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

FSD Beta : Steering Wheel Jerking

This site may earn commission on affiliate links.
It's an intervention when FSD is performing correctly causing noise in the data sent to Tesla.
I completely disagree. If the path planning is not consistent that is something that needs to be immediately addressed. It points to a perception or planning problem.

I am fairly certain that disengagements like this will not distract Tesla or hurt development.
 
  • Like
Reactions: bedoig
I completely disagree. If the path planning is not consistent that is something that needs to be immediately addressed. It points to a perception or planning problem.

I am fairly certain that disengagements like this will not distract Tesla or hurt development.
The path planner is generating thousands of paths per millisecond with a monte carlo algorithm. There can be small deviations left or right for each one. And as you progress through the turn, new paths are built from a noisy drivable area and different perspective. It's the combination of noise and the slow speed that leads to the rapid wheel movement. Stop human-splaining and just let it drive the car.
 
  • Funny
Reactions: AlanSubie4Life
generating thousands of paths per millisecond with a monte carlo algorithm. There can be small deviations left or right for each one. And as you progress through the turn, new paths are built from a noisy drivable area and different perspective. It the combination of noise and the slow speed that leads to the rapid wheel movement.
Almost like they added too much dither and have degraded the signal-to-noise ratio!

I am not human-splaining. I am just trying to ensure it drives the correct path. And I don’t see what is wrong with that. If their path planning/predictions are so randomized that they jump everywhere, that points to a problem with the degree of certainty that the system has. And that will have to be addressed. I suspect that having excellent perception with a high degree of certainty will be helpful in minimizing path variability.

For a human, the drivable area and road markings are NOT noisy. We know where they are, we know they are fixed in position, and we drive accordingly. So I think having excessive noise on the perception (or excessive noise on the decision making) is going to lead to negative results.

Note I am talking about simple cases here of well marked roads with minimal traffic - a good solid environment where it should be possible to drive smoothly, consistently, and make turns.
 
Last edited:
  • Like
Reactions: EVNow
Why is it bad behavior?

It's uncomfortable, it reduces confidence in the system, it increases wear on the steering system, it increases wear on the tires, it makes other drivers question what you're doing, it potentially results in getting pulled over, it gives Tesla owners a bad image.... How is it not bad behavior? Not trying to be snarky, but I don't get how any of it is good.
 
It sounds like human-splaining. Why is your path correct?

Because I am following the road markings. It does not make sense to jerk the steering wheel to the right in the middle of a left turn with multiple lanes when there is a clearly marked line in the intersection showing where to go, and where your lane is, for example. That’s an error.

For an unprotected left, with a clearly marked center lane area to turn into, it does not make sense to spin the wheel back and forth away from that marked area when waiting to make the turn. It makes sense to point the wheels in that direction as the turn begins. Turning the wheel suddenly to the right, far away from that desired vector, is an error. For now, I am letting it do its weird “pre-turning” before moving, in cases where it is safe, but I intervene if it points the wrong way.

Anyway, back to Elon tweets. I guess this started with comments about jerk? Speaking of jerks, I wonder whether Elon has opined on the look of anyone else’s pp today? And could that be related to FSD? Let me check.
 
The path planner is generating thousands of paths per millisecond with a monte carlo algorithm. There can be small deviations left or right for each one. And as you progress through the turn, new paths are built from a noisy drivable area and different perspective. It's the combination of noise and the slow speed that leads to the rapid wheel movement. Stop human-splaining and just let it drive the car.
Actually that doesn't explain the behavior.

I just counted - it rapidly jerked left & right 6 times yesterday on a simple unprotected right turn with clean visibility on both sides of a T junction. The shown path itself wasn't changing much - so there was no real reason to jerk the wheel.

Besides, it is supposed to find the right cost optimized path. Definitely jerking the wheel around is not reducing the cost - it has a high cost in terms of comfort. Even if it is not n the right path - it just needs to find a new path from the current position to the desired location - and it should be smooth. Not jerky.

The jerky behavior is making FSD very unsafe. It is extremely difficult to figure out where it is going and whether it is driving right.

To me, jerking of the wheel is a clear bug. BTW, when it is turning right without a stop sign, there is no jerky behavior. It is fairly smooth.

ps : Yes - we should move all these wheel jerking to a new thread. Its a very important bug they should fix.
 
Because I am following the road markings. It does not make sense to jerk the steering wheel to the right in the middle of a left turn with multiple lanes when there is a clearly marked line in the intersection showing where to go, and where your lane is, for example. That’s an error.

For an unprotected left, with a clearly marked center lane area to turn into, it does not make sense to spin the wheel back and forth away from that marked area when waiting to make the turn. It makes sense to point the wheels in that direction as the turn begins. Turning the wheel suddenly to the right, far away from that desired vector, is an error. For now, I am letting it do its weird “pre-turning” before moving, in cases where it is safe, but I intervene if it points the wrong way.
So says you. FSD is finding what it determines is a "good" path and following it. It all makes sense if you are try to follow the current path.
It's uncomfortable, it reduces confidence in the system, it increases wear on the steering system, it increases wear on the tires, it makes other drivers question what you're doing, it potentially results in getting pulled over, it gives Tesla owners a bad image.... How is it not bad behavior? Not trying to be snarky, but I don't get how any of it is good.
My wife says my driving makes her uncomfortable.

These are million mile cars. A few extra turns of the wheel aren't going to decrease the utility for most users.

Tesla owners already have a bad image to some. A little wheel turning won't matter much.

I don't think you'll get pulled over for deviating a few inches left or right.

I don't really disagree with your complaints. It just seems like overreaction. I mean so what?

The good is the method gets the car around the corner safely and can be optimized in the future.
 
  • Funny
Reactions: AlanSubie4Life
I don't think you'll get pulled over for deviating a few inches left or right.

Tell that to this guy :)


I don't really disagree with your complaints. It just seems like overreaction. I mean so what?

As a development exercise it's fine and I'm not complaining. However, it's being developed to comfortably transport humans on a roadway used predominantly by humans expecting other users to act like humans, so for now, it's wrong :)

I cringe when my wife drives too, so FSD isn't doing that bad, lol.
 
Tell that to this guy :)




As a development exercise it's fine and I'm not complaining. However, it's being developed to comfortably transport humans on a roadway used predominantly by humans expecting other users to act like humans, so for now, it's wrong :)

I cringe when my wife drives too, so FSD isn't doing that bad, lol.
I think that guy interrupted FSD. He panicked when he saw the cop.
 
I had an incident on 10.4 while slowing (prob ~25mph) and coming to a red light. My path was proceeding straight and not turning. FSDbeta made a burst of very rapid movements (<0.5s) on the steering wheel before it stopped, but it didn't really affect the direction of travel.

I go through that intersection virtually every day and this never happened before on 10.3 or AP so it's a regression.

Actually that doesn't explain the behavior.

I just counted - it rapidly jerked left & right 6 times yesterday on a simple unprotected right turn with clean visibility on both sides of a T junction. The shown path itself wasn't changing much - so there was no real reason to jerk the wheel.

Besides, it is supposed to find the right cost optimized path. Definitely jerking the wheel around is not reducing the cost - it has a high cost in terms of comfort. Even if it is not n the right path - it just needs to find a new path from the current position to the desired location - and it should be smooth. Not jerky.

The jerky behavior is making FSD very unsafe. It is extremely difficult to figure out where it is going and whether it is driving right.

To me, jerking of the wheel is a clear bug. BTW, when it is turning right without a stop sign, there is no jerky behavior. It is fairly smooth.

ps : Yes - we should move all these wheel jerking to a new thread. Its a very important bug they should fix.
 
I had an incident on 10.4 while slowing (prob ~25mph) and coming to a red light. My path was proceeding straight and not turning. FSDbeta made a burst of very rapid movements (<0.5s) on the steering wheel before it stopped, but it didn't really affect the direction of travel.

I go through that intersection virtually every day and this never happened before on 10.3 or AP so it's a regression.
It's not necessarily a regression unless it happens frequently on 10.4. Every time you approach that intersection it's a different set of random path segments. Usually they'll be pretty similar, but you can randomly get a bad set.
 
  • Funny
Reactions: AlanSubie4Life
I just tried to engage FSD for one right turn on the way home. It jerked the wheel so hard to the left that it instantly disengaged bc I was holding the wheel. It was an easy 10 mph right turn with no tricks. I have no idea where it was going turning left. That was the 0.25 sec of FSD engagement for me on my way home
 
Tell that to this guy :)




As a development exercise it's fine and I'm not complaining. However, it's being developed to comfortably transport humans on a roadway used predominantly by humans expecting other users to act like humans, so for now, it's wrong :)

I cringe when my wife drives too, so FSD isn't doing that bad, lol.
Video looks fake to me…
 
It's not a bug. You are carrying enough forward speed that it only takes small corrections with the steering wheel to get you back to the current path quickly.
Its unclear it’s a “bug” per-say, but it’s certainly a failing of the system to work in a reasonable way. I won’t be using it long term if that’s the best if ever does.
 
its behaving differently than a vast majority of users would expect
Everyone used to believe the earth was flat. If you took a vote, would the world be flat? Just because it's not behaving as you think it should doesn't mean it's a bug.
its not the design intention
How do you know this? The car is designed to follow a path and that's what it's doing.

While making a right turn, if you came around a corner and there were cinder blocks in your lane, you would have to plot a new course around them. You would probably turn the steering wheel rapidly to the left. You're still making a right turn, but you are briefly turning left to follow your new path around the blocks.

If instead you came to a stop close the the bricks, you would have to turn the wheel sharply to go around them, assuming you have no reverse.
 
  • Funny
Reactions: AlanSubie4Life