I think, if the wide release is truly a point release as suggested, based on prior patterns I’m not so sure (eg 10.69.3 to employees / early testers, 10.69.3.1 to refine, etc)
-
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
Tesla's 22Q3 shareholder deck showed some updated cumulative FSD Beta miles (estimated from the chart):
It seems like there's a group of vehicles "kept back" on 2022.28.2 (instead of upgrading to 2022.36.x) maybe in preparation for addition with the upcoming FSD Beta 10.69.3.x potentially merged to 2022.28+ release branch. So for those that missed out on 10.69.2, presumably they'll get added as part of 10.69.3 if still meeting requirements, e.g., 80+ Safety Score, then there's potentially enough time for a 10.69.4 for a wide release late November not requiring Safety Score anymore.
For reference, before this year, cumulative miles wasn't even 9M, and the fleet is doing at least that every month now. And with wide release, the same 9M miles could be accumulating every week by the end of this year.
- June 35M (+7M)
- July 42M (+7M)
- August 48M (+6M)
- September 57M (+9M)
It seems like there's a group of vehicles "kept back" on 2022.28.2 (instead of upgrading to 2022.36.x) maybe in preparation for addition with the upcoming FSD Beta 10.69.3.x potentially merged to 2022.28+ release branch. So for those that missed out on 10.69.2, presumably they'll get added as part of 10.69.3 if still meeting requirements, e.g., 80+ Safety Score, then there's potentially enough time for a 10.69.4 for a wide release late November not requiring Safety Score anymore.
For reference, before this year, cumulative miles wasn't even 9M, and the fleet is doing at least that every month now. And with wide release, the same 9M miles could be accumulating every week by the end of this year.
AlanSubie4Life
Efficiency Obsessed Member
Yes. Intervene within 0.5 seconds when it does stupid things! This is the way.
Full disengagement not always required.
Beta tried to go when a light turned green with cars in front of me. Kinda strange and not sure if it would have stopped but I hit the brakes. Here is the (crappy iPhone) video that is a little long and the incident is near the end. Was coming up for a left turn and lead car changed into the left lane and then moved back. My car went into the left lane and I (CLEARLY) saw that a bus was broken down in the lane. My car just continued towards it as if it was stopped at the red light. At the last moment I activated the right turn signal to forced Beta to switch lanes or I would have been blocked by traffic. When the light changed Beta then accelerated forward like it was going to hit the car in front of me.
The first part is just showing some of the crazy obstacles faced one after another when driving in a city. Oddly the traffic was kinda light (mid morning). When Beta turned 3 of 5 lanes were blocked. The first lane by Utility and Beta moved to the 2ed lane which was the best. It then moved over a lane while a construction truck pulled out. Would have been better to stay in lane 2 but…….
The first part is just showing some of the crazy obstacles faced one after another when driving in a city. Oddly the traffic was kinda light (mid morning). When Beta turned 3 of 5 lanes were blocked. The first lane by Utility and Beta moved to the 2ed lane which was the best. It then moved over a lane while a construction truck pulled out. Would have been better to stay in lane 2 but…….
DarkForest
10.69 made me a shill
That’s pretty wild. Beta did something like the first part for me a couple of weeks ago. One lane was blocked with a truck and some workers doing maintenance. I saw it a few blocks ahead, but Beta didn’t do anything about it before I decided to hit the brakes at the last moment.
It plays chicken with other vehicles (stopped and moving) way too much. I happen to know it’ll usually work out, but other people on the road or in the car don’t. It needs to do more than path finding. It needs to show intention as well.
That last bit where it started moving when the light turned green is also super odd. I’ve been wondering if I should be ready to brake at traffic lights, and your clip demonstrates the answer is an astounding yes.
That's my perspective! I'm on 2022.28.4 and in the FSDb queue with a SS of 99 for the past several months. When 2022.36.1 first hit, I was briefly notified this version was available for download (which I had planned to ignore), but then minutes later it disappeared from my system and hasn't come back. To me, this seems like an indication that Tesla intentionally pulled it back and won't push it my way since I've been tagged to receive FSDb in the next push (10.69.3). Just speculation, though...
sleepydoc
Well-Known Member
Nope - the laws are clear. The right/proper/legal choice was to stop. The problem is it’s like coming to a complete stop at a stop sign. very few people do it so other drivers get surprised when it happens. Just because FSDb follows the law when others don’t doesn’t make it wrong.
sleepydoc
Well-Known Member
Well, technically roundabouts are not protected. They’re unprotected with a yield sign, but that’s still a problem. FSDb will frequently stop at yield signs with there is clearly no need to do so.
AlanSubie4Life
Efficiency Obsessed Member
Incorrect. There were no pedestrians (except ones on the other side of the street, not yet in the median). Clear overreaction by FSD beta (not an issue, deliberate design choice), overridden by accelerator.
willow_hiller
Well-Known Member
If there weren't cars following me so closely, I would have chosen to brake for those pedestrians as well. I don't consider them to have been on the other side of the street, I considered them to be in the middle of crossing the entire multi-lane street.
The Federal term for what they were within is called a "Pedestrian Refuge Island", which is designed to allow pedestrians to cross one half of the road at a time, but I cannot find any DC-specific guidance on whether a pedestrian in a refuge island is considered actively crossing or not. But here's a tweet about a new traffic law from nearby VA that they describe pretty clearly:
Emphasis on "This includes *all* lanes & both directions if it’s 2-way road, even with a median."
If I see a pedestrian in the median and I can safely stop, I do. The question is either FSDb is considering the speed of approaching traffic from the rear before considering stopping...
AlanSubie4Life
Efficiency Obsessed Member
Yes, now that is what I call unambiguous.
Some states are not so explicit and the prior verbiage you provided is clearly substantially less strict.
Interestingly, it is fairly nonsensical - since they also say (in Tweet replies) you can go as soon as the pedestrian clears YOUR lane. Kind of crazy that is ok while someone 5 lanes over on the other side of the median you have to stop for. But rules are rules.
AlanSubie4Life
Efficiency Obsessed Member
It also stopped far too quickly regardless. It’s not like you didn’t know 100 feet prior.
All California law states is that vehicles must yield to peds in crosswalks. It also says that peds must use caution and not cross while there is traffic. I was taught if a ped is in a crosswalk, I must stop and wait until they have cleared the walk completely before proceeding. However, that was many years ago.
willow_hiller
Well-Known Member
Agreed. Slowing early gives a signal to pedestrians you're planning to stop, and lets them know to start looking at other lanes to see if other drivers will do the same.
FSDb likely doesn't care about pedestrians across medians, and only stopped because it had a last-minute trajectory projection that placed them in the road before it would have crossed the crosswalk.
I ran it again this morning as the lead vehicle but this time it stopped 1x and well before reaching the center of the intersection.
I forgot to mention the intersection has a relatively large crown for rain flow so that and the dip exacerbate vehicle front/back rocking/porpoising and I assume decorrelates large portions of the camera's vertical field of view. It's a WAG not sure what else it could be.
sleepydoc
Well-Known Member
First, there were pedestrians waiting in the median. Second, they don’t have to be in the street - at least in MN the law requires you to stop for pedestrians waiting to cross. Think About it - if you don’t have to stop until they are in the flow of traffic then you’re requiring the pedestrians to step out into traffic in the hopes that people will stop. How safe is that?
FSD Beta shouldn't be impacted by dips, as the cameras are sampling tens of times a second. It's also why speed bumps don't cause issues with cameras losing their position. I can't tell you why the car is stopping in the middle of the intersection. If there was a warning on the screen, such as a red wheel of death, it would mean the system lost confidence and needed you to take over. Absent any warnings, stopping usually means that the car sensed something it thought was a danger, such as another car or pedestrian. In visualizations, objects the car is watching turn blue. When proceeding through the intersection, are there any objects highlighted in blue (including pedestrians)? It also could be an error detecting signal lights. When proceeding through the intersection, do any traffic signals facing you appear red?
EVNow
Well-Known Member
Yes - but people expect you to just go if there is no vehicle coming. Especially true of normally busy roundabouts. I usually disengage or press the accelerator if there is a vehicle behind me.
Interestingly, FSDb proceeds without stopping if there is a lead vehicle in the front, from my experience. It is also in these cases where it seems to not stop even when a vehicle is coming from the left causing a disengagement.
EVNow
Well-Known Member
Either way - I don't think 69.3.x will be the wide release.
Yes, its possible we won't get 69.3, but probably after a few fixes.
Similar threads
- Article
