In fact, I did see that post before I replied, and my superlative comment was over 4 drives and ~60 miles (I literally meant the day, not a drive). Also, FWIW, in my experience with all previous versions, it has seemed better on earlier days than later days outside of the minor improvements everyone notices as whatever version you are on seems to remember things. That having been said, 10.6.1 could have installed faultily for me, or it might behave more normal from now on. Regardless, I felt that I had enough drives and miles to justify my comment, which may weakly support a theory suggesting that improvements in locale can often be paired with detriments in another. If so, it would also exemplify just how big of a challenge truly autonomous driving really is.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Wiki MASTER THREAD: Actual FSD Beta downloads and experiences
- Thread starter jebinc
- Start date
coreylista
Member
Still haven’t gotten the update to 10.6.1 I’ve usually gotten it the day after most people get it, usually wake up to a nice “update” notice.
Seems they might have slowed down the rollout for some reason… phantom slowdowns hitting the rollouts now… just can’t have nice things…
Seems they might have slowed down the rollout for some reason… phantom slowdowns hitting the rollouts now… just can’t have nice things…
Last edited:
I think a generalist solution for FSD is far, far more difficult than a localized one with the same competencies and capabilities. Tesla is doing the former with no intra-country geofencing. So, Tesla is trying to solve the entire problem at once. I agree this could produce uneven results in different locations during its development.
I've definitely noticed that updates seem to cause waves of good experiences and bad experiences across the group. An update hits and Group A has good experiences while Group B says their drives have regressed. Next update hits and the drives are bad for Group A while Group B talks about how much better it is than the last update.
We see this type of variation within the United States with top tier infrastructure by global standards, just imagine the complexity in a worldwide solution
We see this type of variation within the United States with top tier infrastructure by global standards, just imagine the complexity in a worldwide solution
mtndrew1
Active Member
10.6.1 was another step forward for my use case. My test route, which is very challenging, had a 25% reduction in disengagements over 10.5.
It’s getting there. Lots of work to go but it’s getting there. Bring on 10.7!
It’s getting there. Lots of work to go but it’s getting there. Bring on 10.7!
@mtndrew` I find the same behavior that I would have disengaged before I now use the go-pedal to smooth out the drive or let the behavior play out because I've become familiar with FSD's quirks. Makes it difficult to really judge whether the newest installed build is actually better. Do I believe the releases over time have improved FSD yes, but trying to quantify with a number, say 30% is difficult. This makes me wonder when Tesla looks at the fleets disengagement statistics how much of the change is actually attributed to FSD updates vs increased driver experience with FSD.
@mtndrew How much of the 25% disengagements you mention could be attributed to just getting more comfortable with FSD?
@mtndrew How much of the 25% disengagements you mention could be attributed to just getting more comfortable with FSD?
Last edited:
j0shm1lls
Member
We had some heavy rain this morning in SoCal, and on the way to work I got a "Full Self-Driving is unavailable - Poor weather detected" message along with the GUI reverting to the NoA UI. I engaged NoA anyway and when there was a break in the rain it switched back to the FSD Beta UI and starting making decisions again other than 'drive straight'. 10.6.1 felt great in the rain, it did drive a little slower than usual, but smooth lane changes, smooth accel/decel. There was a slightly flooded area off to the right side of the street that I decided to disengage FSD for and move over to the left and as soon as I did, it went back to the NoA UI and FSD didn't return for the remainder of the drive (it was raining heavily though).
It's strange to me that I would get that message when I've seen YT videos of FSD testers driving in some pretty bad conditions and FSD just doesn't care and engages anyway. I wonder if this is new behavior for 10.6.1...
It's strange to me that I would get that message when I've seen YT videos of FSD testers driving in some pretty bad conditions and FSD just doesn't care and engages anyway. I wonder if this is new behavior for 10.6.1...
EVNow
Well-Known Member
So, in what ways is 10.6.1 worse ? The reason I ask is so that others can check those too. I feel - the changes are not location based - but rather scenario / use case based. So, if in a particular situation if you find something is worse, likely it will be worse for others (atleast some others) in other places that are similar too.
EVNow
Well-Known Member
Its been happening for sometime. Whether FSD Beta engages in rain - depends on how heavy the rain is and how well NN can see the lines etc, I guess. I sometimes can engage FSD but not at others. Also many times I can engage FSD - but it will bail out soon.
EVNow
Well-Known Member
Yes - it looks like after all the complaints the car
- Doesn't turn a bit to the left for a right turn / vice-versa.
- Takes a sharper turn so as to reduce crossing the yellow line.
I noticed it in my drives, but couldn't confirm from the videos. DirtyTesla mentioned this in his drive too. I'm going to look for this in upcoming drives.
So,
- First there are reports of the FSD curbing wheels, so they change FSD to take wider turns
- There are reports of unsafe crossing of lines, so now they tighten the turn.
Hopefully they have got an optimized middle and not gone back to curbing the wheel !
mtndrew1
Active Member
I don’t believe many/any of the improvements are due to me becoming more comfortable. My route is extremely challenging and disengagements are solely to prevent the risk of collisions.
I also track interventions (applying accelerator, manually going around obstacles) and they increased inversely to disengagements. So where in the past an event could have risked a collision, now FSD just needs some help with the accelerator and usually that’s just for the comfort of other motorists.
I’m pretty methodical at tracking disengagements for that specific reason. The system is becoming more capable.
FSD Tracking
Log Interventions,Disengagements Date,Firmware,FSD Version,Map Version,S1 Int,S2 Int,S3 Int,S4 Int,S5 Int,S6 Int,S7 Int,Total Interventions ,Int/mi,S1 Dis,S2 Dis,S3 Dis,S4 Dis,S5 Dis,S6 Dis,S7 Dis 11/1/21,2021.36.5.3,NA-2020.48-12628,0,4,2,2,2,1,2,13,0.56,3,3,5,2,5,2,0 11/8/21,2021.36.8.5,NA-202...
EVNow
Well-Known Member
Wow - this great.I don’t believe many/any of the improvements are due to me becoming more comfortable. My route is extremely challenging and disengagements are solely to prevent the risk of collisions.
I also track interventions (applying accelerator, manually going around obstacles) and they increased inversely to disengagements. So where in the past an event could have risked a collision, now FSD just needs some help with the accelerator and usually that’s just for the comfort of other motorists.
I’m pretty methodical at tracking disengagements for that specific reason. The system is becoming more capable.
FSD Tracking
Log Interventions,Disengagements Date,Firmware,FSD Version,Map Version,S1 Int,S2 Int,S3 Int,S4 Int,S5 Int,S6 Int,S7 Int,Total Interventions ,Int/mi,S1 Dis,S2 Dis,S3 Dis,S4 Dis,S5 Dis,S6 Dis,S7 Dis 11/1/21,2021.36.5.3,NA-2020.48-12628,0,4,2,2,2,1,2,13,0.56,3,3,5,2,5,2,0 11/8/21,2021.36.8.5,NA-202...docs.google.com
I'd suggest a few runs with each release to rule out just random variations.
mtndrew1
Active Member
While I would like that, the route takes nearly two hours through miserable LA rush hour traffic. I just don’t have enough time on my hands to do it more frequently than once per release.
In any event, it’s becoming more capable on my drives. LOTS of improvements needed but it’s improving.
EVNow
Well-Known Member
Didn't realize it was such a time consuming route.
Hopefully the length and complexity in the route itself provides a cushion against random variations.
Similar to NoA, it seems to appear/disappear based on wiper speed - as soon as you hit that 2nd or 3rd level of wiper frequency for long enough, it is no longer available. However, if you engage FSD before the car senses the downpour, or when you're stopped under a bridge, it seems to stay engaged, which is probably how we get those videos...
Haven't had a poor drive in awhile until today. Drove in an old Massachusetts city settled in 1652 and the streets as you can imagine are difficult to say the least. I had to constantly disengage and finally just quit using FSD all together. A key problem was the route it took. Maybe it was a minute or two faster but the streets were narrow with a whole host of difficult lane changes. I know the city well and always take a much easier route. Tesla should consider modifying routes to reduce the complexity for FSD in challenging locations like this.
Last edited:
I believe early FSD Beta (from last year) did not have the FSD poor weather message and seemed to behave good enough (but it did notify of NoA poor weather). My guess is FSD Beta introduced 360º with memory neural networks a few months ago, and the training data has focused mostly on good weather or at least does not have enough diversity with poor weather, so that's why FSD Beta actually ends up turning itself off, which also seems to suggest Tesla believes regular Autopilot is more capable in poor weather for now.
Ditto. Looking at TeslaScope it looks like the 10.6.1 rollout has been paused.
Sky is clear for Bay Area, finally able to test 10.6.1.
I am experience the aggressive lane change to pass slower car. In one instance, it was about less than half a mile for a left turn, I was in the left lane, speed limit was 40, we were going about 35 and increasing, FSD decide to change lane to the right and make the pass. I ended up had to disengage because the car it was trying to pass got up to 40, so I slowed down and reengaged FSD.
Another is getting confused in bike lane/street parking lane.
I was making a right turn on to Branham Ln, initially, the car turned ok, but right before it straighten out, it try to go into the bike lane, then adjust back, and back to bike lane again before I finally took over.
10.5.0 didn't have this issue for me. But, I don't know if 10.5 never "see" the bike lane and just ignore it, and 10.6.1 can see the lane, and is just confused on what to do; OR, 10.5 see it, knows what to do, and 10.6.1 broke it.
I am experience the aggressive lane change to pass slower car. In one instance, it was about less than half a mile for a left turn, I was in the left lane, speed limit was 40, we were going about 35 and increasing, FSD decide to change lane to the right and make the pass. I ended up had to disengage because the car it was trying to pass got up to 40, so I slowed down and reengaged FSD.
Another is getting confused in bike lane/street parking lane.
I was making a right turn on to Branham Ln, initially, the car turned ok, but right before it straighten out, it try to go into the bike lane, then adjust back, and back to bike lane again before I finally took over.
10.5.0 didn't have this issue for me. But, I don't know if 10.5 never "see" the bike lane and just ignore it, and 10.6.1 can see the lane, and is just confused on what to do; OR, 10.5 see it, knows what to do, and 10.6.1 broke it.
Similar threads
