you never get to 100.

    So you solve vision to say 95. LIDAR is still back there at 85 and it’s not going to contribute towards the finish line because we already baked its capabilities into the 85%-capable system. We’ve passed LIDAR’s limits. So we’re at 95 on the scale with vision, and LIDAR is done contributing in a meaningful manner, and vision is doing all the heavy lifting in nearly every scenario. If we don’t pass 95 then no L5 for anyone and LIDAR isn’t going to help. It’s already accounted for.

    Now what? How do you start creeping past 99? It ain’t LIDAR. It’s vision. And to do this vision must have already surpassed LIDAR’s early leapfrog on the timeline. We don’t need LIDAR anymore because we fundamentally can’t need it to solve this problem. We need something else.

    While I remain skeptical that this problem will ever be solved I am solidly convinced that LIDAR will not be contributing past its leapfrog point. I’m convinced that only enormous data ingestion feeding neural net development (plus radar and ultrasonics) will ultimately get us to the 100 mark on this scale. And I’m convinced that Tesla is the only company with the fleet and tools deployed to pull this off, if it’s possible at all.
     
    • Love x 4
    • Helpful x 2
    • Like x 2
  9. diplomat33

    diplomat33 Well-Known Member

    Joined:
    Aug 3, 2017
    Messages:
    6,832
    Location:
    Terre Haute, IN USA
    Thanks for posting. This helps a lot.

    You don't need to reply. I am trying to quit too. :)

    But your analogy helps illustrate the different points of view.

    Some (like you maybe) believe that vision + lidar gets to you 85 but it is possible to get vision to 99 by "solving it" and then you don't need lidar anymore. That is why you all are arguing that the best approach is just to focus on vision and not bother with lidar at all.

    I am arguing that no matter how much "solving" you do, vision will always be stuck at a max of say 85 so the ONLY way to get to 99 is vision + lidar.

    That's the key difference.

    Thanks again for sharing.

    Peace out!
     
    • Helpful x 1
    • Disagree x 1
  10. Bladerskb

    Bladerskb Senior Software Engineer

    Joined:
    Oct 24, 2016
    Messages:
    2,068
    Location:
    Michigan
    Yes because there's no such thing as turning on red.
     
    • Disagree x 2
  11. Soda Popinski

    Soda Popinski Member

    Joined:
    May 9, 2018
    Messages:
    351
    Location:
    Los Angeles, CA
    I'm not sure of the utility of a robotaxi that can only make right turns at intersections.
     
    • Like x 2
    • Funny x 1
  12. Bladerskb

    Bladerskb Senior Software Engineer

    Joined:
    Oct 24, 2016
    Messages:
    2,068
    Location:
    Michigan
    This is fundamental misunderstanding of how any of this thing work.
    Perception needs to be at 99.99999% rate of success. Vision alone is currently IMPOSSIBLE to get to that level of accuracy for another decade or so.

    Get it through your head. VISION ONLY IS CURRENTLY IMPOSSIBLE.
     
    • Like x 1
    • Disagree x 1
  13. MP3Mike

    MP3Mike Well-Known Member

    Joined:
    Feb 1, 2016
    Messages:
    14,925
    Location:
    Oregon
    #453 MP3Mike, May 13, 2020
    Last edited: May 13, 2020
    You can't turn right on red at every intersection. In fact at some intersections there isn't even an option to turn right at all.

    So I guess you just make sure you never make a route that goes past any one way streets, or past a T intersection, or has intersections where right on red is restricted. I guess that really cuts down on you ODD and makes things much simpler. :rolleyes:

    Even so, you hit your first one, and then you are stuck with the only option being to just keep circling the block forever. (Until you get pulled over for illegal cruising or run out of fuel/energy.)
     
    • Like x 1
  14. Bladerskb

    Bladerskb Senior Software Engineer

    Joined:
    Oct 24, 2016
    Messages:
    2,068
    Location:
    Michigan
    Its not Lidar vs. Camera. 23 pages and its still the same circular argument
     
    • Like x 2
    • Disagree x 1
  15. diplomat33

    diplomat33 Well-Known Member

    Joined:
    Aug 3, 2017
    Messages:
    6,832
    Location:
    Terre Haute, IN USA
    #455 diplomat33, May 13, 2020
    Last edited: May 13, 2020
    Except, we are NOT talking about a robotaxi that is designed to only turn right at intersections though. You are right that such a robotaxi would be absurd. In context, we were talking about what happens if the camera vision fails on a "camera + lidar" car. If the camera vision fails, could the lidar still drive the car? The argument was that lidar is pointless because you need to "solve vision" and if the camera fails, lidar adds nothing, because the car would not be able to handle an intersection. The counter argument is that in that rare failure case, yes, the lidar would add something, because the car could navigate an intersection by turning right. So the lidar car could still handle an intersection in a pinch if the camera vision failed whereas the camera-only car would definitely fail the intersection if the camera vision failed. So lidar does add fall-back value.
     
    • Disagree x 2
    • Like x 1
  16. Bladerskb

    Bladerskb Senior Software Engineer

    Joined:
    Oct 24, 2016
    Messages:
    2,068
    Location:
    Michigan
    In MI I can confidently say the amount of intersections with traffic light with a 'no turn on red' sign are less than 5%.

    Most one-way streets don't have traffic lights because they are mostly side streets. Additional most intersection don't have traffic lights, they have stop signs.


    Not at all because none of your points compute. In addition you don't even have to turn at an intersection. There's something called side streets. Nonetheless I know what I'm talking about because i used to work as a premise technician for AT&T in MI and it was basically against the rules to make a left turn.
     
    • Disagree x 3
  17. MP3Mike

    MP3Mike Well-Known Member

    Joined:
    Feb 1, 2016
    Messages:
    14,925
    Location:
    Oregon
    I guess you have never been in a downtown area.

    Yes, but that isn't relevant to the conversation. As a tech were you allowed to go straight? Because your crippled Robotaxi couldn't. (Without possibly breaking the law.)
     
    • Like x 1
  18. scaesare

    scaesare Well-Known Member

    Joined:
    Mar 14, 2013
    Messages:
    8,178
    Location:
    NoVA
    You need to read the context of a post being responded to before rebutting a point not being made.
     
  19. scaesare

    scaesare Well-Known Member

    Joined:
    Mar 14, 2013
    Messages:
    8,178
    Location:
    NoVA
    #459 scaesare, May 13, 2020
    Last edited: May 13, 2020
    Only able to turn right on red... that sounds like Level 5 if I've ever heard it....
     
  20. scaesare

    scaesare Well-Known Member

    Joined:
    Mar 14, 2013
    Messages:
    8,178
    Location:
    NoVA
    How many times are you going to unleave this thread? :confused:

    Do we really need to spell out all the cases this is a completely absurd assertion?
     
    • Funny x 2
    • Like x 1

Share This Page

  • About Us

    Formed in 2006, Tesla Motors Club (TMC) was the first independent online Tesla community. Today it remains the largest and most dynamic community of Tesla enthusiasts. Learn more.
  • Do you value your experience at TMC? Consider becoming a Supporting Member of Tesla Motors Club. As a thank you for your contribution, you'll get nearly no ads in the Community and Groups sections. Additional perks are available depending on the level of contribution. Please visit the Account Upgrades page for more details.


    SUPPORT TMC