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

This site may earn commission on affiliate links.
I thought you were talking about your driving. What insurance company do you have?

State Farm.

It’s not aggressive at accelerating by any stretch of the imagination, so not sure how your insurance company is measuring this. I wonder if they are detecting initial jerk. There are situations where it will accelerate rapidly for very brief periods of time, and perhaps that general jerkiness is causing this.
I'm sure the initial jerk is some of the problem, but I think it is just based on G-force, so it doesn't matter if the acceleration is a for a millisecond or minute. Each "event" counts against the score.

Very surprised at this low acceleration score, since FSD’s about as aggressive as a sedated golden retriever.
That certainly isn't my experience. It nails it out of corners. (It starts to creep around them slowly, determines it is clear and then "floors" it. Accelerating WAY faster than I would.) It is also agressive when the stoplight turns green and I'm the first vehicle.

In a 13 mile trip it got dinged for 2 accelerations events around turns, and 2 from stop lights. (But weirdly it was never enough to trigger an aggressive cornering event, which I occasionally do when driving manually.)

And it got nailed for 2 hard braking events, both prior to turns. (Not the same corners that it got nailed for acceleration on.)
 
LOL.
The first time I drove this road my son-in-law who owns a Model Y was driving and he quickly decided to disable FSD for the rest of the drive. Today my son was with me who also owns a Model Y and he is thinking maybe he should wait before subscribing. I was a bit surprised FSD still failed after 8 months of updates.

American roadways are proving to be a tough nut to crack and they've only just begun to 'roll it out.'

Even level 2 is a pickle. It's impossible to train every real world roadway scenario. It's impossible to know how many edge cases there are as there will always be more over time. And so far NN generalization isn't getting it done. It might be like the old movie Jaws - we're gonna need a bigger NN, hardware design, Dojo computer, memory, sensor suite, power source, etc.
 
  • Like
Reactions: sleepydoc
American roadways are proving to be a tough nut to crack and they've only just begun to 'roll it out.'

Even level 2 is a pickle. It's impossible to train every real world roadway scenario. It's impossible to know how many edge cases there are as there will always be more over time. And so far NN generalization isn't getting it done. It might be like the old movie Jaws - we're gonna need a bigger NN, hardware design, Dojo computer, memory, sensor suite, power source, etc.
All you are saying is that you do not know how to solve the problem.
 
  • Disagree
Reactions: Misterbee
It nails it out of corners. (It starts to creep around them slowly, determines it is clear and then "floors" it. Accelerating WAY faster than I would.) It is also agressive when the stoplight turns green and I'm the first vehicle.

I guess everyone has their own definitions!

Not sure what the insurance company’s threshold is but I guess about 0.4g seems about right for typical driving. (7 seconds to 60mph.)

Telsa’s drivetrains are so ridiculously smooth though that you can do quite a bit higher without feeling too aggressive. Once it gets above about 0.6g then it starts to feel pretty brisk I guess, but still depends on the roll-in of the acceleration (jerk).

Seems to me the insurance company is likely picking up the jerky spikes - I am not even sure that 0.3g is sustained for any length of time with FSD Beta (have not measured, maybe someone with a decent accelerometer could).

Good to see the performance of FSD documented in this manner in any case. It really is horribly rough, and it’s a shame, because it seems completely unnecessary, and you don’t even get fast driving as a tradeoff. I usually cannot determine why the system is being jerky. The control systems stuff must be a complete mess. May be time for a complete rewrite!

Using FSD beta with a non-Tesla driver monitoring device does not sound like a wise choice
I would conclude that using the State Farm app in a Tesla is a means of extinguishing all joy from ownership, based on the parameters used for acceleration. Not limited to FSD Beta.
 
Last edited:
  • Funny
Reactions: Daniel in SD
We did a long drive today - all through city roads. Good thing is I could select a non-highway route using alternate routes !!! Well done, Tesla.

Bad news - wife drove one way i.e. no FSD. I used FSD while coming back - but disengaged at potentially complicated intersections for WAF. Funny thing - kids kept telling my wife she drove worse than FSD ;)
 
We did a long drive today - all through city roads. Good thing is I could select a non-highway route using alternate routes !!! Well done, Tesla.

Bad news - wife drove one way i.e. no FSD. I used FSD while coming back - but disengaged at potentially complicated intersections for WAF. Funny thing - kids kept telling my wife she drove worse than FSD ;)
Hilarious 😂, my kids tell my wife exactly the same thing!
 
  • Like
  • Funny
Reactions: JohnB007 and EVNow
Disappointing FSD drive. 6 months ago FSD approached this intersection where it should have kept to the left, instead the car jerked back and forth so badly I had to disengaged. First drive on 10..69.3.1 and it was just as bad if not worse. I let FSD continue since no cars were near me and FSD just stopped in the red box right in the intersection

The mapping in this old Massachusetts mill town (Lowell) is just awful. One of the big reasons there is so much variability in how FSD performs.

Funny, I had a major fail in Lowell today as well. FSDb botched this intersection pretty badly, but it's just a typical weird intersection here in New England:


I was heading southbound on Aiken Ave. Aiken and Essex both have stop signs, and first person to stop has the right of way. Irrelevant today as there was no one around but me. Car stopped at the line, then proceeded to accelerate normally to continue on Aiken. Suddenly it slams on the brake right in the center of the intersection. I can see a creep wall on the visualization that represents the edge of Essex. Basically just like on the map view where the car is in the intersection, that's where my car stopped, thinking it was crossing Essex. In reality, it's just continuing on Aiken.

Maybe it's a map issue. But I wouldn't have rated this intersection hard for FSDb.
 
Good to see the performance of FSD documented in this manner in any case.
I'll probably test this trip at least once on each new version I get to see if they ahve improved the acceleration or braking control.

The other interesting thing, that is likely just coincidence, is that it took a significantly different route on FSDb, than it ever has for me before I had FSDb. Maybe traffic on all of the three normal routes was backed up, or maybe FSDb routes differently. (Maybe it tries to route using roads that it doesn't have a lot of data for?)
 
Using FSD beta with a non-Tesla driver monitoring device does not sound like a wise choice. I would discontinue using one or the other for the foreseeable future.
Not really - State Farm runs their discount program so get a discount/rebate based on your driving. Worst case you end up where you were if you didint’ use it all. I basically quit Worrying about it and almost never look at the app.
 
Funny, I had a major fail in Lowell today as well. FSDb botched this intersection pretty badly, but it's just a typical weird intersection here in New England:


I was heading southbound on Aiken Ave. Aiken and Essex both have stop signs, and first person to stop has the right of way. Irrelevant today as there was no one around but me. Car stopped at the line, then proceeded to accelerate normally to continue on Aiken. Suddenly it slams on the brake right in the center of the intersection. I can see a creep wall on the visualization that represents the edge of Essex. Basically just like on the map view where the car is in the intersection, that's where my car stopped, thinking it was crossing Essex. In reality, it's just continuing on Aiken.

Maybe it's a map issue. But I wouldn't have rated this intersection hard for FSDb.
Did you push the report button? Oh, right, there is no report button!

Are only Chuck Cook style videos on Twitter taken in now? They are awesome, but also exit-ordinary, i.e. very, vary rare.

With mechanical issues, we have Tesla Service deal with them, so Tesla does get the feedback about what is going wrong, which is essential to making improvements. Now that the report button is gone, how is Tesla going to know that FSD screws up here? This is an edge case, but we all see some edge cases. It seem unlikely that Tesla can notice these among all the discomfort and wrong route interventions. The "FSD Oops" button was brilliant.

If Tesla does not have some way to catch and log these failures, they will not be able to prioritize and make changes to prevent them in future updates. It is a flood of data, but some system to capture and catalog and evaluate these is essential to the success of FSD.

Kaizen (wiki) works as well on software and AI as it did for Toyota's manufacturing.

SW
 
I'll probably test this trip at least once on each new version I get to see if they have improved the acceleration or braking control.

Seems like a good plan.
Improved control would be a long time coming. Lots of complaints about this over many months as I am sure everyone knows.

Maybe now that they have achieved a rollout to a somewhat wider group, they can work on smoothing things out, prior to wide release.

It is a bit worrisome and mysterious why it is still so rough. Hopefully it is intentional and not due to spurious or noisy detections that get fed into the control loops.

I tend to think it could be a couple years before it is smooth, and even then I am not sure…
 
  • Like
Reactions: Jeff N
Funny, I had a major fail in Lowell today as well. FSDb botched this intersection pretty badly, but it's just a typical weird intersection here in New England:


I was heading southbound on Aiken Ave. Aiken and Essex both have stop signs, and first person to stop has the right of way. Irrelevant today as there was no one around but me. Car stopped at the line, then proceeded to accelerate normally to continue on Aiken. Suddenly it slams on the brake right in the center of the intersection. I can see a creep wall on the visualization that represents the edge of Essex. Basically just like on the map view where the car is in the intersection, that's where my car stopped, thinking it was crossing Essex. In reality, it's just continuing on Aiken.

Maybe it's a map issue. But I wouldn't have rated this intersection hard for FSDb.
The intersections do have a similar geometry where there is a "triangle" in front of the car but as you stated neither appear to be particularly difficult which is why I think incorrect mapping may be the cause. I will try it again later today and see if it fails for the 3rd time in a row.
Affordable Solar Lowell · 137 Adie Way, Lowell, MA 01852
 

Attachments

  • Tiangle Intersections Screen Shot 2.png
    Tiangle Intersections Screen Shot 2.png
    436.8 KB · Views: 70
  • Informative
Reactions: FSDtester#1
Honestly, I can’t imagine they ever reviewed anything we sent them. Maybe select clips that people followed up on with emails.
My assumption was they had to have some sort of automated filtering criteria and/or would review clips if a location was reported repeatedly Or by multiple drivers. There’s no way they could manually review everything, though.
 
Here is my favorite Massachusetts intersection after a year of using FSD. What is yours?

The car approaches the Vinal Square intersection where the light controls 2 streets, (Dunstable & Groton Rd). That alone is bizarre. When you're on Dunstable street whether the light is red or green you don't proceed directly up to the light you turn almost 90 degrees to the right and join Groton Rd. Sometimes FSD tries to drive over the slightly raised concrete barrier directly in front of you but FSD never turns fully to the right like it should. When the light is red cars back up on Groton Rd forcing FSD to either wait until all traffic passes or to fit into the line as you turn left. What always happens is a car on Groton Rd will stop to let you in usually waving you on. Since FSD has no concept of this driving behavior it just sits there. However cars behind you expect you to turn right or if you're going thru the intersection to merge into the line of cars since that is the anticipated behavior. Even when someone stops on Groton Rd leaving you plenty of space to fit in FSD won't merge into that space since there is car already in the road. 100% disengagement rate so far. I know it will fail but hope by sending a few emails to Tesla eventually Tesla will solve it.

I'm sure the original cows and horses who traveled this "path" did better than FSD. And routing you around that intersection is difficult since there is a river to the left and to the right roads are restricted during rush hour traffic to keep cars off narrow neighborhood roads.

Google Maps
 

Attachments

  • Vinal Square Intersection.png
    Vinal Square Intersection.png
    1.2 MB · Views: 44
Last edited: