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
Start a Discussionhttps://teslamotorsclub.com/tmc/tags/

HW2 not enough for self-driving?

Discussion in 'Tesla' started by SoCal Tsla, May 23, 2017.

Tags:
  1. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    So my wife sent me a Chinese post about NVIDIA’s Drive PX 2 is not enough for level 5. I'm not really surprised, as when we were purchasing my MX it was mentioned to us specifically that the GPU can be upgraded easily. I wonder if that will be extra cost in addition to the $4000 upgrade?
     
    • Disagree x 2
    • Like x 1
  2. JHWJR

    JHWJR Member

    Joined:
    Jan 31, 2017
    Messages:
    344
    Location:
    Pittsburgh PA
    Tesla has not definitively stated it's plans should the current computer end up being insufficient, but I think the implication is strong enough that, if you paid for FSD, and that requires an upgrade in processing power, you won't be charged for the upgraded computer.
     
    • Like x 3
    • Funny x 1
  3. Saghost

    Saghost Active Member

    Joined:
    Oct 9, 2013
    Messages:
    4,773
    Location:
    Delaware
    #3 Saghost, May 23, 2017
    Last edited: May 23, 2017
    That $4k isn't fixed. Tesla may well move it up or down at any point in the future. They might even choose to stop offering it, though I'd be surprised.

    I've read that a lot of people think Tesla will move it up a bunch after they prove FSDC works and get approval for it. (Not at all sure they are right.)

    Certainly if more hardware they weren't expecting is required, I'd expect to see the price raised to account for that.

    If you buy it on delivery or ahead of any changes, I'd expect any surprise hardware upgrade to be provided for free.
     
    • Like x 1
    • Funny x 1
  4. chillaban

    chillaban Active Member

    Joined:
    May 5, 2016
    Messages:
    1,903
    Location:
    Bay Area
    I hate speculating on whether or not a certain piece of hardware is enough or not enough to do something based off how we feel.

    When AP1 came out, there were some that believed that it can't even be an adaptive cruise + lane keeping system, because it has one camera and one radar, while incumbents on the market already had 2-4 radar systems and Mercedes had just started touting their dual stereoscopic cameras + 2 radar adaptive cruise solution.

    Fast forward a year or two, I'm sure most of us agree that AP1 is an industry leading system that uses surprisingly fewer sensor components compared to competitors that do less.


    Will AP2 be the same way? Who knows. But I think you gotta take a big grain of salt when the CEO of NVidia says Tesla needs more nvidia hardware to get the job done (who stands to benefit from that transaction? And which other Drive PX customers are listening and changing their orders to dual-GPU models?).

    The one thing for certain is that Elon advertised Full Self Driving and many of us bought the package. If it involves us paying more money or never follows through, Tesla would have substantial legal problems.
     
    • Like x 5
    • Funny x 1
  5. Tam

    Tam Active Member

    Joined:
    Nov 25, 2012
    Messages:
    2,717
    Location:
    Visalia, CA
    We paid for driverless option so we will get that function sooner or later, there's no question about it.

    However, not all driverless options are the same. Tesla's goal is to be twice better than average human while other companies will not settle for that: Toyota wants zero deaths from their future system.

    So the current hardware that is claimed to be 40 times more powerful than AP1 will do but don't expect it to reach Toyota's goal which will require you to pay up much more money for more robust hardware.
     
  6. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    That's my understanding too, and I'll be really disappointed if they decide otherwise
     
  7. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    NVidia has 3 chips and the one for Tesla HW2 is customized and between 1&2. their chip 3 is enough for FSD and I think Tesla should have an idea about how much that will cost
     
  8. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    That's what I think too. On the other hand, current GPU definitely cannot handle all the sensors if the chinese post was correct. I was look all over for an English version of the post but I cannot find it. I think they basically took the GPU out and did some research.
     
  9. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    0 death? That's not possible, at least in the near future. My understanding is if Toyota won't release the system till they achieve that, they will be swiped out of the market. If they will release something before that, then essentially they are doing exactly what Tesla is doing right now...
     
  10. Tam

    Tam Active Member

    Joined:
    Nov 25, 2012
    Messages:
    2,717
    Location:
    Visalia, CA
    Whatever you see now is what you got: 40 times more powerful than AP1 and it is irrelevant that you see others have more chips than Tesla's capability.

    It is also irrelevant that you see Uber, Google have LIDAR and Tesla doesn't. You didn't pay for those specs.

    It is also irrelevant that you see others have more than 1 RADAR and Tesla has only 1. You didn't pay for those specs.

    What you paid for is:

    1) 40 times more powerful than AP1
    2) Driverless
    3) As good as twice human driver.

    And I believe although it is missing additional RADAR, LIDAR, Chips...., it will do what the specs say.
     
  11. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    Found the link:
    Look inside Tesla’s onboard Nvidia supercomputer for self-driving

    "Therefore, it appears to be a custom solution for Tesla and it is not as powerful as Nvidia’s most powerful solution for self-driving, which the chipmaker itself doesn’t believe can enable fully self-driving."

    So it's not driverless. It's not about 0 death or high/low standard. It won't achieve driverless without an HW upgrade. Jensen Huang said it can achieve level 3 but need two chips like that to do level4-5...

    I barely did any research in this area so I could be wrong, but I thought level 4 is only work as designed and level 5 is as good as human, what's the definition of as good as twice human? (again no sarcasm just eager to learn)
     
  12. lunitiks

    lunitiks ˭ ˭ ʽʽʽʽʽʽʽʽʽ ʭ ʼʼʼʼʼʼʼʼʼ ˭ ˭

    Joined:
    Nov 19, 2016
    Messages:
    1,474
    Location:
    Prawn Island, VC
    #12 lunitiks, May 23, 2017
    Last edited: May 23, 2017
    This will probably help your thinking. Official definition chart:

    [​IMG]

    It is kind of abstract, but important to understand. The key here is to know what "driving modes" means. By this, SAE means operational domains with characteristic task requirements, like expressway merging, high-speed cruising, low- speed traffic jams, etc. An L5 vehicle must be able to handle all such modes just as well as a human driver. In contrast, an L4 vehicle has one or more driving mode restrictions. For example, the L4 vehicle cannot do self parking. Or driving when its dark. Or leave the city of L.A. Or do a hill start. Some kind of restriction on full automation. OTOH, within the driving domains that the L4 vehicle does handle, it must be just as good as an L5 system. I.e. the driver must not be required to do anything at all. The car must be able to perform 'perfectly', and if it somehow figures out that it should not continue to drive, it must be able to safely do a fallback manuever, for example slow down, pull over, put on warning lights and maybe call for help.

    Anyway, back to your question: AFAIK, when automakers talk about the system being «better» than a human driver -- that is, better than the average human driver -- they're not talking about the level of automation (L3, L4 or L5), but instead they're talking about how much lower the probability of an accident is when the system is engaged compared to when a human is doing all of the driving tasks. For example, an L2 system -- like Teslas AP1 or EAP -- could possibly perform better (i.e. safer) than an average human driver in the specific driving mode. Think of the simple highway lane keeping feature: It's not difficult to imagine that the L2 system performs better (safer) than your average person when it specifically comes to lane keeping. Yet, the L2 system has by definition no fallback procedure or might not be designed to watch out for flying car wheels, roadkill or man holes, so the person behind the wheel is expected to monitor and intervene at any moment.
     
    • Informative x 3
  13. Tam

    Tam Active Member

    Joined:
    Nov 25, 2012
    Messages:
    2,717
    Location:
    Visalia, CA
    You are still listening to others who say Tesla just can't do it just because you need more chips, more radars, more lidars....
     
  14. Tam

    Tam Active Member

    Joined:
    Nov 25, 2012
    Messages:
    2,717
    Location:
    Visalia, CA
    #14 Tam, May 23, 2017
    Last edited: May 23, 2017
    There are about 1.3 million people die from car accidents annually.

    If your Tesla works as its spec of twice better than human, it would save 650,000 lives.

    On the other hand, there are still the rest of 650,000 people who will die because the spec is only twice as good as human driver.
     
  15. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    Wait is that how the math is done? That's gonna be hard to prove...
     
  16. Saghost

    Saghost Active Member

    Joined:
    Oct 9, 2013
    Messages:
    4,773
    Location:
    Delaware
    I actually think the "safer than" math includes all accidents, not just fatalities. It is, as you say, very fuzzy math, especially when something like Autopilot operates for only a portion of the miles - and the easiest miles to drive, at that.
     
  17. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    Thanks! This info is really useful and solved many of my questions lol
     
  18. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    After reading @lunitiks's reply it's making more sense to me now. And that math sound like no math to me coz you can hardly quantify many aspects of the driving condition, thus making it sound more like a marketing phrase to me(which is often misleading)... something like some toothpaste is "twice as effective than other brands" but turns out only true in some small aspect....
     
  19. Tam

    Tam Active Member

    Joined:
    Nov 25, 2012
    Messages:
    2,717
    Location:
    Visalia, CA

    Since your Tesla is logged, it is easy to compile a spreadsheet to see do how many collisions are detected when the automation is on VS how many when the automation is off (classical non-AP are always off).
     
  20. SoCal Tsla

    SoCal Tsla Member

    Joined:
    Apr 27, 2017
    Messages:
    257
    Location:
    Irvine, California
    yes but as @Saghost pointed out, AP is used in easier scenarios. My 5yr old daughter does 1+1=2 better than me doing calculus, but that doesn't mean she is better at math....
     

Share This Page