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

FSD Beta 10.69

Let's goooo

100% success in heavy traffic

To me, that's impressive! Clearly some improvements have made it since the current version.

I feel Tesla has improved enough on Chuck's ULT to move on to other challenging scenarios such as roundabouts (especially multi-lane ones). Oh yeah, and that super slow slow-downs! This has got to be an easy fix.

Hopefully, someone will soon report a bunch of black Model Ys showing up one morning on a big a*s complex roundabouts on their neighborhood...
 

AlanSubie4Life

Efficiency Obsessed Member
Oct 22, 2018
14,910
19,264
San Diego
Clearly some improvements have made it since the current version.
I don't think this looks significantly different than 10.69.2. Maybe there are some subtle differences (specifically the creep limit maybe?). Unfortunately never got to see the ULT in traffic with 10.69.2. But I would guess they behave nearly identically.

It does seem like there have been tweaks since the initial roll of 10.69 though.
 
To me, that's impressive! Clearly some improvements have made it since the current version.

I feel Tesla has improved enough on Chuck's ULT to move on to other challenging scenarios such as roundabouts (especially multi-lane ones). Oh yeah, and that super slow slow-downs! This has got to be an easy fix.

Hopefully, someone will soon report a bunch of black Model Ys showing up one morning on a big a*s complex roundabouts on their neighborhood...
A basic roundabout is nothing more than a right turn at a yield sign. This seems like it should be easier to program than Chuck's unprotected left turns. 2 land roundabouts get more complex but even the majority of those don't seem like they should be that tough.

Just from my experience with FSD it seems to have trouble and take too long for unprotected right turns/right turns on yield signs which fits with it having trouble with roundabouts, too.

FWIW, FSD handled 4 roundabouts perfectly for me yesterday. They were straightforward one & 2 lane roundabouts, nothing fancy, but it was still an improvement.
 
I feel like the latest has both improvements and many regressions. I've noticed that it will handle certain lane choices better than it used to, which is nice. Certain turning paths seem improved as well. On the other hand the following things are regressions on common roads i drive:
  • it seems to want to slam on the brakes and come to a full stop for yellow flashing lights that it never did before
  • On a road which it needed to travel straight on for a long ways it randomly decided that it should change into the middle turn lane to "follow the route"
  • The navigation route wanted to make a right at a T intersection, it confidently made a left turn instead, with no traffic present
  • Sitting at a traffic light, the light turned green and it just sat there (I waited about 5 seconds and then used the accelerator to get it going again)
 
  • Like
Reactions: RabidYak
I don't think this looks significantly different than 10.69.2. Maybe there are some subtle differences (specifically the creep limit maybe?). Unfortunately never got to see the ULT in traffic with 10.69.2. But I would guess they behave nearly identically.

It does seem like there have been tweaks since the initial roll of 10.69 though.
I agree. Very minor, but he did mention that it seems to hold creep more firmly before committing. I guess it needs more testing to see if this is the case.
 
  • Like
Reactions: AlanSubie4Life

AlanSubie4Life

Efficiency Obsessed Member
Oct 22, 2018
14,910
19,264
San Diego
I looked at the timings here a little bit more closely:

Minimum gap used: Approximately 5 seconds. (10:06 - 10:11). (One of the issues with this test was that there were relatively few tests of the minimum gap, just by chance, the way the traffic was moving.). Note it actually has 6 seconds, because it starts rolling before the traffic fully passes.

Crossing time, when not stopping in median: 4.5 seconds (10:05 - 10:09.5). Max speed 14mph while crossing. (So 1.5 seconds until car arrives)

Crossing time, when stopping in median: 6-7 seconds (2:28 - 2:34/2:35). Max speed 12mph.

I worry about the inconsistency here (between gap and crossing time). It may use smaller gaps on the near side if it knows it is going to roll it. So maybe if it is not going to roll it it requires a larger gap (this may be why we saw a "failure to go" failure on the last attempt - it may have decided the single vehicle on the far side was going to prevent proceeding)?

Still, I'd like to see more assertive acceleration after the initial commit (to allow driver to intervene), to help reduce this crossing time.

I'd like to have more than 1.5 seconds margin. In my experience driving the a busy ULT recently, that's just too tight to be comfortable.

Seems like an easy solution. It doesn't have to rocket into the road (dangerous for monitoring); it just has to feed in a lot of acceleration progressively and get these crossing times down. And a reasonably fast halt in the median is fine, as long as it is smooth. Get up to 15-20mph, and more quickly! Should be able to get both of these maneuvers below 4 seconds.
 
Last edited:
  • Disagree
Reactions: sleepydoc
More 10.69.2.2 rolling out on teslafi.
Thanks for the heads-up. Here it is as of now with 640 pending.
1663610121607.png


Edit: Just two minutes after posting this my download started...nice. Hopefully this means we are getting close to pushing to 80+ safety scores.
 
After driving on 10.69 for a week and having it do several commutes and side drives, I can honestly say it is one of the biggest regressions I have experienced in a FSDb update. It may be confirmation bias, but I can't really see anything that is noticeably better. On the other hand, there are a number of things that are significantly worse, primarily:
1. The problem I have in bumper-to-bumper traffic on the highway in NoA has made its way to Autosteer on City Streets ("AoCS") - the car wants to be accelerating or braking constantly. If it's not happily going down an empty street at the max speed, it's braking hard or accelerating to adjust speed based on the conditions around it. Reacting to lights turning green, pedestrians entering view, proceeding through intersections, etc. - it has gotten extremely jerky and lost all "subtly" in its driving ability. Some may say this is somehow inherently safer, but it's very uncomfortable for passengers (including the operator) and makes AoCS much less useful as a feature.
2. Lane selection has gotten orders of magnitude worse. You can tell it has migrated away from using map data (as Elon touted) and instead wants to read the lane lines and street markings. However, for many of my drives, that means it is constantly getting into the wrong lane. It will make, e.g., a right turn going into the leftmost lane on the new road, then put on the signal to move to the right lane, then immediately signal to move back to the left lane for an upcoming left turn. And at many places it finds itself in the wrong lane to proceed, the white line goes solid and the car won't cross, and it's screwed and I have to intervene. To be fair, some of these situations require "local knowledge" - having driven there before and knowing the lanes split the way that they do. But, because it can't learn local routes and no longer relies on map data, it will never be better at these intersections or conditions, which is very disheartening.
3. Unprotected left turns and right-turn on red are just really bad now. I don't know if it can no longer see oncoming traffic or if its aggressively "creeping," but I no longer feel safe allowing it to turn right on red or make unprotected left turns if there is any oncoming traffic. It just feels like it wants to lurch out and get hit by the oncoming cars, and I don't have the guts to let it go and see what's going to happen. Again at this stage in the game, it makes AoCS basically unusable as a feature because it can't be trusted at all in these situations.

I don't remember enough about some early versions of FSDb to be able to say if 10.69 is the worst version of FSDb ever, but I am confident in saying this is the worst step back in a single version upgrade I have experienced since v10.5 last November. As far as all those saying it's a significant improvement? I am simply not having the same experience here in Atlanta. I have said several times in the past I would wait for v11 and if it wasn't any good, I would back out of FSD Beta testing. It's getting harder and harder to wait for v11, though (if it is ever coming).
I will say that the update did give me back Sentry and Dashcam recordings to my USB key, so I am excited about that!
 

Yelobird

Active Member
Aug 24, 2020
1,572
2,101
Illinois
I looked at the timings here a little bit more closely:

Minimum gap used: Approximately 5 seconds. (10:06 - 10:11). (One of the issues with this test was that there were relatively few tests of the minimum gap, just by chance, the way the traffic was moving.). Note it actually has 6 seconds, because it starts rolling before the traffic fully passes.

Crossing time, when not stopping in median: 4.5 seconds (10:05 - 10:09.5). Max speed 14mph while crossing. (So 1.5 seconds until car arrives)

Crossing time, when stopping in median: 6-7 seconds (2:28 - 2:34/2:35). Max speed 12mph.

I worry about the inconsistency here (between gap and crossing time). It may use smaller gaps on the near side if it knows it is going to roll it. So maybe if it is not going to roll it it requires a larger gap (maybe why we saw a "failure to go" failure on the last attempt - it may have decided the single vehicle on the far side was going to prevent proceeding)?

Still, I'd like to see more assertive acceleration after the initial commit (to allow driver to intervene), to help reduce this crossing time.

I'd like to have more than 1.5 seconds margin. In my experience driving the a busy ULT recently, that's just too tight to be comfortable.

Seems like an easy solution. It doesn't have to rocket into the road (dangerous for monitoring); it just has to feed in a lot of acceleration progressively and get these crossing times down. And a reasonably fast halt in the median is fine, as long as it is smooth. Get up to 15-20mph, and more quickly! Should be able to get both of these maneuvers below 4 seconds.
The way you micro analyze each milla-second I question if you will ever be satisfied with any result or progress. For most, simply completing a turn they were not able to accomplish Safely is an acceptable result even if it could have been done .0002 seconds faster lol. And just saying something is “easy” doesn’t make it possible. Sometimes good is good enough as long as it can Safely accomplish the task. So many valuable contributions and feedback on this thread but sometimes the micro analysis with numbers that really mean nothing devalue the context. JMO
 

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
Top