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

Wiki FSD Beta 10.4

This site may earn commission on affiliate links.
Why in the world would it even want to go to the left lane when GPS knows it was a right turn coming up?
I think at some point based on traffic it needs to get on the correct lane early on.
And stay there.

Even in "chill" mode, FSD tries to change lanes too much when there is slow traffic. BUT, usually at turns, in turn lanes, the traffic IS slow - duh !

They need to add a heavy cost to changing lanes to get into the wrong lane close to turns. That way the cost optimizer will not try to change. Looks like the highest cost now is time - and FSD would rather try to zig-zag to save time than just stay put.

They should also add traffic congestion as a parameter that increases the cost of changing lanes - and also forces the car to get into the turn lane early.
 
  • Like
Reactions: Tez~La
After a month of driving with FSD - I can see how 0 disengagement driving is possible. There are routes in which I can do 0 disengagement drives as well. Just have to let FSD do a few weird but not unsafe things without disengaging.
Agree, especially allowing for "weird" things. But for me at least it would have to be a cherry picked route(s) since the vast majority of drives have both interventions as well is disengagement. I would also have to be willing to annoy other drivers unless done with no traffic.
 
  • Like
Reactions: EVNow
3 lanes going southbound and we are in the middle lane, GPS says “turn right in 1 mile.“
At the .5 mile mark it changes to the right lane?
at the .2 mark it goes back to the middle lane and at this point there is no room to get in the far right lane.
At the last minute it tries to jump in the far right lane and it was wat to close to a car that was in that lane already so disengaged and skipped that turn.

Why in the world would it even want to go to the left lane when GPS knows it was a right turn coming up?
I think at some point based on traffic it needs to get on the correct lane early on.
This is my biggest complaint and seems pretty easy to change. I see 2 reasons to change lanes. 1 is because you are coming up on a turn and need to change lanes to follow GPS. 2 is because you are behind slow traffic. I believe they need a option for #2 for the driver to ALLOW it to happen. This is how NOA works now on freeways. We need same option for FSD Beta on surface streets. This would make a huge difference for me as the system gets better.
 
But it is not yet feature complete. Are you expecting it to be? This is like giving a student a final grade halfway through the semester. Your expectations are unreasonable and if these are your expectations you should not be part of the program.
I think you have it wrong. Here's why-

I paid for FSD and was very satisfied for what it did at the time I paid. It will be nice when it does more. I later volunteered to be a beta test pilot and do the testing as a volunteer. That work did not require me nor expect me to make excuses for the work in progress that in my opinion failed to work. You seem to feel I don't understand that it is a work in progress. But I surely do. While most of my career I worked in industry as an engineer and owned my own business for a number of years, I also taught at the college level part time. But aside from my past experience, nowhere in my agreement to be a beta tester was I expected to make excuses for what doesn't work. I'm here to both learn what others are experiencing and offer my observations. When I had a student do poorly on a test he failed. When I saw the whole class did poorly then I failed.
 
  • Like
Reactions: K5TRX and sids911
I think both options are valid. I've reported a erratic steering a lot - now I just let it be erratic and see what it does next - so I can report on any further problems.
At first I did this but took a tip from some of the veteran beta testers. I always reported the erratic steering behavior, but now I disengage and once through the turn I reengage AP. Safer that way. I refuse to damage my car for the sake of volunteer testing. Hitting a curb or another car with bad behavior is not in my operation manual.

Someone who knows more than me mentioned a belt that can have excessive wear with this erratic steering. Sounds logical and that something to consider. Same goes for tires and brakes. Erratic driving behavior seems to tax the mechanics of the car. I have observed that when I drive my city test route I get 215wh/mile when I run FSD beta it jumps to 260wh/mile. Is that an indication that FSD on city streets is less efficient too ( at this stage of development)?

On higher speed highway and rural roads the FSD beta and manual driving is much closer in wh/mile.
 
  • Like
Reactions: K5TRX
At first I did this but took a tip from some of the veteran beta testers. I always reported the erratic steering behavior, but now I disengage and once through the turn I reengage AP. Safer that way. I refuse to damage my car for the sake of volunteer testing. Hitting a curb or another car with bad behavior is not in my operation manual.
Perfectly valid. This would be like using AP.
 
Agree, especially allowing for "weird" things. But for me at least it would have to be a cherry picked route(s) since the vast majority of drives have both interventions as well is disengagement. I would also have to be willing to annoy other drivers unless done with no traffic.
My test route is 10 miles of typical suburb high traffic, lights, stop signs, traffic circles, residential streets with one lane due to parallel parking both sides and speed bumps. I have to run in in 6 legs due to no waypoints. It does not do downtown which is worse challenge. I also turn it on occasionally when I am in new places but much less challenging than my local test route. I have yet to do any run where it drove 100% off the highway.

I suppose I could find a route where it would do 100% but so far that hasn't happened.
 
  • Like
Reactions: K5TRX
I think this 10.4 is seriously defective in basic ways that I have not heard testers of previous betas mention. I have turned the toggle off and recommend that until the next update. The frantic steering reversals cannot be good for the steering belt.
 
While most of my career I worked in industry as an engineer and owned my own business for a number of years, I also taught at the college level part time. But aside from my past experience, nowhere in my agreement to be a beta tester was I expected to make excuses for what doesn't work. I'm here to both learn what others are experiencing and offer my observations. When I had a student do poorly on a test he failed. When I saw the whole class did poorly then I failed.
+1. In fact, you SHOULD be critical and report issues .. that's precisely what the beta is for! :)
 
I have 10.4 FSD (2021.36.8.5) and had a question about the Safety Score. Since getting the latest software and firmware, the Safety Score no longer appears in the iOS app. Do I no longer need to be concerned about it?

The reason I'm asking is my wife stopped driving my car because she kept dragging my scores down. After 2 weeks of her not driving, I went from 91 to 100, and got into the Beta program. She has asked if she can now drive again, and I want her to, but honestly don't know if her 91 driving will kick me out. Sorry to bring up the Safety Score again, but I could not find the answer elsewhere on this forum.

The Safety Score is no longer applicable and isn't even recorded anymore.

Now this doesn't mean that your wife can't get you kicked out. She can easily get you kicked out by enabling the FSD Beta in her profile, and completely ignoring the nags.

This would quickly get the vehicle kicked out of the FSD Beta program.
 
+1. In fact, you SHOULD be critical and report issues .. that's precisely what the beta is for! :)

That isn't what the beta is at all.

The beta is a strategy Elon is using to buy time. A way of giving FSD owners something to hold them over.

How do we know this? There is no feedback loop on any of this or any kind searchable database of bugs/location/etc.

All you're doing is spending lots of time/effort sending an email into the ether. Maybe someone will read it, and maybe someone won't.

You can't even report mapping/navigation issues, and get them fixed.

It takes maybe 1-2 days for Apple to fix an Apple Maps issue when I report it, but if I report a similar issue to Tesla its never fixed.

Tesla could easily fix this by having regional representatives to filter FSD issues by region, and to improve the data but they don't.
 
How do we know this? There is no feedback loop on any of this or any kind searchable database of bugs/location/etc.
You don't "know" this. You are just speculating.

Yes, it would be nice if there were an army of analysts to triage all the issues and track them to closure. I guess not a high priority for Tesla. They are using data and telemetry from the field to figure out whats happening and fix issues. Just the way companies do A/B testing. No close loop of issue tracking is necessary.
 
You don't "know" this. You are just speculating.

Yes, it would be nice if there were an army of analysts to triage all the issues and track them to closure. I guess not a high priority for Tesla. They are using data and telemetry from the field to figure out whats happening and fix issues. Just the way companies do A/B testing. No close loop of issue tracking is necessary.

Speculation is the result of an open loop so its what we're all forced to do. There is just some button were supposed to push, but does this button do anything?

It's not a true beta without a feedback loop. A closed feedback loop is necessary when want to get quality data from those beta testing your product. If you don't close the loop you'll lose your greatest asset.

Data and Telemetry won't tell them all the issues that a beta will have.

In fact I can't even test FSD Beta completely on my drive to work because the navigation data is wrong, and instead I have to break the usage into pieces.

Ultimately it doesn't matter what reason they have for not valuing human input. All we know if they don't value it, and so there is no reason to waste out time doing so.
 
Last edited:
  • Disagree
Reactions: K5TRX and Yelobird
10.5 needs to be some big improvements or I will be really disappointed. I'm really tired of the erratic steering wheel movements and hesitations in crossing intersections or making turns. Maybe that will be v11. is that 12 weeks away?

I don't understand why Tesla is promising score 98 in the program when veteran beta testers are complaining most of their problems haven't been fixed yet. Do the small number of programmers need more data? Do they need more of the same reports. Just more testers that will be disappointed in the current state of the art on FSD. Elon said 10.3 will blow our minds. It did that. It amazed me how bad it was and how far we really are from the skilled public release so people can depend on it the same as they do the brake pedal.
I’m out, after making the request last Friday. Enjoy all! ARNR for me, until beta v11. Ciao
 
  • Like
Reactions: Phlier
40FD946A-71C1-4562-9B3F-278F01330D18.jpeg

More fun with interstate traffic control. Here it sees traffic signals on the overpass and tells me it’s preparing to stop for traffic control in 500 feet. Pre-FSD I turned off traffic control response bc it would stop for train control signals on the highway, but with FSD enabled traffic control response is always enabled. This particular example didn’t actually result in braking, but FSD still misclassifies the situation and is confusing
 
  • Informative
Reactions: K5TRX and Phlier