GlmnAlyAirCar
Active Member
Why would you signal when two lanes merge into one? There is no ambiguity there.
-
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
Ran past this intersection a couple times today and 10.12.2 seems to handle it without stopping short. So definitely possible it did react to the car... will have to see once I get 10.69.x on mine
For merges in which the sign indicates that a lane ends, anybody in the lane that is ending is changing lanes, so should either signal or yield to the traffic in the other lane with right of way.
But even in an alternate merge, there are decisions to make, so it can be helpful to make it clear to the person behind that you're ready to take the merged lane and they shouldn't try to overtake.
When your lane is the one merging, you should signal out of courtesy, and FSD should aspire to be "better than average human", right?
You must signal if you cross a white line - I was taught - such as a dotted white line at a merge point. This video has no line at this particular merge point (some do), but since our lane is ending it is courtesy to signal. For one thing it confers information to cars behind you that something is happening causing you to leave the lane. They are thus advised even if they are well back before the signs.
Like the signalling in / out of roundabouts, there are rules, courtesy, and some people may not know/obey any of them but they still exist for the "above average driver".
Tesla/Elon focusing on Chuck's ULT was a great move from their perspective purely because of the focus on that turn and the reception now received, this is really top tier marketing. Throw in a price increase on the back of the release to YouTube reviewers and a tweetstorm and man, it doesn't get any better. I'm honestly in awe at the effectiveness.
My pessimistic self says this should be worrying because Tesla needed to secretly send a team to this turn after Chuck has been grinding away at it for over a year. And Tesla very clearly does not want people to know it's happening, in no small part because I think it pokes holes in the idea that mass public testing is even required in the first place. Chuck was doing that, pressing the camera button endlessly as we all watched, and now it's magically solved after we know Tesla had a team there.
The mystery and lack of understanding around this technology plays heavily in Tesla's favor
My pessimistic self says this should be worrying because Tesla needed to secretly send a team to this turn after Chuck has been grinding away at it for over a year. And Tesla very clearly does not want people to know it's happening, in no small part because I think it pokes holes in the idea that mass public testing is even required in the first place. Chuck was doing that, pressing the camera button endlessly as we all watched, and now it's magically solved after we know Tesla had a team there.
The mystery and lack of understanding around this technology plays heavily in Tesla's favor
I don't get why the geometry and acceleration needed to do Chuck's turn wasn't nailed down in simulation. It's clear that the car had more problems than just live traffic & intersection perception in pre-10.69 versions. It stuttered left & right, hesitated, drove slowly across oncoming traffic. Don't they simulate and solve?
I see you got laughed at for saying this, but I think it's right and I posted a very similar opinion up thread.
It's true that Elon has been overly optimistic, to say the least, in his predictions of upcoming FSD accomplishments. And he should learn that, and be a lot more careful in his public comments, but it's just not his style. One thing I've noticed, particularly when he's speaking on the earnings calls, is that he'll explain one of his mantras like "prototypes are easy and production is hard", and then repeat it almost verbatim, two or three times more as the meeting goes on. A more polished communicator would not just replay the tape, but formulate the point in different words, and coming from different angles, as a reinforcement and an indication of its importance.
Despite this observation, I wouldn't trade the accomplishments he's brought for someone else's smooth communication of lackluster results. Peter Rawlinson of Lucid (a company I'm rooting for but starting to lose hope) gave his very articulate Masterpiece Theater explication of how they would be immune from Elon's Production Hell scenario, due to Lucid's superior planning (and plenty of funding). Following this, they've proceeded to underperform dramatically vs their production-ramp plan, and their quarterly results meeting was pretty embarrassing, with lots of articulate and highly vague dithering. I'm no financial expert, but it's pretty easy to see which one of these two electric car CEOs is building a track record of success.
Last edited:
Virtually all issues that FSD has today existed in Oct 2020. Tesla will work them when and if they decide to work them. You don't necessarily work all the simple ones first.I wonder if Elon Musk knows that James Locke has been sending back video snapshots of FSD Beta not signaling for 2 lanes merging into 1 as recently as March 2022 but also going back now literally years to December 2020 (and probably October 2020). I would not be surprised if Tesla has over 1TB of video snapshots (with actual ongoing costs to Tesla) from his vehicle just for this one priority problem of his to use the turn signal.
Hopefully this doesn't mean 10.69.1/.2 will be delayed for the rest of us.
Simulation is not a substitute for the real thing. It's a tool that (hopefully) reduces the amount of real world testing.
Huh? How was Tesla trying to hide their activities? Did I miss where they asked/requested that Chuck not share that he saw Tesla testing his turn?
Simulation is also needed and will be more important as the march of 9's increases. Need to simulate the unlikely events that don't occur irl much.
Well they have touted the quality of their simulation environment. I would expect that upon arriving at Chuck's turn, a well-prepared software version would just have to map the intersection, look at the traffic and have a pretty good idea how to proceed. I would say it didn't appear to be well-prepared at all, so why wasn't simulation done beforehand. Or why wasn't it very effective? Seemingly anyway.
Maybe I'm moreso just thinking about Chuck trying to respect their privacy, although he tweeted that they were clearly given a "when confronted by Chuck" coaching in terms of what they can say.
But really I'm taking shots at the idea that big data is the key here, which is the public perception, and then the key to solving this turn was still sending their own human drivers to the turn despite having likely the most dedicated FSD Beta tester running it for a year prior and providing all the clips you could probably imagine.
I really want to see this rolled out more widely so we can get a deeper look and allay any fears that the system has been "overfit" to this one turn
But seems like you would work the broad spectrum problems first before working on the more particular issues. For example, I would think there are a lot more people experiencing, for example:
1. Curvy two-lane road with double yellow line (FSD can't stay in lane), or
2. School zones (FSD doesn't recognize or slow down), or
3. Rail-road crossings (FSD doesn't slow down to speed bump speeds), or
4. Cars stopped in two-lane road waiting for oncoming traffic before making a left turn (FSD wants to go around), or
5. School busses stopped to pick up/drop off children (FSD wants to go around),
etc. - none of which Autosteer on City Streets can handle today - instead of unprotected lefts across fast-moving multi-lane highway with occluded left-hand view and a median. I can't even think of where such an intersection exists in the city of Atlanta.
If all it took was simulation, then there wouldn't be 100,000 beta testers.
AlanSubie4Life
Efficiency Obsessed Member
It’s hugely improved, but not solved, to be clear. And maybe with another couple weeks of tweaking it will work every time; I don’t know. We’ve seen two clear failures so far (one other debatable).
The failures are odd, since all the capabilities seem to be there - which is great, and wonderful; it might even be
.It is weird that they (apparently?) can’t just simulate this though. They probably are, and of course they will do actual “on-the-ground” validation testing as well, but if that’s the case they should have seen (and fixed) all the issues we have seen so far in simulation…shouldn’t they? I can’t make sense of that, because I think they actually likely are simulating this.
Yeah well, you & I are missing understanding each other here. I'm saying simulate then test, for some reason I'm getting the sense you feel I'm saying all it takes is simulation. What I'm suggesting is their simulation seems poor if the car can't perform acceptably in live testing, on something that really should already have been ironed out in simulation.
Similar threads
- Article
