The rear camera needs a physical re-design. When driving in rain it quickly gets covered with distorting water drops so FSD can't see clearly and disables.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
FSD Beta 10.69
- Thread starter Buckminster
- Start date
I'm in the OG 100 group and haven't been sent the update so they are limiting access quite a bit so far.Haven't gotten the 69.1.1 version yet: I'm part of the May crowd.
But if Tesla is being slow about letting me in, I understand.
I'm in Central NJ, one of the most congested places in the U.S.. Our roads weren't laid out by drunken cows (that's Boston's story, and it's sticking to it), but we got our fair share of Colonial-era ideas about how to route roads; they're not pretty. On top of that, NJ has this historical bit where, as competing railroads extended themselves west, every little po-dunk rail stop acquired a small town which promptly seceded from the larger townships. The larger township's farmers would maintain the roads in the townships in lieu of taxes; the small towns each thought they were the next best thing to NYC or Philly. As a result, there's lots of vaguely straight roads between NYC, Philly, and points south, but going NE or SW.. no two towns ever agreed on a straight road that went between them. A major E-W road terminates on one of those SW-NE roads, people hang a left (or right), then go into the neverwoods (there's actually a town by that name..) to go cross-country. Road maintenance and planning was, shall we say, an extreme afterthought.
So, my 20-odd mile (as the crow flies) commute is along a NW<->SE axis. There's easily, depending upon where the latest accident was on I-78, the GSP, I-287, the NJT, local roads, etc., and normal variable congestion, a dozen ways to get from point A to B and/or back. Local roads, 4-way stop signs, 3-way stop signs, six-lane local roads with lights, occasional rotaries, narrow roads going over hills with heavy woods on both sides and no lane markings and everybody going 20 mph faster than the limit, one lane roads splitting into three lanes as it goes over a hill, where one must be in the middle lane to go straight (and 12.2 always gets the lane wrong), interstates with a two-lane on-ramp merging onto the five-lane main artery, and one must shift over three lanes to go straight, and it's always congested, and so on and so on. It's no surprise that on a given day that little video recorder button get hit anywhere between 5 and 20 times a trip, never mind both ways.
I don't know who it was at Tesla that let me into the Beta back in May, but, by now, I'm betting they sincerely regret that move. (Assuming that it wasn't a robot. Do robots feel regrets?) So, if I'm dead last on the 69.1.1 brigade, it's not going to be surprise. I mean, Tesla wants feedback.. but there must be limits.
It's a problem with any exterior camera, and interior camera without some sort of wiper. It's the reason that all the AVs (Waymo, Cruise, Zoox) don't operate autonomously in the rain. Even LIDAR has problems, since water can accumulate on the sensor and refract the incoming laser. There is work being done to compensate for this type of interference in LIDAR. It will take time for cameras.
AlanSubie4Life
Efficiency Obsessed Member
He did not say anything about new features. They’re probably just playing around with the parameters of existing features to see what they get. I’d expect there will be regressions, but not clear if they’ll be obvious enough to prevent rolling it out wide.
Tweaking isn’t great and closed-form solutions are ideal, but seems like inevitably engineering ends up here.
At this point Elon definitely seems to be in the phase of recognizing that all the problems haven’t been magically solved by the new (and very important and significant and crucial for future progress) features. But he’s still clinging to hope that a couple of parameter changes will get things over the finish line. Unbelievable, but that is where we are!
Imagine parameters like:
Turn on creep network earlier
Adjust creep parameter describing how much view is required to stop creeping.
(As mentioned) Gap requirements
Adjust creep speed downwards.
Make stopping more aggressive.
Reduce delay between stopping and enabling creeping (see item 1)
Adjust jerk profiles (I’ve got to be in there somewhere, haha).
Increase max crossing speed.
Etc. obviously this is just ULTs.
Last edited:
powertoold
Active Member
Whatever Tesla is doing, the practical improvement from 10.69 to 1.1 seemed significant (from YouTubers). So if the jump from 1.1 to .2 is similar from .69 to 1.1, then that'd be great.
Perhaps they've done things like narrow the required gaps for ULTs. That would likely fall in the tweak category once they have determined that it's safe to do so.
AlanSubie4Life
Efficiency Obsessed Member
Would be interesting to see more of CULT, just so we can measure more easily (probably possible from some of the other existing videos already though), and eliminate some variables. But my guess is we won’t get that change (and any test of it) until .2
AlanSubie4Life
Efficiency Obsessed Member
Tried my ULT/UPL again today (see link in prior detailed post). The creeping line was farther back (weird how it is different). But then it completely botched the entrance to the merging turn lane, missing it entirely and slowly lurching into the far lanes of traffic (no one was coming, it was completely cleared, so no risk and no intervention. This is what I expected: inconsistency! I made the mistake of using it with my wife in the car:
“I feel unsafe when you use this, so why do you keep using it with me in the car?”
So I guess that puts an end to that use case!
This inconsistency that we see everywhere is so interesting to me. What is it about their approach that leads to such variability? And how do they address this (it’s definitely not good even if it allows adaptability to many subtly different situations - you need consistency and adaptability!)?
“I feel unsafe when you use this, so why do you keep using it with me in the car?”
So I guess that puts an end to that use case!
This inconsistency that we see everywhere is so interesting to me. What is it about their approach that leads to such variability? And how do they address this (it’s definitely not good even if it allows adaptability to many subtly different situations - you need consistency and adaptability!)?
Spouse Acceptance Factor (SAF) is below the minimum. Another parameter to tweak?
FSDtester#1
Craves Electrolytes
I have noticed today that teslafi has added about 10 more cars over the last few hours... this happened yesterday as well. Do we think they may just keep pushing it out little by little?? 
well this is just peachy - the 12v in my Model 3 just decided to start to fail.
My car has nice messages like this
Schedule service to replace low voltage battery
Software will not update until battery is replaced
I'm presuming it's saying that because there is a software update pending. Oh well.
My car has nice messages like this
Schedule service to replace low voltage battery
Software will not update until battery is replaced
I'm presuming it's saying that because there is a software update pending. Oh well.
EVNow
Well-Known Member
Why do those tweaks in 69.2 instead of 69.1 or 69.1.1. ?
Seems to me they are waiting for some changes to be completed and tested internally before rolling them out.
Get the battery replaced quickly - I have a friend with an M3 where the battery died completely, and getting into the frunk to replace it was a bit of a pain because you cannot open the frunk from your app or inside the car without the 12V.
Saying this cautiously (as in, nobody really knows for sure how Tesla's organized internally) I get the impression that there's more than one group doing development over there. That's not unusual: when there's, say, 20 people working on the final release of 3.6, there's another bunch working on 3.7, a smaller crowd on 3.8, and one or two on 3.9. At the same time the systems engineers have dumped out the inspected 4.0 requirements and are busy messing around with features on 5.0. At least, that's the way I've seen it run on Big Development Projects, and those projects were likely smaller than what Tesla's R&D efforts are like.
I wouldn't be overly surprised that Elon is only tweeting about, using the above example, the 3.x tree, seeing as that's the one with visible results, safe enough for the CEO to drive around in, and there's a high likelihood, given how close to release a 3.x package might be, that a delivery time may be forecastable. If he did mention the 4.0 or 4.5 releases, it'd be really fuzzy statements, if anything.
Thing is, though, is that every time Tesla puts out a major release, they learn stuff and the systems engineers get busy again. It's quite possible that a 3.x (especially 3.9.1.1
) is doing the manipulating-the-variables local minimum trick in a particular architecture. But a different, 4.0 or 5.0 architecture may have substantial improvements over 3.anything.The above is definitely an optimistic, glass-half-full view of what Tesla is up to. As FSD-b testers, we only see what's come before and what we have right now; the future is somewhat obtuse. For all we know Tesla may have realized that there's no more there, there, and they're stuck. Get stuck badly enough and it's going to affect profitability and negate quite a few of Tesla's statements. Which would make the SEC really, really irritated. In such a case, a whistleblower could make quite a bit of hay, NDA or no NDA.
But there haven't been any whistleblowers like that, at all, from the inside. So, for now, I'd put my money on Tesla's people chugging along and coming out with better and better stuff. Maybe even hitting Level 3 despite those filings with the California DMV. (Remember: There's likely people rolling the dice inside Tesla as well as outside. What's better: Saying one can get to Level 3, then failing, or saying that one is going to do Level 2, but then managing to pull Level 3 off? The first attracts regulatory attention; the second, not so much.)
We'll see when we get there.
Interestingly, I have never had an issue with the repeater cameras nor the b-pillar cameras. The front cameras are taken care of by the main wipers, so it's really just the rear that's significantly occluded. Worse in winter when it's always caked in salt. There could be a way to design a surround that creates air currents that minimizes grime from affecting the camera. Barring that, a water+air jet combo seems like an ok solution.
Usually speed/wind are enough to move water off the pillar camera cover, and same for blind spot cameras. The rear doesn't get wind as much from its location.
Yeah, you’re pretty much right at the four year mark (warranty expiration) and that seems to be when they are failing often.. kind of a PITB but at least it’s predictable.
FSDtester#1
Craves Electrolytes
The "SAF" factor is 0 at my house. I have had the Plaid for a year. She has actually been inside it 2 times. Once me driving her to pick up her minivan from service and the other me picking up her and her best friend after they were drinking.
"Electric cars are stupid!"... and that's a quote from her...
She may wake up from her anti EV media induced coma in the future.
Here's to hoping...
Edit: My 3 kids love it and so do their friends, they see the future!
AlanSubie4Life
Efficiency Obsessed Member
For safety, and to be cautious, as he said. .2 will be a staged rollout like every rollout, too, so they’ll get to check the results in real world as well as their engineering vehicles, before it goes very wide.
It makes perfect sense. Certainly makes a lot more sense to adjust parameters in a point release than to add features!
I doubt it.
FSD is insignificant for Tesla as long as there is no competition for these features, which there currently is not. No one else is doing this so whether or not they are successful doesn’t matter much. It’s not clear anyone will be able to do it anytime soon. And none of Tesla’s statements should be materially impacted by any schedule slip (since they definitely are not expecting anything any time soon!).
Their business is building and selling cars (and hopefully trucks!), not automating them!
no, the rear gets a ton of wind, just the wrong type. There's a drag vortex back there which is responsible for that whole back side getting dirtier than the rest of the car. I would think there's a way to design a spoiler just for the camera to prevent so much grime from affecting it.
Regardless, it's really just the rear camera on the current design that poses a dirtiness problem for FSDb.
Similar threads
- Article
