3.75 m parts + SAS = Death Wobble

[Jouni]

It does, in fact. No matter how many times you declare it doesn't, it continues to function poorly. It can be tested and seen with a command pod in orbit. There's really no point to arguing it with you further since it can be easily demonstrated without any accompanying "design" to influence it's behavior.

SAS doesn't function poorly, unless you expect it to be a magical mechanism that stabilizes the ship and points it to the right direction. In order to use it properly, you first have to understand how it works. This includes understanding that it's impossible to select a single set of parameters that works well for all designs in all situations. If SAS doesn't work correctly, you have to change the parameters, the situation, or the ship.

A command pod oscillates in orbit, because the fixed set of parameters doesn't work with objects that have so much torque compared to their moments of inertia. A wobbly rocket wobbles, because SAS falsely assumes that the rocket is a single rigid object, and applies torque accordingly. These two are unrelated issues that have nothing to do with each other.

I'm not saying that the SAS is already perfect and can't be improved. I'm just saying that there are no significant issues with it, as long as you're willing to learn to use it.

[Alshain]

There are certain parts in both the 2.5m and 3.5m range that seem to be very weak. Most of them in the 3.5m range, but the 2.5m batteries and the SAS modules themselves have a tendency to be the pivot point for wobbling. The Poodle in stack can also cause this.

[Korizan]

This might appear to be a unrelated but maybe not.

A couple weeks ago I was having some weird things happen.

Ships wobbling for no reason and and yes the SAS just made it worse.

At one point it got so bad that a station that was sitting stable completely disintegrated as soon as I warped away from it.

I had service bays that would cause a wobble if the doors where closed.

The list goes on

When I started looking for a bug I found someone reported that the claw can break the physics engine when it gets engaged.

I thought it was a bit far fetched but as I had one that was currently in use and it would not be much work to remove it so I went ahead and decoupled it and destroyed it as debris.

Funny enough everything stopped I haven't had an issue sense.

Note - NONE of the ships or stations that had the problem had a claw mounted on them.

Almost like a a bug gets injected into the physics engine and gets replicated to every ship as you use them.

Like I said it may not be the issue and it doesn't happen all the time but there is definitely something weird with the claw and how it interacts in the game.

For now I just don't use the claw

Edited June 11, 2015 by Korizan

[Guest]

SAS doesn't function poorly, unless you expect it to be a magical mechanism that stabilizes the ship and points it to the right direction.

Isn't that what the Stability Assistance System is supposed to be in the first place?

A mechanism that stabilizes your ship, and keeps it pointing in the right direction?

Let me get this straight:

First people complained the SAS wasn't good enough.

In version .21, SQUAD fixed the SAS dramatically increasing the stability of the system.

Everyone was happy.

Then SQUAD arbitrarily decided the SAS was "too good" and turned it into a drunken overcorrecting slob.

Now people are complaining again.

Is that about right?

- - - Updated - - -

Ships wobbling for no reason and and yes the SAS just made it worse.

At one point it got so bad that a station that was sitting stable completely disintegrated as soon as I warped away from it.

I had service bays that would cause a wobble if the doors where closed.

You need to be EXTREMELY careful building things with service bays. Part clipping, or attaching to the wrong node inside the bay will cause a vibration that will tear your ship apart.

[Jouni]

Isn't that what the Stability Assistance System is supposed to be in the first place?

A mechanism that stabilizes your ship, and keeps it pointing in the right direction?

It's a mechanism for stabilizing the ship, but not a magical one. Rotating and stabilizing a spacecraft is a gameplay element, and it's not supposed to be too easy.

I went through MechJeb source code quickly to see what SmartA.S.S. does. As far as I understood it, the core component of SmartA.S.S. is also a PID controller. While stock SAS uses a fixed set of parameters, SmartA.S.S. constantly tunes the parameters, based on available torque and the moment of inertia of the ship. This means that SmartA.S.S. has more or less the same strengths and weaknesses as stock SAS in a ship with the right amount of torque. Badly designed ships will still wobble, while borderline cases remain controllable due to good enough PID parameters.

Then SQUAD arbitrarily decided the SAS was "too good" and turned it into a drunken overcorrecting slob.

When did this happen? I've been playing KSP since 0.22, and while there have been some SAS changes, I've always adjusted to them after a couple of hours.

Edited June 11, 2015 by Jouni

[TimePeriod]

I choose wobble because I want challenge.

I want to sit at the edge of my chair, grinding my teeth, taking frantic mental notes and in a panic spilling my soda.

I choose wobble, for the sake of glory, challenge and the kerbal style of life.

Give me wobble or give me death.

[Guest]

When did this happen? I've been playing KSP since 0.22, and while there have been some SAS changes, I've always adjusted to them after a couple of hours.

Well, the first major change happened in v.21 when the old regular SAS (providing torque only), and the ASAS (providing torque and flight control manipulation), were merged together to make the current system that combines the functions of both. You started playing KSP after these changes were already implemented so you have no experience with the earlier form of SAS and ASAS.

The "wobbly overcorrecting drunken behavior" has only cropped up in v1.0 and later.

Personally I am able to fly around the problem but the fact that it is noticeable, and requires specific corrective actions to compensate for the flaws in the system, leads me to believe this is a problem that needs correcting.

Personally I think the experience level of a Kerbal Pilot, or the tech level of a probe controller, (with the exception of the Fly-By-Wire system which should be absolutely perfect to balance out it's significant cost), should have an in-game function to reduce this SAS behavior.

By the way, I am talking about a properly designed and built rocket having wobbling issues, not some noob hack job.

[Jouni]

By the way, I am talking about a properly designed and built rocket having wobbling issues, not some noob hack job.

The design issues aren't always obvious. Some time ago, I built three similar rockets for a thread like this. All of them are tall and have relatively high launch TWR. Only the third one is structurally sound.

The first rocket is initially fine. Near Mach 1, it starts to wobble a bit but remains controllable. A bit later, the center of mass moves into the upper stage, and the rocket starts to wobble uncontrollably. In retrospect, the fault is obvious. Of course the rocket will wobble, if the primary source of torque isn't rigidly connected to the center of mass.

I added a probe core to the first stage and tried to control the rocket from there. The rocket starts to wobble uncontrollably as soon as I start the gravity turn, even if I disable the reaction wheels in the payload. The addition of the probe core made the joint between the two heavy sections (lower stage+boosters, upper stage+payload) too bendy for the force the lower section was pushing the upper section and the rate I was trying to turn at.

Turning the 2.5-stage rocket into an 1.5-stage rocket solved all problems.

One reason why KSP rockets tend to be wobbly is that their payloads are huge compared to the size of the lower stage. This makes it harder to build structurally sound rockets.

[cybersol]

There is definitely a huge regression in SAS behavior from 0.90 to 1.0. Ships wobble more, plus I experience a lot more orbital jitter due to SAS interacting with gimbals. The regression is present even in orbit, so its not just a new aero thing. I suspect something changed in either the SAS or the gimble code to create this. In this sense 1.0 is much closer to 0.22 than to 0.90.

[Guest]

One reason why KSP rockets tend to be wobbly is that their payloads are huge compared to the size of the lower stage. This makes it harder to build structurally sound rockets.

As I said before I am able to fly around the problem, that being said I very carefully test every rocket I build. You can see the results of my work in my Mission Report thread for the AAP Challenge. My problem is not with rocket design, but the excessive amount of overcorrection induced by the SAS system's interaction with reaction wheels and engine gimbals.

This is new behavior, not old behavior.

[Jouni]

My problem is not with rocket design, but the excessive amount of overcorrection induced by the SAS system's interaction with reaction wheels and engine gimbals.

This is new behavior, not old behavior.

My rockets rarely have this problem, because I try to avoid using unnecessary reaction wheels and fins. Launch vehicles steer with thrust vectoring, which I sometimes tweak down before launch. Most orbital vehicles can manage with command pod torque, along with the occasional use of RCS. I rarely use the reaction wheel parts, except for unmanned vehicles and large ships. Apart from planes and landers, spacecraft generally have all the time in the world, so they don't have to be able to rotate 180 degrees in 30 seconds.

[tutike2000]

As far as I understood it, the core component of SmartA.S.S. is also a PID controller. While stock SAS uses a fixed set of parameters, SmartA.S.S. constantly tunes the parameters, based on available torque and the moment of inertia of the ship.

This is what people mean when they say that the stock SAS is crappy. SAS has wobble even with a trivial construction that is impossible to be 'badly designed' (3 parts: pod + tank + engine), SmartA.S.S. does not.

[DancesWithSquirrels]

SAS does not overcorrect badly, except for individual designs...

Actually, it does. The PID settings used for 'hold orientation' are reasonable - all other orientations (prograde, retrograde, normal, target, etc.) use different PID parameters that lead to really nasty oscillations. This has been confirmed by checking the actual values, which apparently have far too high a derivative scalar.

There's a mod for this very issue - SAS Tuning Fix - along with a discussion of the PID parameters.

Edit: To demonstrate the difference, disable any reaction wheel torque and use RCS for orientation. Set to 'hold', watch your (slow) consumption. Now set to 'prograde' or any other vector, and watch your monopropellant evaporate due to the SAS thrashing...

Edited June 12, 2015 by DancesWithSquirrels

[Jouni]

SAS has wobble even with a trivial construction that is impossible to be 'badly designed' (3 parts: pod + tank + engine), SmartA.S.S. does not.

I've never seen that happen. The closest thing I've seen is that ships with too much control authority often oscillate when using SAS in other modes than stability assist. That oscillation may sometimes turn into wobble, but it won't happen with a pod+tank+engine combination, because it's a rigid structure.

Edit: I'm distinguishing between oscillation and wobble, because their gameplay effects are different. Oscillation may sometimes waste resources due to oversteering, but otherwise it's primarily an aesthetic problem. Wobble, on the other hand, often leads to loss of control.

Edited June 12, 2015 by Jouni

[Sharpy]

You must understand before 1.0 all Kerbals were made equal, and so SAS was as good as any kerbal.

With 1.0 you got specializations and "stars" which decide how good a pilot given Kerbal is. And since normally YOU're the pilot, the quality of Kerbal pilots is reflected by various things like SAS stability or speed of finding the node on the navball when set in SAS. And so, Probes' piloting capabilities were adjusted as well.

I want Retrograde NOW, about to hit Minmus ("too good" an encounter) and I watch the Navball roll through Radial, Normal, Antiradial, Antinormal, Radial again...

Obviously probe cores aren't very good pilots.

[magnemoe]

It's a mechanism for stabilizing the ship, but not a magical one. Rotating and stabilizing a spacecraft is a gameplay element, and it's not supposed to be too easy.

I went through MechJeb source code quickly to see what SmartA.S.S. does. As far as I understood it, the core component of SmartA.S.S. is also a PID controller. While stock SAS uses a fixed set of parameters, SmartA.S.S. constantly tunes the parameters, based on available torque and the moment of inertia of the ship. This means that SmartA.S.S. has more or less the same strengths and weaknesses as stock SAS in a ship with the right amount of torque. Badly designed ships will still wobble, while borderline cases remain controllable due to good enough PID parameters.

I found that hold retrograde and similar using the in game system uses an lot of power compared to mechjeb smartass. I had an problem with runing out of power then returning from Mun and aerobraked solar pannels retracted before entering atmosphere. MK1 pod 2x100 batteries, probe and mechjeb. stock tend to run out just after Pe, mechjeb uses perhaps 30% of electrisity.

[GusTurbo]

I've never seen that happen. The closest thing I've seen is that ships with too much control authority often oscillate when using SAS in other modes than stability assist. That oscillation may sometimes turn into wobble, but it won't happen with a pod+tank+engine combination, because it's a rigid structure.

Edit: I'm distinguishing between oscillation and wobble, because their gameplay effects are different. Oscillation may sometimes waste resources due to oversteering, but otherwise it's primarily an aesthetic problem. Wobble, on the other hand, often leads to loss of control.

Oscillation should not happen just because SAS is active. I think that is the main issue people have.

[tutike2000]

I've never seen that happen. The closest thing I've seen is that ships with too much control authority often oscillate when using SAS in other modes than stability assist. That oscillation may sometimes turn into wobble, but it won't happen with a pod+tank+engine combination, because it's a rigid structure.

Edit: I'm distinguishing between oscillation and wobble, because their gameplay effects are different. Oscillation may sometimes waste resources due to oversteering, but otherwise it's primarily an aesthetic problem. Wobble, on the other hand, often leads to loss of control.

Ah, ok, I meant oscillation. Which I find annoying.

And yes, sometimes it turns into wobble, which sometimes gets amplified in a positive feedback kind of way until the ship rips apart (doesn't happen with pod+tank+engine, but does with larger craft).

[rkman]

Edit: I'm distinguishing between oscillation and wobble, because their gameplay effects are different. Oscillation may sometimes waste resources due to oversteering, but otherwise it's primarily an aesthetic problem. Wobble, on the other hand, often leads to loss of control.

I think to most people "wobble" is any obvious, non-damped oscillation. Based on dictionary definitions wobble is just another word for oscillation.

The amplitude of the oscillation being so strong that it causes loss of control or disintegration is just a worse kind of oscillation/wobble.

The issue is that default part configurations can easily cause oscillations so strong that it causes mission failure - even more so than it does in previous versions - and the game gives no hint how to solve those control-loop issues. Solutions are self-evident only to people with some sort of experience in that direction, which obviously most people don't have.

[Jouni]

I think to most people "wobble" is any obvious, non-damped oscillation. Based on dictionary definitions wobble is just another word for oscillation.

The amplitude of the oscillation being so strong that it causes loss of control or disintegration is just a worse kind of oscillation/wobble.

I was talking about qualitative differences, not quantitative ones. In what I call "oscillation", the entire rocket oscillates as a single rigid body. This is primarily an aesthetic problem in most situations. "Wobble" is when different sections of the rocket oscillate with different periods due to a structurally unsound design. This used to be a more serious problem before 0.23.5, and it can easily lead to loss of control or disintegration.

While oscillation is caused by bad PID parameters, wobble is usually caused by both SAS and SmartA.S.S. treating the rocket as a single rigid body. When the wobble starts, their actions may actually drive it instead of damping it.

[hirschhornsalz]

Oscillation is not a purely aesthetical problem, it consumes electricity or monopropellant.

Also, rockets which fly perfectly straight with "Hold", immediatley starts oscillating/wobbling if switched to "Prograde", even if the rocket already points prograde. This has nothing to do with bad rocket design, it's a badly tuned SAS (yes, it's OVERCOMPENSATING :-))

[rkman]

While oscillation is caused by bad PID parameters, wobble is usually caused by both SAS and SmartA.S.S. treating the rocket as a single rigid body. When the wobble starts, their actions may actually drive it instead of damping it.

What you call wobble can be reduced by adjusting the same PID parameters that cause what you call oscillation. And besides adjusting PID parameters both can also be reduced by adjusting control authority (gimbal range, reaction wheel force). So both are symptoms of a poorly tuned control loop, but the game does more to hide the control loop than to help the player resolve problems with it.

MJ's smartass can actually be made to behave very well by adjusting Tf max in Attitude Adjustment (Tf auto-tuning on).

[Jouni]

Oscillation is not a purely aesthetical problem, it consumes electricity or monopropellant.

There are situations where the waste of resources can be a problem, but they're quite rare. Atmospheric reentry is perhaps the only situation, where this can be a problem with any regularity. Most ships can generate electricity during maneuvers, while RCS is rarely the best choice for stabilizing a ship.

What you call wobble can be reduced by adjusting the same PID parameters that cause what you call oscillation. And besides adjusting PID parameters both can also be reduced by adjusting control authority (gimbal range, reaction wheel force). So both are symptoms of a poorly tuned control loop, but the game does more to hide the control loop than to help the player resolve problems with it.

Adjusting the PID parameters, reducing control authority, reducing thrust, and changing the control part can alleviate the problem, but they don't solve it. The real issue is that the controller tells the reaction wheels/RCS thrusters/engines to turn to the wrong direction, because different sections of the ship have different periods of oscillation. The directions are based on the phase of oscillation of the control part, while they should be based on the phase of oscillation of the reaction wheel/thruster/engine part.

All the partial fixes I mentioned make it less likely that the ship starts to wobble, but if it does start, a PID controller always makes things worse. That's why the best idea is usually to fly wobbly ships without any stabilization.

[rkman]

but if it does start, a PID controller always makes things worse.

It starts due to a poorly tuned control loop, of which the PID controller is a part (which could in principal be used to solve the problem to begin with), that same controller is then likely to make it worse.

That's why the best idea is usually to fly wobbly ships without any stabilization.

Unless one tunes the control loop and control authority to the rocket.

I fly all my rockets with MJ ascent guidance, and most of those can wobble but don't thanks to tweaking of Tf max.

At any rate, you do seem to agree that stock KSP can use some improvement wrt to attitude control.

[Jouni]

It starts due to a poorly tuned control loop, of which the PID controller is a part (which could in principal be used to solve the problem to begin with), that same controller is then likely to make it worse.

Or due to the player rotating a rocket under thrust. Or due to aerodynamic forces. Or because the ship is already rotating quickly, and the rotation must be stopped soon, or you'll going to miss a maneuver node.

A single PID is simply the wrong tool for stabilizing a non-rigid ship. It just can't do the job, except under very favorable circumstances.

Unless one tunes the control loop and control authority to the rocket.

I fly all my rockets with MJ ascent guidance, and most of those can wobble but don't thanks to tweaking of Tf max.

I've flown my fair share of wobbly ships, because it's very hard to make large modular ships rigid enough (and because sometimes you have to improvise). MechJeb handles them better than stock SAS, but eventually it also starts making things worse.

Last year, I decided relatively late into my big Eve mission to use a fuel tanker to deorbit the lander in order to save a little fuel. Pushing a 1600-tonne lander with something attached to it with an 1.25 m docking port was obviously dangerous, so I decided to test it first. This was what happened with any kind of stabilization. Without stabilization, I was able to run the tanker engine at 20-25% (IIRC) and even maneuver a bit during the burn.

Later last year, I built and tested a Tylo lander in the 6.4x Kerbol System. This was what the ship looked like. The weakest link was the nuclear asparagus lander, whose engines were also used in the second transfer stage. MechJeb was completely hopeless with the ship while the first transfer stage was still attached. When I piloted the ship manually and allowed the heading to wander a bit around the maneuver node, I was able to throttle the engines up until around 0.5g.