Jump to content

Poor SAS performance


Recommended Posts

When I started using SAS, I was a bit surprised how poorly it hunts for the intended target, often overshooting it, especially on craft with poor attitude control. I often have to help it and actually counteract its inputs, slowing it down as it approaches the target while SAS is still trying to accelerate towards it. It looks like the system's gains are tuned for high performance attitude control (working well using vectored engine thrust, with RCS or on very light craft with sufficiently large reaction wheels) but are woefully wrong when control is more limited. I can see it continue to accelerate towards the target while it's totally obvious it's going to overshoot and it ought to be slowing down.

It would be nice if SAS would take the available torque into account so it stopped accelerating towards the target and started decelerating a lot earlier to avoid overshooting.

Link to comment
Share on other sites

When I used the word "target", I meant whatever target direction the SAS system has been tuned to. Prograde, normal, maneuver, whatever. When it's not already pointing that way and, for example, you switch it to hold prograde, it starts turning towards prograde but then overshoots it horribly, comes back, overshoots again, etc, until finally stabilizing. Almost invariably, I actually need to counteract its inputs. While it's rotating to the right and approaching the symbol, I have to push "A" to slow it down before it gets there (or cheat by activating and deactivating time warp when it passes through the desired direction). Otherwise, on a relatively large rocket using standard pod control wheels, it takes forever to stabilize.

Link to comment
Share on other sites

Yeah, I see. I tend to use MJ's SmartA.S.S., that feels a bit better in this respect... but does the same if the craft is huge and has poor controls. Guess it's not that easy to precisely compute. I kinda' do what you do - corect the controls manually, or do it in full manual. Though I always blame my craft when I need to do so - I could have included sufficient RCS or reaction wheels.

Link to comment
Share on other sites

8 hours ago, NathanKell said:

We're aware the PID has trouble. When we have time we'll be working on it.

Thanks, good to hear!

One other issue I have with it, quite aside from the PID tuning:  it doesn't seem to be following great-circle paths.

That is, suppose I'm pointing in direction A (and am stationary), and then I tell SAS to point in direction B (prograde, retrograde, maneuver node, whatever).  I would expect the ship to rotate in a "straight line" from A to B.  That is, if you imagine the navball crosshairs scribing a trail across the navball's surface as the ship rotates, I'd expect it to be a great-circle arc on the surface of the navball.

Instead, what the navball seems to do is to follow some sort of "do X, then Y" behavior.  I've never observed it closely enough to work out exactly what it's doing, but I'd guess it's something like "rotate to make angle #1 match, then angle #2", or something like that.  As if a plane were navigating from Seattle to Miami by heading due southeast until the latitude matches, then sharply turning to head due east to the target-- something like that.

 

Link to comment
Share on other sites

15 minutes ago, Snark said:

I'd guess it's something like "rotate to make angle #1 match, then angle #2", or something like that.

It might be doing the equivalent of a 'dogleg' movement. When some older CNC milling machines do a rapid movement, they can't calculate the correct angle to move straight from coordinate x1, y1 to x2, y2. So instead they move in a 45 degree angle until it hits the first position (x or y, whichever is closer), then goes straight to the second. 

Link to comment
Share on other sites

19 minutes ago, FullMetalMachinist said:

It might be doing the equivalent of a 'dogleg' movement. When some older CNC milling machines do a rapid movement, they can't calculate the correct angle to move straight from coordinate x1, y1 to x2, y2. So instead they move in a 45 degree angle until it hits the first position (x or y, whichever is closer), then goes straight to the second. 

That kind of motion is certainly what it feels like.

Link to comment
Share on other sites

I would have actually said it just has a lot of trouble with being close to 90 degrees out from it's target, as it's passing through about 100-80 degrees of separation that it suddenly dives off in a seemingly random direction (probably indicates it's using Euler angles or Vectors instead of Quaternions). Even with Quat's, if you do the calculations in single precision the roll error gets really jumpy around 90 degrees (found that one out when I was calculating the shortest roll distance for my SAS mod)

Another oddity to note with the current SAS PID modules, the module used for the direction following has damping even when it's only under proportional control. I'd guess this is so the roll controller actually does something without any fixed target, but it does make tuning pitch/yaw for some torque ratio's nigh on impossible.

Edited by Crzyrndm
Link to comment
Share on other sites

On 24.1.2016 at 0:30 PM, Evanitis said:

Yeah, I see. I tend to use MJ's SmartA.S.S., that feels a bit better in this respect... but does the same if the craft is huge and has poor controls. Guess it's not that easy to precisely compute. I kinda' do what you do - corect the controls manually, or do it in full manual. Though I always blame my craft when I need to do so - I could have included sufficient RCS or reaction wheels.

Yes, mechjeb has an better controller who don't overshoot so much. I found that using SAS keep prograde during aerobrake tend to drain batteries hard. 
 

Link to comment
Share on other sites

This thread is quite old. Please consider starting a new thread rather than reviving this one.

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...