Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Search results

  1. G

    FSD v12.x (end to end AI)

    An end-to-end architecture doesn't mean it's only trained on examples from real drivers. Near accidents and the final moments before actual crashes are great examples of things that are nearly impossible to collect real training data for: - Humans don't respond very quickly so they probably...
  2. G

    FSD v12.x (end to end AI)

    Teslascope on X was saying it had improvements to in-lane positioning (which has been a minor problem area for me). We'll see when I get one of the point releases...
  3. G

    FSD v12.x (end to end AI)

    Tesla isn't exactly unbiased on their own assessment though. It's preferable from a regulatory standpoint for Tesla if everyone considers FSD a level 2 system. That was four years ago - clearly FSD was going to be L2 for the forseeable future in 2020 (Elon's comments aside, lol). If they get to...
  4. G

    FSD v12.x (end to end AI)

    Definitely less predictable. I'm noticing the "random" in-lane offsetting a lot. Lower predictability is the cost of moving to an AI based planner. In traditional code you can know, precisely, what conditions will make the car do something. Neural nets are more probabilistic in their behaviors...
  5. G

    FSD 12.3 loss of auto park

    We really don't know with much confidence. Probably in the next month or so as they get a V12 build on a 2024 branch that can include the new autopark.
  6. G

    FSD v12.x (end to end AI)

    Matches my experience. It's pretty clear V12.3 is not the release where it learns how to drive in winter. That said, the dirt road performance (which is excellent) gives me some hope that the current architecture is amenable to further training and improvement on snow conditions. Dirt and snow...
  7. G

    FSD v12.x (end to end AI)

    Hopefully just simplifying things. Indirectly over a longer time horizon we’re “training” it by offering feedback via disengagements to identify areas of insufficient training. A lot of people have some big misconceptions about how the feedback and training loop works…
  8. G

    FSD v12.x (end to end AI)

    "Lane lines are just suggestions for peasants" -probably V12 Joking aside, the E2E architecture in 12.3 is introducing some weird regressions that V11 would never do. I happened to catch two of them back to back in a 30 second clip: - First it crosses over a double yellow for no reason - Then...
  9. G

    FSD v12.x (end to end AI)

    Oh good!
  10. G

    FSD v12.x (end to end AI)

    Snow storm is rolling in here in Colorado so planning to do a bit of testing (just on 12.3, not 12.3.1 yet) shortly. I expect we'll have 2-3" of snow for testing. With just a light covering of snow a bit ago initial results are a little sus. On a 50mph highway (2 lanes in my direction) the car...
  11. G

    FSD v12.x (end to end AI)

    I may have missed this in past discussion and videos, but FSD got good at dirt roads! As in nearly perfect. V11 was… unrelaxing… (albeit safe). It would hunt around, unable to hold a straight line and decide where on a wide dirt road to go. V12 is stable and smooth, not too far left so other...
  12. G

    FSD v12.x (end to end AI)

    I stand corrected. V12 doesn’t care what we want. The robots have won 🤣
  13. G

    FSD v12.x (end to end AI)

    Agreed. Chill / Average / Assertive has pretty minimal impact on auto-speed, from my testing. For what it's worth, if you just lightly tap the accelerator auto-speed will bump up by about 5 mph. That increased speed will stick for a while - it seems until auto-speed predicts a different speed...
  14. G

    FSD v12.x (end to end AI)

    I found a strange corner case (not a very important one, but made me laugh): - we're approaching an intersection - navigation calls for a left turn - blinker is on (to the left), car is on track for a normal left turn - I rerouted the navigation so we needed a right turn - blinker stays on to...
  15. G

    FSD v12.x (end to end AI)

    It’s a side affect of an AI based speed control system. V11 logic was go exactly the target speed (unless something exceptional happens) V12 logic is AI predicts the correct speed based on, well, anything and everything. (It won’t exceed your target if you have one set but otherwise the...
  16. G

    FSD v12.x (end to end AI)

    Yup! You described it better than I could have, but that's exactly what I see. Mostly at sub-25 mph as well. Annoying as an attentive driver; but looks fine in all the V12 videos on youtube LOL. Today I tested a bunch around Boulder CO and on roads with lots of bikes. Definitely major...
  17. G

    FSD v12.x (end to end AI)

    After about 90 miles of testing V12.3, this release seems like a very interesting but mixed bag. From videos and reviews, I think I ended up coming in with too high of expectations for smoothness and assertiveness arounds pedestrians. It's more like a 5 steps forward and 1-2 steps back for me...
  18. G

    FSD v12.x (end to end AI)

    If a car enters the inner lane of the roundabout, what do you think they are allowed to do then? Any route from the inner lane involves crossing over the outer lane.
  19. G

    FSD v12.x (end to end AI)

    someone’s got to try it 😄 I generally only test in snow without other cars around and am super attentive
  20. G

    FSD v12.x (end to end AI)

    Indeed. I've clipped a curb once on FSD, and it was in a huge snowstorm with a foot a snow on an early FSD release. FSD was struggling but to be fair I couldn't see the curb under there anyway 🤣. The scrapped up wheel is a badge of honor and my contribution to FSD, lol. I'm pretty carefully...
  21. G

    FSD v12.x (end to end AI)

    The State Farm safety score is pretty ridiculously over sensitive, IMO (compared to Tesla Insurance for instance). It's a lot more fun if you aim for the lowest score you can (I bought the whole zoom zoom pedal I'm going to use the whole zoom zoom pedal)
  22. G

    FSD v12.x (end to end AI)

    Finally got it in Colorado - heading out testing soon! This is the most interested in an FSD release I've ever been, mostly to feel if it's actually as good as people are describing.
  23. G

    FSD v12.x (end to end AI)

    You're not going to get 2023.44.30.25, unfortunately, so not much hope for the next few days. Hard to know how soon a 2024 branch based V12 is ready but it could be pretty soon. #TwoWeeksMaybe
  24. G

    FSD v12.x (end to end AI)

    To fallback on TACC on steering override it does need to keep TACC running continuously, so it can instantly switch over. Otherwise you'd get a weird jerky cutover between when FSD drops out and TACC starts up. (Obviously I don't have an real inside insight into Tesla's software to confirm...
  25. G

    FSD v12.x (end to end AI)

    A theory I've heard about why it's gone is that they don't have enough compute to run: - V12 (actually driving the car) - V11 stack (for visualizations and probably automatic emergency braking) - V-whatever legacy stack that controls TACC (as fallback when you override steer and the car reverts...
  26. G

    FSD v12.x (end to end AI)

    I personally really like this behavior. This is exactly how FSD should be driving. It's a bit more relevant to highway driving, but I personally always aim to keep up with traffic. I don't really care about the speed limit (I'll happily go 15-18 mph over here in CO, when everyone else is, or...
  27. G

    FSD v12.x (end to end AI)

    Does this imply that the auto speed neural nets have some memory of previous speeds? I guess that makes perfect sense. I wonder how long it's lookback period is and how this actually works. Is historical velocity an input to the neural net?
  28. G

    Anyone charge at 800V using a Tesla V4 Supercharger?

    Tesla themselves have made this confusing by using the term V4 Supercharger stalls to refer to chargers with V3 speeds and power electronics. (Also this post below is from over a year ago! Time flies, and I'm amazed V4 cabinets are not out yet a whole year later)
  29. G

    FSD v12.x (end to end AI)

    Hopefully they’re going after parking lots and parking next. That feels a natural next large opportunity
  30. G

    Discrepancy in charge prices

    I could be wrong on this but I thought in car and in app Tesla always shows the energy that went in to the battery. I know my car on L2 charging I can see through Tessie it uses energy to warm the battery or run HVAC that is not shown in the Tesla app. Regardless, it sounds like a completely...
  31. G

    Discrepancy in charge prices

    Yes, exactly. You're charged for what the charger delivers (energy into battery + losses). The app shows Energy Delivered which doesn't include the losses. I'm trying to potentially explain why the Energy Delivered is unexpectedly low.
  32. G

    FSD v12.x (end to end AI)

    To me the "how fast can they iterate" question is the question about V12. How did the car learn to stop so far back anyway? Humans typically err completely in the other direction stopping late, so doesn't seem like they'd accidentally end up with human sourced clips of it stopping early...
  33. G

    $89 Tera Level 2 Charger on Amazon

    Good EVSEs have temperature monitoring. The Tesla mobile connector for instance has temperature sensors in the wall end connect, right at the plug that goes in the wall. The gist of what you're saying is correct - there's no significant power electronics in the EVSE.
  34. G

    FSD v12.x (end to end AI)

    I noticed this as well. Related: to me it looks like the car is stopping a few feet short of the stop line as well. Check out this screenshot from the video. The car is stopped (0 mph indicated!) and if you look at the drone view there's obviously space in front of the car to the stop line...
  35. G

    Discrepancy in charge prices

    FWIW I got similar numbers assuming a very pessimistic charging curve: (5 minutes * 5kw) + (12 minutes * 60kw) + (8 minutes * 100kw) + (3 minutes * 180kw) = 35kwh But we're both ignoring losses. Power into the vehicle != energy getting stored in the battery. There's various sources of loss: -...
  36. G

    FSD v12.x (end to end AI)

    I'd argue areas where humans find driving harder are harder to supervise FSD and other drivers are more likely to make mistakes. Both of these increase risk - which may be motivating Tesla to avoid these areas. It doesn't necessarily have anything to do with how "hard" it is for the car, or...
  37. G

    FSD v12.x (end to end AI)

    This geographically segmented rollout is new for FSD, right? The installs on Teslafi show a pretty clear focus on California (of course), Texas (and neighbors), and the east coast (Virginia, North Carolina, Georgia). I thought it was silly of Tesla to send V12 to new inexperienced testers, but...
  38. G

    FSD v12.x (end to end AI)

    True. And it’s been pretty clear from the videos and experiences that V12 is off highway only. Don’t think there’s really been too much debate over that. It’s how FSD Beta has been historically as well.
  39. G

    FSD v12.x (end to end AI)

    Anyone in Colorado get it, or are they avoiding us due to recent blizzard and snowy conditions?
  40. G

    Is radar used in cars that have it?

    HW2.5 and earlier cars still use the radar
  41. G

    FSD v12.x (end to end AI)

    I know! The number of smart phone videos of the cars screen playing dashcam footage is appalling. Where's the "Share to social media" button? If I drive an iPad on wheels why can't it do iPad things? (Probably a better solution would be able to access dash cam footage via mobile app, if you...
  42. G

    FSD v12.x (end to end AI)

    There's been whisperings about actually smart summon for a while. I'm not sure whether that's more focused on the car coming to you or the car parking on it's own, but at least it seems some work is happening with respect to parking lot navigation. FSD has never really had any ability to drive...
  43. G

    FSD v12.x (end to end AI)

    Is V12 even real?
  44. G

    Colorado Superchargers (location speculation, discussion)

    GJ has an old 6 stall 120 kw V2 install, yes. There are better third part CCS charging options to consider using instead.
  45. G

    Out of Spec doing a Cybertruck vs. R1T vs. Lightning vs. Silverado EV cross country race. Who wins?

    It's rare. But (from what I've heard) if all the factors come together in just the right (or wrong) way it's possible to hit thermal limits on V3 cables: - warm weather - large battery Tesla (CT or newer S/X) - plug in at very state of charge to pull very high current for the longest time...
  46. G

    Really getting annoyed at Tesla - delivery experience

    Unfortunately, this just sounds like a case of the sales and delivery staff being straight up wrong. Tesla has never offered any post delivery FSD transfers. It has always required signing the transfer paperwork and committing to it before delivery. But unless you follow online discussions very...
  47. G

    Out of Spec doing a Cybertruck vs. R1T vs. Lightning vs. Silverado EV cross country race. Who wins?

    I'm gonna go out on a limb here and say CT pulls out a win. I know the range and charge curve aren't as good as we'd like, but the overall charge curve + efficiency makes it reasonably efficient. Even without CCS support I think the CT probably has the best charging options (given it's...
  48. G

    Another Battery Range Concern

    Not sure what you’re talking about here - you can’t set a target arrival percentage
  49. G

    Supercharger - Copper Mountain, CO

    Not too surprised, construction looks like it be a pain winter up there.
  50. G

    Another Battery Range Concern

    As others have shared, that sounds normal. Short trips are both worse for efficiency and, by virtue of being short, more prone large swings in consumption (battery estimate errors, heavy acceleration, heavy braking, etc). If you're really interested in studying the efficiency of the vehicle...