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 or making a Paypal contribution here: paypal.me/SupportTMC

Is Model X (or was?) getting an "AP 1.5" before full-autonomous "AP 2.0" suite upgrade?

Discussion in 'Model X' started by AnxietyRanger, Jul 28, 2016.

  1. AnxietyRanger

    AnxietyRanger Active Member

    Joined:
    Aug 22, 2014
    Messages:
    2,228
    Location:
    EU
    #1 AnxietyRanger, Jul 28, 2016
    Last edited: Jul 28, 2016
    With the recent news of Tesla and/or Mobileye dumping each other for future products (an interesting he said/she said in its own right) I got to thinking about the Auto-Pilot upgrades in Model X once more. Some of you may remember me charting the Model X/S mules for clues on new cameras, radars etc. back in 2015.

    Of course we know an Auto-Pilot hardware upgrade is in the works. The best clue is the competition which has been placing far more hardware into their cars for years already. A high-end Audi or a Mercedes Benz has far more hardware situational awarence than any production Tesla to date, with multiple cameras, multiple radards back and front complementing the ultrasonics. Tesla only has one camera, one radar plus the ultrasonics. Yet, for whatever reasons (like the DrivePilot fiasco points out) Tesla is actually bringing all this together better, but still, there are limits to what the current hardware can do (for example, a Tesla can not detect cars approaching from the back until very close). We also know Audi is coming out with Level 3 autonomous driving in 2017.

    The second clue has been the various mules we've seen. We have seen both surround cameras as well as Tesla's (includind Model X mules) with what look like rear-bumper radars similar to what the Germans have. We have even seen Lidar setups on Model X as well as very advanced looking computing gear in a Model S mule suggesting advanced autonomous driving testing - much of this already in 2015. Tesla also showed initial Model X press shots with dual front cameras in the windshield as well as what looked like a high-up rear camera between the falcon wing doors, possibly related to a Gentex camera rear-view mirror and/or future autonomous driving updates.

    Mobileye also hinted at a trifocal camera coming to Tesla within a year. And of course, also back in 2015, Elon Musk also had this to say:
    Then, of course, Model X launched with the same AP 1.0 as Model S, but with a windshield camera housing for two cameras, suggesting something more was or at least had been in the works during the much delayed Model X development process. Next clue came this year, with the Model S facelift launch, when leaked schematics showed triple frontal cameras for Model S. These schematics suggest only three additional cameras over the current AP hardware, not a all-around seeing full suite. Yet, again, the Model S facelift shipped without these cameras. We also possibly hear from inside Tesla that current AP is already considered legacy and that a replacement comes this year.

    On the other hand, we have Elon Musk going on record that Tesla Model 3 likely to ship “autonomous ready” with new sensor suite and other quotes from him suggesting that full-autonomous requires more than just frontal cameras. Finally, with the breaking up of the Mobileye partnership, a lot of talk has come out suggesting Tesla wants to be more self-sufficient with their full-autonomous Autopilot, making it clear Mobileye will play no part in that phase of autonomous Teslas.

    Putting all this together, it seems plausible to me, the three camera AP upgrade is not the full autonomous suite. It possibly is/was an evolutionary step for the AP 1.0, perhaps an "AP 1.5", that included an upgraded Mobileye system and was/is intended to ship with the Model X and the Model S facelift - if the full-autonomous capable hardware (with Level 3/4 software updates coming out in the future) is the "AP 2.0" Tesla wishes to launch before Model 3 shipping. It is also possible given the talk, that this "AP 1.5" was/is either delayed and/or not fully satisfactory to Tesla and/or AP 2.0 hardare (Tesla's in-house efforts?) is expected to be ready sooner than initially planned.

    This begs the question: Will this "AP 1.5" still come out - or has it been cancelled entirely? The permanence in talks between Tesla and Mobileye, and the fact that Tesla has been upping the talk (with Mobileye concurring) on significant software upgrades coming to the current hardware, certainly makes it possible "AP 1.5" has been cancelled and the end of Tesla's and Mobileye's co-operation happening sooner than expected. On the other hand, the wording is vague enough to support the possibility that the "AP 1.5" upgrade might still happen - and Tesla's in-house efforts coming only a bit later (unless Tesla also has some other replacement for "AP 1.5" that is ready sooner).

    But certainly, if the "AP 1.5" is really an intermediate step to a full-autonomous "AP 2.0", skipping it might sense if it is delayed too much compared to the launch of the full suite. Very little point in introducing a new hardware suite if it is to be replaced by another in a less than year. If "AP 1.5" is cancelled, it might mean we won't see an upgraded AP suite until sometime next year. If "AP 1.5" is still a go, I would expect to see it in 2016 for new upcoming Model X/S facelift production.

    What do you guys think? Anything to this admittedly wild speculation?
     
    • Like x 3
  2. Alketi

    Alketi Member

    Joined:
    May 3, 2016
    Messages:
    320
    Location:
    Boston, North Shore
    My educated guess is that AP 1.5 (if it existed) has been cancelled. And that the drop-dead date for AP 2.0 is Model 3 production or sooner for the S/X, if its ready.

    I base this on:

    - The MobileEye breakup
    - The fact that Tesla has hired a number of high-profile chip designers
    - Elon's recent tweet about the techniques ("temporal smoothing" and "point-cloud") that would be possible with the current hardware

    This indicates to me that Tesla is going to design their own processing chip, and is also looking at longer term software solutions to support/improve the current hardware.

    Further, an intermediate (1.5) step makes almost no sense from a cost/support standpoint for Tesla.

    I'm slightly worried that bringing the chip design in-house might actually threaten the Model 3 launch. Not that they'd delay the launch, but rather that they'd ship the Model 3 with AP 1.0 if they absolutely had to. Elon mentioned at the All Things D conference that any "cool features" not ready for Model 3 would just be pushed to Model 3 2.0.
     
    • Like x 2
  3. AnxietyRanger

    AnxietyRanger Active Member

    Joined:
    Aug 22, 2014
    Messages:
    2,228
    Location:
    EU
    Thank you for the great summary, Alketi. I agree it all seems plausible. The move away from MobilEye - and thus the three camera setup too? - gets more possible confirmation from the earnings call where Elon Musk says they are gunning for an internal solution:

    Tesla probably had planned the MobilEye co-operation to last longer and had an interim upgrade to the Auto-Pilot for Model X and Model S facelift in the work, but are now aiming for the full-autonomous capable hardware suite in time for Model 3 at least (and launch it first when ready on Model S/X, if ready before Model 3 launch).

    It seems less likely we'll see an Auto-Pilot hardware suite upgrade in 2016. I also share the concern that delays in this development might lead to launching Model 3 with AP 1.0 initially, just like they did with Model X and Model S facelift which clearly were developed with an AP upprade in mind (Model X dual-camera casing, Model S facelift three-camera schematics), but didn't get it for whatever reason.
     
  4. Alketi

    Alketi Member

    Joined:
    May 3, 2016
    Messages:
    320
    Location:
    Boston, North Shore
    Based on Elon's comments yesterday, about the in-house AP blowing him away but that they need to get the software to run on a smaller computer, I draw two conclusions:

    1. They've likely got the next-generation sensor suite they think they need.

    2. The challenge is to now run the software on computing hardware that can fit and operate in the car.

    I think #2 points back to their in-house chip design.

    A wild guess is that they're not going to design an ARM-based CPU, but rather a dedicated FPGA solution. Such targeted solutions can be many times more powerful than even the latest graphics cards. You see this in the BitCoin arena, where people are selling targeted FPGA-based solutions that outperform everything else available.

    The next 12 months could be the most exciting technology-wise since the automobile first overtook the horse. Can't wait...
     
    • Like x 1
  5. J1mbo

    J1mbo Member

    Joined:
    Aug 20, 2013
    Messages:
    610
    Location:
    UK
    I would imagine Nvidia's Drive PX would be a far more attractive solution over a new chip design. Even with the best chip designers in the world, Tesla are not going to be able to compete with Nvidia on power/performance/price by the time the first customer's M3 rolls off the production line.
     
  6. AnxietyRanger

    AnxietyRanger Active Member

    Joined:
    Aug 22, 2014
    Messages:
    2,228
    Location:
    EU
    I wonder if the "triple camera housing" everyone is talking about for a potential AP 1.5 or AP 2.0 or whatever actually just means dual camera housing in the windshield and the third wide-angle camera in the nose of the car that would see side traffic. (Similar to what the Germans and Volvo use?) There is that big gaping hole in the Model X and Model S facelift noses, perfect for a camera bulging out a bit there...

    Model S facelift does not have the dual camera housing. Maybe that is what they are now installing into Model S facelifts starting recently. I wonder if they only installed the old style until parts ran out and now are installing something similar to what the Model X has had from the start. Fitting three cameras into the windshield does not necessarily sound like a useful idea. Two there and one in the nose might be plausible?

    Tesla Autopilot 2.0: next gen Autopilot powered by more radar, new triple camera, some equipment already in production
     
  7. J1mbo

    J1mbo Member

    Joined:
    Aug 20, 2013
    Messages:
    610
    Location:
    UK
    Putting them behind the windscreen means that they are easy to keep clean, plus have the best possible view of the road.
     
  8. AnxietyRanger

    AnxietyRanger Active Member

    Joined:
    Aug 22, 2014
    Messages:
    2,228
    Location:
    EU
    But is putting all three behind the windshield the best idea. I would find two behind the windshield and one in the nose a scenario that would cover more ground.

    The nose of the car can see things in some scenarios, especially if it has a 180 degree lense, that the driver or the windshield of the car can not - ever so important in e.g. T junctions and autonomous drive, no?

     
  9. Matias

    Matias Active Member

    Joined:
    Apr 2, 2014
    Messages:
    1,590
    Location:
    Finland
    I'm layman, but to my understanding, bitcoin hardware is ASIC?
     
  10. J1mbo

    J1mbo Member

    Joined:
    Aug 20, 2013
    Messages:
    610
    Location:
    UK
    So how does your nose cam see anything when it's raining and there is a car in front kicking up spray?

    Being closer to the front of the car does not mean it can see further - just gets a better view of the widest part of the vehicle in front.

    Would have thought that a clean 180 degree lens with a high viewpoint will be more useful for autopilot, no?
     
  11. AnxietyRanger

    AnxietyRanger Active Member

    Joined:
    Aug 22, 2014
    Messages:
    2,228
    Location:
    EU
    #11 AnxietyRanger, Aug 15, 2016
    Last edited: Aug 15, 2016
    What a 180-degree nose-cam does is see to the sides before any other part of the car does. It is located in the most forward-reaching place of the car, after all. Do look at the Volvo video I linked to above, it shows one very clear scenario where nose certainly sees more and wider than the windshield does.

    I guess what it boils down to me is also this: Would three cameras be needed in the windshield area anyway? Couldn't they do it with two? That should cover both wide and narrow angles (say, 120-degrees and 60-degrees) as well as redundancy. Why would they have three there? They can't make a camera bulge out of the windshield anyway, so its field of view would be less than 180-degrees anyway by physical necessity? What would really add a, literally, new angle of information is a third camera located further to the front of the car, bulging out like a fisheye, perhaps 180-degree one...

    I guess, also, it comes down to nobody else doing three cameras in the windshield, but plenty of cars having nose cameras - it would seem there is some merit in having that angle. A nose camera could be kept clean by a water spray pointed at it, like Audi does with their night-vision nose cameras. Or maybe the redundancy of two cameras in the windshield would mean it is OK to leave the nose camera exposed as is.

    Of course a three-camera Tesla windshield setup is perfectly possible. But I am not necessarily reading the tips we have as suggesting that. The third camera might be elsewhere.

    [​IMG]
     

Share This Page