Jittery SAS on small craft

[zolotiyeruki]

Here's a bug I'd love to see fixed:  on craft with low mass, SAS is very jittery on every setting (prograde, retrograde, radial, normal, etc) except for the default (hold attitude).  The craft wiggles around all over the place.  It sure would be nice to not deal with that.

[Snark]

Yes, it's a known issue and KSP players have been complaining about it for quite some time. 

There are mods that deal with it.  You can use Tweakable Everything to adjust the strength of reaction wheels, and I vaguely seem to recall reading somewhere that Claw's stock bug fix collection has a fix for this.

Hopefully it'll make it into the stock game at some point, but until then you either need to keep your torque down, or use a mod to deal with it.

[Warzouz]

A low torque reaction wheel could be added for probes. The existing 0.6m reaction wheel has already too much torque.

Edited March 12, 2016 by Warzouz

[Dafni]

I think of this as Navball drift induced by excessive reaction wheel torque. On my landers that require a lot of reaction wheel torque for some reason I usually assign them to an action group, toggle them off for orbital maneuvers and only let the torque from the pod or the RCS do the work. For flying around on surfaces and stuff the strong torque is very nice a lot of the time, and it only really bothers the system on orbital maneuvers, so its nice to have it available via actiongroups.

[Warzouz]

Ths issue is also with the SAS which over balance the oscillation. MechJeb SAS is much more efficient and doesn't create oscillations.

When I hav a probe where I had to add a small reaction wheel, the stock SAS is useless, so I use MechJeb's.

[GeneralVeers]

My solution to this problem is a nifty bit of top-secret technology I've installed in all my ships.

I call it the "temporal inertial stabilizer". When triggered, the device kicks Time Warp to 5x for a second and then drops it right back down. As soon as the 5x kicks in, all rotations and wobbles are zeroed! Problem solved.

(translation: I hit the > key for a second to stop my ship from spazzing around..... )

[Crocket]

I've also noticed this problem on some of my smaller upper stages (it's especially bad in my Science save for some reason) and I would want it to be fixed at some point.

[Racescort666]

On 3/11/2016 at 9:31 PM, Snark said:

Yes, it's a known issue and KSP players have been complaining about it for quite some time. 

There are mods that deal with it.  You can use Tweakable Everything to adjust the strength of reaction wheels, and I vaguely seem to recall reading somewhere that Claw's stock bug fix collection has a fix for this.

Hopefully it'll make it into the stock game at some point, but until then you either need to keep your torque down, or use a mod to deal with it.

I'm running stock bug fix and it has not fixed this. I have faith in Claw though. Trust in Claw... to bring me to a better place... 

Ok, sorry, today has been a weird day.

[rkman]

On 13-3-2016 at 9:52 PM, GeneralVeers said:

As soon as the 5x kicks in, all rotations and wobbles are zeroed! Problem solved.

And then what happens as soon as you drop out of timewarp and "hold attitude" on your over-torqued spaceship kicks in again?  Problem not solved anymore.

[Snikersnee]

Put it in fine control mode maybe?

[GeneralVeers]

12 hours ago, rkman said:

And then what happens as soon as you drop out of timewarp and "hold attitude" on your over-torqued spaceship kicks in again?

Nothing, cuz SAS is turned off.

[cubinator]

A way to fix this would be to make it so that the SAS generates less and less torque as it approaches the destination node, just like it already does for the 'hold attitude' mode. I think the code is different for the other modes, and that should be changed so that all SAS modes are treated in the same way by the game. Seems like a fairly simple fix, all that would be necessary is for the reaction wheels to generate less torque when approaching the destination orientation instead of going at full power right past it, causing all the jiggliness.

[Racescort666]

10 hours ago, cubinator said:

A way to fix this would be to make it so that the SAS generates less and less torque as it approaches the destination node, just like it already does for the 'hold attitude' mode. I think the code is different for the other modes, and that should be changed so that all SAS modes are treated in the same way by the game. Seems like a fairly simple fix, all that would be necessary is for the reaction wheels to generate less torque when approaching the destination orientation instead of going at full power right past it, causing all the jiggliness.

What you describe is a PID controller, which several people have mentioned, is how SAS should behave anyway. It would be nice to have tweakables for SAS to adjust the sensitivity of each term in the equation but something that didn't overshoot so badly would be just as nice. 

Wikipedia has a pretty good article on PID controllers. We had a whole class on it for engineering and we barely scratched he surface so don't worry if it seems complicated because it is.

[rkman]

On 16-3-2016 at 11:01 PM, GeneralVeers said:

Nothing, cuz SAS is turned off.

Then your contribution to this thread is off-topic.

 

[Enceos]

If SAS is jittery, then it's a design flaw from the user. Don't put reaction wheels for probes which already have built-in SAS. They already have one as weak as you can get in stock to not overcompensate. If designing is hard there's Mechjeb SAS which does the best job.

Edited March 19, 2016 by Enceos

[nli2work]

It's not all user error. The Mk1 command pod jitters non stop with SAS. But on the other hand Mk1 cockpit is rock solid, even though it's SAS is much stronger and not that much heavier than the Mk1 command pod.

Edited March 19, 2016 by nli2work

[GeneralVeers]

4 hours ago, rkman said:

Then your contribution to this thread is off-topic.

Uhh.....no. Saying "I like cheese" or something like that would be off topic (Manchego is delicious, by the way!).

If the SAS is jittery, getting the ship to stop jittering and then turning off the system that causes the jitters is a solution. Probably the best one, in a game that's known to be glitchy. I have not seen the same problems with jitteritis that other folks in here have described. Not with Scott Manley's 2-ton Mun lander, not with space probes fitted with 15-torque SAS wheels. My problems are elsewhere.

As the doctor said:

"It hurts when I do this--OW--"

"Then don't do that!! That'll be $40."

(oh, and I like cheese!)

[zolotiyeruki]

On 3/19/2016 at 9:25 AM, Enceos said:

If SAS is jittery, then it's a design flaw from the user. Don't put reaction wheels for probes which already have built-in SAS. They already have one as weak as you can get in stock to not overcompensate. If designing is hard there's Mechjeb SAS which does the best job.

The OKTO-2 plus a small reaction wheel exhibits this behavior, and that's the smallest one you can get.  As nli2work pointed out, the Mk1 command pod also has the same problem.  Plus, the problem isn't consistent--it doesn't appear on the 'hold atitude' setting, but shows up for everything else.

[Enceos]

1 minute ago, zolotiyeruki said:

The OKTO-2 plus a small reaction wheel exhibits this behavior, and that's the smallest one you can get.  As nli2work pointed out, the Mk1 command pod also has the same problem.  Plus, the problem isn't consistent--it doesn't appear on the 'hold atitude' setting, but shows up for everything else.

Stock SAS markers are indeed terrible, they don't point exactly at the vector but little sideways plus current math doesn't account for overcompensation. SAS is trying to point at the vector as soon as possible which results in overshooting. I use Mechjeb SAS, its what you need as well. I don't know if this issue was tackled in 1.1

[zarakon]

The worst example I've seen is when you just have the Mk1 pod with heatshield and you try to set it to face retrograde on re-entry.  It somehow doesn't even point in the right direction

[Francois424]

46 minutes ago, zarakon said:

The worst example I've seen is when you just have the Mk1 pod with heatshield and you try to set it to face retrograde on re-entry.  It somehow doesn't even point in the right direction

And it'll eat-up your electricity REALLY fast...

[michelcolman]

SAS is indeed awful, but it looks like this should be very easy for Squad to fix.

Current stock SAS flaws (and partial solutions):

- When you turn on a guidance mode (e.g. maneuver), it often starts to turn in the wrong direction initially (up to perpendicular to the required direction), then makes some kind of arc that circles around the target. Strange, because the maneuver arrow is pointing in the correct direction so the SAS should know which direction to apply torque in!

- It keeps accelerating towards the target direction, overshoots it, then comes back, overshoots again, etc... This, too, should be easy to fix. You know how much torque the reaction wheels can produce. When the rotational speed and angular distance to target are such that max torque would slow it down just in time to come to rest at the target angle, stop accelerating and start decelerating. If angular distance is delta, rotational speed is omega, and max angular acceleration is alpha, then you should start decelerating when delta < 1/2 omega^2/alpha. Maybe start at a slightly larger delta and calculate alpha from delta and omega to avoid missing the target due to rounding errors. If you're not directly on course to the target direction, split up the angular velocity in a component towards the target and a perpendicular component and give each a "budget" of alpha/sqrt(2). When the deviation is small (both delta and omega), use only a small alpha but increase it when necessary due to aerodynamic forces.

- It uses a lot of electricity since it's constantly applying large correcting inputs while chasing the target.

Like some people indicated, MechJeb has a much better SAS but it's also quite a bit slower. I think it uses a simple speed feedback loop to limit the speed towards the target. Certainly works a lot better than stock, but I often end up helping it to get there more quickly.

[evileye.x]

Claw's bug fix + all reaction wheels have only 3% of original torque (not so magic wheels)  + persistant rotation. 

[Warzouz]

Yes we need a way to reduce torque. Probes which includes torque have a low value one, so they aren't an issue. Burt Okto2 which as none has to be fitted with the smallest reaction wheel which as a torque of 5. It's way too powerful and SAS code handles it very bad.

The SAS code should auto-dim the torque when approaching centre value.

[michelcolman]

Also, it's not just SAS. Manually controlling a small pod can be next to impossible even with caps lock on. Instead of just reaction wheels on/off, we should have a slider controlling their torque so we can turn it way down if the craft has become much lighter.