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

Two Paths (AP On-Ramp to Off-Ramp)

Discussion in 'Autonomous Vehicles' started by J1mbo, Jun 7, 2018.

  1. J1mbo

    J1mbo Active Member

    Joined:
    Aug 20, 2013
    Messages:
    1,265
    Location:
    UK
    EM quote (via Teslarati):

    “There’s a new version of Autopilot that’s rolling out, I think, this week which I think is quite a significant improvement. What you’ll see is that the reliability and capability of Autopilot will increase exponentially over the next 6-12 months. The improvements are very, very rapid.”

    “I was just testing that last night at about 1 a.m. I think we might be able to release something in a couple of months that can do that. We’ve been pursuing two paths. One really complicated path that I think isn’t working that great. And then a simple path that I think will work pretty well.”

    “I was able to able to drive last night, going from highway on-ramp to highway off-ramp using the simplified version of the control system. And I think with some further effort, we can get that out in the next couple months.”

    What could these "two paths" be? AI vs Conventional?
     
    • Funny x 1
  2. daktari

    daktari Member

    Joined:
    Jan 21, 2017
    Messages:
    614
    Location:
    Norway
    EM quote (via Teslarati):

    “ ...exponentially over the next 6-12 months.”
    "...we can get that out in the next couple months.”

    So we now know that the 3 months probably, 6 months definetly from February 2017 is at least 16 months and counting. Then add the 6 to 12 months muliplicated with a factor 3 to 9 to convert from EM-time to normal time.

    Whatever it is, I am pretty sure I will not be able to use it as long as I own the car.

    Sorry for the sour mood.
     
    • Funny x 3
    • Like x 1
    • Disagree x 1
  3. jimmy_d

    jimmy_d Deep Learning Dork

    Joined:
    Jan 26, 2016
    Messages:
    386
    Location:
    San Francisco, CA
    That caught my attention too. We don't know which part of the system was different between the complicated and simple versions (perception? planning? control?) but presumably it was the part that was giving them the most trouble. On these boards there's a strong bias towards seeing perception as the limiting issue and there's some merit to that position. So lets run with that for now.

    Tesla's statements and my own review of AP code and outputs suggests that neural networks are a central and probably the limiting component of their vision perception system today. So what makes a neural network vision perception system complicated or simple? The network design, network training, and network use are all candidates here as is the non-network code used in the vehicle.

    Something that is not well appreciated about using NNs in deployed systems is that the NN code itself is rarely most of the code or most of the complexity in a system. There are lots of NN reference designs that are implemented in a single page of code and even the networks used in systems as advanced as AlphaGo generally reduce to just a few pages of code.

    An example of this is Karpathy's own "Pong to Pixels" blog entry: Deep Reinforcement Learning: Pong from Pixels where he implements from scratch a state of the art TRPO reinforcement learning algorithm (which is one of the more complex approaches being used BTW) to solve the same problem that DeepMind addressed in their 2015 Nature paper. The entire code base for that, excluding math libraries, is 130 lines of python:

    Training a Neural Network ATARI Pong agent with Policy Gradients from raw pixels

    The complexity in a neural network isn't written or designed, it's the part that is *learned* by the network. The code itself is usually relatively simple.

    But for real world applications there's often a huge amount of interface and management code surrounding this small neural network codebase that makes the NN itself usable for the overall application. The size and complexity of this surrounding code depends on what role the NN itself is playing in the application and how well the NN's capabilities match up to what the application needs.

    I'd speculate that the difference between 'simple' and 'complex' in this case comes from reducing the non-NN code either by letting the NN take on more of the overall job or by redefining the job to be a better match to how the NN is currently performing. This makes sense mainly when the NN is working better than initially planned for, or when you are capitulating on some aspect of your original objective.

    Elon's comment was so terse that he could have been talking about almost anything. We have very close to zero specific context here. But this is what ran through my mind when I heard him talk about the simple version working better.
     
    • Informative x 7
    • Like x 4
  4. ewoodrick

    ewoodrick Active Member

    Joined:
    Apr 13, 2018
    Messages:
    1,099
    Location:
    Buford, GA
    It's probably along the same path that Cadillac takes. With a well mapped Interstate system, you can do a lot of things, because you don't need everything. Ramp to ramp would indicate that you don't have to worry about stop signs and pedestrians and the such. But provide routing so that the car can change lanes, switch between interstates and the such. Just about completely providing autonomous driving on limited access roads.

    The hard one is driving on surface streets, where there is no guarantee that the road even exists, and cars can be parked in the middle of the roads, people cross wherever they feel and lines may be hard to follow even for a trained driver.
     
  5. J1mbo

    J1mbo Active Member

    Joined:
    Aug 20, 2013
    Messages:
    1,265
    Location:
    UK
    In my view, the "gold standard" for on-ramp to off-ramp would be that you hit the on-ramp, switch AP on, then the car will drive you to the end of the off-ramp, taking into account any lane changes necessary to get you there. In between that, there may be anything AP handles today, plus automated overtaking.

    I guess this approach would be more complex, as the NN would need a bunch of new skills, and they would probably want to do some sensor fusion with the rear-facing cams to be able to judge speed and direction of the traffic behind the car.

    A more simplistic approach might rely on the driver for overtaking and lane discipline, only adding specific functionality for the on-ramp, road branching/merging (i.e. the first link between AP and navigation) and the off-ramp. Minimum viable product? Delivers something, leaves room for agile (ugh) improvement to get to the gold standard one day.

    IMO the big challenge with this whole thing is being able to safely merge from the on-ramp into lane one without rear corner radars. Maybe they will keep this bit manual, and make "on-ramp" = the point where the car has just entered lane one.
     
  6. croman

    croman Active Member

    Joined:
    Nov 21, 2016
    Messages:
    3,305
    Location:
    Chicago, IL
    Many ramps around me have pedestrian crossings about 10 feet from the mouth of the ramp. Granted, pedestrians must be very vigilant as well but every single ramp onto 90/94 (the biggest highway in Chicago) in the loop (heart of the city) has pedestrians crossing frequently. Many other ramps in suburban areas to that same highway also feature pedestrian and bike lanes.
     
  7. Kanting

    Kanting Member

    Joined:
    Apr 21, 2016
    Messages:
    420
    Location:
    US
    Hopefully he was conducting a good test at 1AM, not just riding the 405 southbound from Wilshire on-ramp to Santa Monica Blvd off-ramp. :)
     
    • Funny x 1
  8. J1mbo

    J1mbo Active Member

    Joined:
    Aug 20, 2013
    Messages:
    1,265
    Location:
    UK
    Coast to coast? :)
     
  9. R.S

    R.S Active Member

    Joined:
    Mar 8, 2015
    Messages:
    1,110
    Location:
    Munich, Bavaria, Germany
    Maybe coast to coast on a very tiny island...

    It's always the same, Elon quotes they make very good progress we will see very soon and soon after that FSD. Then we get some minor improvements and the date for FSD is shifted.

    Back in 2015 he thought FSD would be done in 2 years, because it's an already solved problem. I don't know what he would say if you asked him right now, probably also 2 years.

    I guess eventually we will get there, I am actually pretty sure. But if Elon says exponential progress, all I expect is that when I get my 3 sometime in early 2019, it might be able to read speed signs, like my current AP1 Model S, but probably a bit more accurate.

    That should make it a slightly enhanced autopilot. But still enhanced!
     
    • Like x 1
  10. Kanting

    Kanting Member

    Joined:
    Apr 21, 2016
    Messages:
    420
    Location:
    US
    It should be a lot more complex what they currently have/show and they will need breakthroughs on all fronts.

    Anyway I hope the simple path does not mean a target specification change or a compromise. The description on the Autopilot page uses the words "determine which lane you need to be," "intended exit" and "move to a faster lane," so let's not forget the On-Ramp to Off-ramp essentially means a Level 5 on freeway! The only thing the driver needs to do manually is to set the navigation route and maneuver the cars to the on-ramp, turn the AP on, sit there, and maneuver the car from the off-ramp to the final destination.

    "On-ramp to Off-ramp
    Once on the freeway, your Tesla will determine which lane you need to be in and when. In addition to ensuring you reach your intended exit, Autopilot will watch for opportunities to move to a faster lane when you're caught behind slower traffic. When you reach your exit, your Tesla will depart the freeway, slow down and transition control back to you."

     
    • Like x 3
  11. R.S

    R.S Active Member

    Joined:
    Mar 8, 2015
    Messages:
    1,110
    Location:
    Munich, Bavaria, Germany
    #11 R.S, Jun 9, 2018
    Last edited: Jun 9, 2018
    There is no level 5 on freeways. If it's only freeways, it's geofenced to freeways and geofenced means level 4. But they also say it's a driver assistance feature, with the driver having to be in full control over the car at all times, which demotes it to level 2.

    I hope they will at least make EAP level 3 at some point and drop the full control of the car line, as soon as it's ready. But right now if you buy EAP, no matter how good it gets, you will always be legally required to monitor the car and environment to avoid accidents.

    Edit: That also means lane changing on itself, exiting and entering the freeway doesn't have to be very good. It only working some times is good enough.
     
    • Like x 1
  12. jkennebeck

    jkennebeck Member

    Joined:
    Apr 29, 2016
    Messages:
    92
    Location:
    Richmond, VA
    Was this clipped from the Tesla website? Just curious. As a HW1 owner, I've read various posts saying that On to Off Ramp was a promise for HW1, but have never seen any confirmation.
     

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