FSD Beta 10.69

[AlanSubie4Life]

Strong disagree here. I would have not have slowed down at all.

As I said in the post, that was another option; there is plenty of room. I would say the Miata pulled only when the Tesla started making its move though.

I would definitely like that feature in FSD. Maybe that polish & shine will be added in 10.69.2.

 

[clydeiii]

Alan Subie, it doesn’t have to be perfect. I am fine babysitting a teenage FSD brain in exchange for it driving me most of the time. Great trade off.

 

[Dunsel]

Strong disagree here. I would have not have slowed down at all. The Miata clearly saw the green car coming and pulled way to the right to allow both cars to pass unimpeded. I would never have engaged FSD in the first place and gotten to my destination faster and not angry that the richest man in the world suckered me into paying to beta test his software.

And, most importantly, posted about how you didn’t use FSD beta to better inform us all how FSD beta works.

 

[nvx1977]

Alan Subie, it doesn’t have to be perfect. I am fine babysitting a teenage FSD brain in exchange for it driving me most of the time. Great trade off.

I've always felt this way about AP/NoA too. The system has gotten better over time (with a few hiccups in between), but it has made so many of my drives less stressful. FSDb will get there. It's not there yet, but I don't think they are far from decent L2 on local roads.

@AlanSubie4Life I kinda think at this point Tesla is not going to spend any energy on a more meritocratic system. If FSDb is going to be deployed to everyone who's purchased/subscribed to FSD soon, I think the safety score and strike system will be eliminated. I could be wrong. But I think we'll see something more akin to AP jail. I do agree with your sentiment though. I worked hard for the 100 score right from the beginning. Have never gotten a strike. Been diligent with snapshots and emails. I suspect you are the same? Seems like Tesla would benefit prioritizing initial releases out to this kind of profile.

But I've long stopped caring. I get it when I get it. Back to the old mentality of tesla OTAs.

 

[powertoold]

don't see how you calculated this. The car initially reacted at 9:10. This was 4-5 seconds after the oncoming car was visible and a clear issue!

Nah, it actually saw the oncoming car at around 9:07 and was planning to overtake the first parked car and then move over.

Change the video to 4k and do frame by frame to see when the oncoming car first appears on the visualization.

After about 2 seconds, fsd beta decided it's too risky to squeeze in with the oncoming car, so it decided to stop and wait until it passed.

 

[powertoold]

Unfortunately it is an example of what a human would never do.

Sorry, but I just don't like when people go here. This is obviously wrong and just being contrarian for the sake of hating. Someone who doesn't routinely drive that road would definitely wait for the oncoming car to pass.

Edit: btw, if you watch Cyberowner's videos, he's routinely picky and critical of every little thing fsd beta does. He wasn't in this case, and he lives here, so that is something.

 

[Daniel in SD]

Nah, it actually saw the oncoming car at around 9:07 and was planning to overtake the first parked car and then move over.

Change the video to 4k and do frame by frame to see when the oncoming car first appears on the visualization.

After about 2 seconds, fsd beta decided it's too risky to squeeze in with the oncoming car, so it decided to stop and wait until it passed.

What changed that made it decide it was too risky?

 

[powertoold]

What changed that made it decide it was too risky?

Initially, it thought there was only 1 parked car, but once it moved over to the left a little, it saw there were multiple parked cars, so it couldn't pass all the parked cars in time to avoid the oncoming car.

I don't like taking so much time analyzing these nitpicky details, lol. The details change with every new version, so it's pointless to spend so much time analyzing them. My point was that fsd's planning is improving. It seems to be doing it more in advance and more gracefully than before. I even mentioned that it's not as graceful as the best humans.

 

[AndrewZ]

Alan Subie, it doesn’t have to be perfect. I am fine babysitting a teenage FSD brain in exchange for it driving me most of the time. Great trade off.

Could be fun once in a while if it’s included as a quirky feature, but I definitely would not pay extra for such functionality

I’d pay good money for a Level 3 system that will do long highway hauls and allow me to stop staring out the windshield for hours on end

 

[Daniel in SD]

Initially, it thought there was only 1 parked car, but once it moved over to the left a little, it saw there were multiple parked cars, so it couldn't pass all the parked cars in time to avoid the oncoming car.

I don't like taking so much time analyzing these nitpicky details, lol. The details change with every new version, so it's pointless to spend so much time analyzing them. My point was that fsd's planning is improving. It seems to be doing it more in advance and more gracefully than before. I even mentioned that it's not as graceful as the best humans.

At 9:00 you can clearly see all the parked cars. Though maybe it can't see them as well afterwards. I don't understand why Tesla didn't put the camera in the upper left corner of the window so it doesn't have to poke out so much more than a human.
Yes, it's very cool as an engineering exercise. Hopefully it will be useful someday.

 

[aronth5]

Alan Subie, it doesn’t have to be perfect. I am fine babysitting a teenage FSD brain in exchange for it driving me most of the time. Great trade off.

@clydeiii, Agree. Being obsessively obsessive as you point out doesn't help. If you had 10 different drivers in the same situation they would very likely make different decisions on what to do.

Would the decision by FSD prevent FSD from successfully?
1. Driving my elderly patients safely to a medical appointment
2. Safely navigate to a local store

IMO nope.

 

[Silicon Desert]

My only hope and expectation is that I will notice more improvement upgrading to 10.69 then most releases I've gotten in the past year. Minor point releases aka .1 & .2 are likely to be just minor fixes without any real added "shine".

yup, that is what I have notice. No .2 yet, but .1 was only an minor incremental step upward. I only noticed a change in one area and that was on a round-about where it did a bit better.

 

[powertoold]

it doesn't have to poke out so much more than a human.

This is my final post on this lol. Alan said no human would poke out, but there's a whole list of reasons to slightly poke out:

1) Show passing intent for the oncoming car and any car coming from behind

2) Have a clear view of the oncoming car in case it decides to park or yield to ego car

3) Ego car didn't cross double line with any part of its car, giving the entire lane for the oncoming car

4) Comfort: Make it easier to move out once oncoming car passes

5) Have clear view of the road in case ANOTHER oncoming car comes

6) More reasons I can't come up with in 10 seconds

 

[tivoboy]

This is my final post on this lol. Alan said no human would poke out, but there's a whole list of reasons to slightly poke out:

1) Show passing intent for the oncoming car and any car coming from behind

2) Have a clear view of the oncoming car in case it decides to park or yield to ego car

3) Ego car didn't cross double line with any part of its car, giving the entire lane for the oncoming car

4) Comfort: Make it easier to move out once oncoming car passes

5) Have clear view of the road in case ANOTHER oncoming car comes

6) More reasons I can't come up with in 10 seconds

At this point, any CAR trying to interpret this type of situation IS going to have to “poke out” more than a human. there is just no world where the CAR/AI/DATASET is going to be able to do the volume of parallel processing of current situation input, past situational inputs, INABILITY to actually SEE/VIEW/INTERPRET inputs as WELL or FAST as a modeerjly experienced human can, so a bit more poking out in a KNOWN difficult situation is going to be warranted, expected and prudent.

Which brings up another point.. as the known challenges, difficulties and RISKS of certain intersections or even scenarios develops, evolves and is codified through the accumulation of data, some type of system FLAG will have to be made that indicates to the onboard computational system that THIS (insert challenging intersection here) specific location, with this orientation of the car at this point in time, with these components (TOD, LIGHT, environmental conditions) needs to be approached (no pun intended) with slightly more caution and care than any other intersection or condition than normal.

 

[momo3605]

The car stopping somewhat gracefully to let the Mazda Miata get through:

1) There's car on right
2) I have to cross double line
3) Car ahead, calculate speed, trajectory, and likelihood of having enough space 4 seconds later to pass
4) Should proceed or not? When to stop and where, is it enough room for other car, etc.

New video from Chuck kind of shows the opposite behavior. Very unsure when car ahead is coming and defaults to stopping and weirding out the other driver even though there is plenty of space to move through. Seems it has trouble calculating space

 

[AlanSubie4Life]

Alan said no human would poke out,

In this situation it was not needed unless you were going to do the squeeze. It’s not a big deal to move over a bit like the car did, but if you do it because you intend to go (which is why it did), that is not good. An oncoming vehicle will understand. In this case the Tesla crossed onto the double yellow (and possibly even over it), which is too far, and indicates intent. This is not complicated stuff - it's stuff we do every day.

My point was that fsd's planning is improving. It seems to be doing it more in advance and more gracefully than before.

And my point was that I don't really see any evidence of that. It looks like it can plan to avoid taking right of way from an oncoming car, once it is very close.

I don't understand why Tesla didn't put the camera in the upper left corner of the window so it doesn't have to poke out so much more than a human.

Would definitely be nice in some situations but was not needed here. Tesla will just have to stop a bit further back than a human to get the sight lines, I guess.

it actually saw the oncoming car at around 9:07

And then forgot about it (it disappeared at 9:08, about two seconds after the human first saw it at 9:06) and saw it again a couple seconds later! I’m not sure this observation makes things better! Perhaps this is the reason it did not respond until 9:10? Why is it not consistently visualizing approaching vehicles it has already detected? You can see the path planner does not go to faded blue until 9:10, by the way - it was perhaps fully intending to proceed until then (I'm not sure of the significance of the faded blue vs. dark blue pathing).

What is also apparent in this video is the incredibly fast stop (20-25mph in a couple seconds). Another reason to make the correct decision earlier! Why is the car using the brakes??? (You can't see them in the video, but they are definitely there (you can see elsewhere in the video it's hard to see the brake lights in the visualization with this aspect).)

This is another one of those examples of a very abrupt move happening, which is not apparent in the video (because it's a video). Passengers would be upset. The driver was not in this case, because he knew what was going to happen and he was in control.

I've always felt this way about AP/NoA too. The system has gotten better over time (with a few hiccups in between), but it has made so many of my drives less stressful. FSDb will get there. It's not there yet, but I don't think they are far from decent L2 on local roads.

AP is great most of the time; it's wonderful for long drives, and has improved over time - but I would say even in 2018 it was "close", before a bunch of improvements. I'm not as sure about FSDb (and never have been). In some limited application scenarios, it may be fine, but I just am not seeing the wonderful comfort of being chauffeured around an area with a bunch of intersections. Maybe some day, but I actually think it is kind of far from decent L2 on most local roads. Of course, I hope I'm wrong and they can iron out these difficulties and end up with a nice, natural driving vehicle.

This is obviously wrong and just being contrarian for the sake of hating.

I've made it very clear that my comments are not aimed at being "hating." I just want a system that I want to use, and am able to use with passengers. (Even though I think that's a very difficult task, and not the reason I bought FSD (fortunately).)

And I don't like to claim abilities for FSD Beta that it does not have (yet). We have to be realistic and extremely rigorous (looking at timing and calculating g forces) when analyzing these videos.

 

[Supcom]

New video from Chuck kind of shows the opposite behavior. Very unsure when car ahead is coming and defaults to stopping and weirding out the other driver even though there is plenty of space to move through. Seems it has trouble calculating space

The car acts like it interprets a narrow unmarked road as being one-lane and acts as if an approaching car was on a collision course. So, it takes the best evasive action it can without actually running off the road. Coming to a full stop seems unwarranted. If the other car were a Tesla using FSD beta, I guess the two cars would sit there indefinitely!

This is curious since there are plenty of videos showing Teslas squeezing through very tight traffic.

 

[TS.42]

Give it to me and I will tell you if it works or not. I’m not a paid shill or a YouTube superstar, just a guy with a lot of driving experience as you all have as well and a lot of poor FSD experiences as you. My feedback would probably be more valuable TBH but that release has not hit me yet. I should start a channel.

 

[tivoboy]

It doesn’t appear that well, really anybody (still single digits) has gotten this update this week at all? Other than possibly more employees or extreme insiders - who have no Teslafi or Teslascope or other integrated tracking. Does that mean its just delayed or deferred due to unforeseen reason on function or readiness?

Still hoping to be happily surprised for a more robust roll-out of AT LEAST the “10,000 first week or two” roll out prior to any wider roll out.

 

[hamoneaster]

Yeah I was hoping that 10k roll out would of happened by now. But it looks like still mostly only employees and youtubers. Sadness! Maybe this weekened -- they did use to push new builds during the wee hours of Saturdays. Heh!