[PART, 1.0.2] Anatid Robotics / MuMech - MechJeb - Autopilot - Historical thread

[Gaius]

Sorry, I wasn't trying to be snarky, just clear. I think last time I made a request, I got exactly what I asked for, rather than what I really wanted, due to poor spec phrasing on my part.

I wonder if turning off all Smart A.S.S. but the active one when you dock might be an easier way to accomplish this, rather than mucking around at undock time.

Edited December 31, 2013 by Gaius

[BARCLONE]

Sarbian,

Unintended consequences: Launch vehicles that were perfectly stable in 149 now do perfectly-complete back-flips before they reach 5Km...

I'm also seeing two instances of the Crew Manifest icon on the screen, on independent taskbars. I've looked for multiple instances of dll's, but haven't found any.

[DevL]

I'm also seeing two instances of the Crew Manifest icon on the screen, on independent taskbars. I've looked for multiple instances of dll's, but haven't found any.

Same here.

[Galane]

Not sure if this is a bug or not, but I never encountered it in 0.22 which is why I'm posting about it. With the latest stable release on 0.23, when using the rendezvous autopilot if a capsule develops a roll, mechjeb will not auto correct it. I need to press the roll keys A or D to get mechjeb to hold the roll.

Use the force... roll. In the docking autopilot, force roll aligns the axis of the control part of the docking ship (which normally should be the port you're going to dock) with the target docking port on the ship you're docking to.

Smart A.S.S. force roll appears to align to the SOI the ship is in, with the way the front of the ship points dependent on the OBT or TGT selection or if docking autopilot is active. It definitely does not align roll to a target ship. (Unless that's been changed recently.) Switch between the two while using TGT+ to aim at another ship and you'll see the roll angle change.

[AbeS]

The ones with two crew manifest icons check navyfish's folder there should be a toolbar.dll in there that shouldn't

[Galane]

Yup. I had the same problem last night: decoupled two spacecraft in space, and MechJeb was unable to control the separated craft until I did the "Control from here" thing -- the Navball was properly reflecting the orientation of the ship, but all of the Smart A.S.S. commands were doing things relative to the other ship and freaking out, doing a kind of confused wobble in place as it tried to turn, but the other ship wasn't moving.

I had that with my Kethane satellites at Duna. Apparently when I decoupled them in .22, KSP never set the one and only part on them with control function to be 'control from here' yet earlier versions of MechJeb would control them. Now it requires 'control from here' to be explicitly set or it'll either do nothing at all or will just wobble it around a bit. Dunno if it might also have something to do with changes in .23 upsetting it.

At least the issue is fixable simply by setting control to a part.

Logically when there's only *one* part on a craft that can be set as control, KSP should set that automatically, or if there are more than one part that can be set to control but only one command part (manned pod or RGU) then the command part should get control. If there's more than one command part then if one is the root part it should get control. If no command part is root then the highest one up the tree should get it.

Would be nice to know the actual method the game uses to set which part has control when a ship is undocked or decoupled, and why when a probe with only one command/control part is decoupled that control from here doesn't get automatically set.

Would it be possible for MechJeb to report that no part has control from here set? "Warning! This craft has no part set to Control From Here. MechJeb cannot control this craft until that is set."

[BARCLONE]

The ones with two crew manifest icons check navyfish's folder there should be a toolbar.dll in there that shouldn't

I'm only finding one instance of "toolbar.dll" in the entire game folder, located in "000_Toolbar".

However, I'm now finding five instances of the Crew Manifest icon on the screen. It's breeding...

Spotted something: A new icon appears on the screen each time I go to the launch pad for a launch. There is one icon at the far left of what I'll call the "master toolbar", which also contains the FAR and AlarmClock icons. But there are also (at this moment) four separate and distinct CM icons. They're being created but they're not being destroyed.

Edited December 31, 2013 by BARCLONE

[BARCLONE]

As for the vehicle going unstable, the log only shows this:

[LOG 19:46:28.099] [Progress Node Reached]: FirstLaunch

[LOG 19:46:28.099] [Progress Node Complete]: FirstLaunch

[LOG 19:47:33.837] Packing Frontiersman LV Debris for orbit

[LOG 19:47:33.837] Packing Frontiersman LV Debris for orbit

[LOG 19:47:33.882] Packing Frontiersman LV Debris for orbit

[LOG 19:47:33.882] Packing Frontiersman LV Debris for orbit

[LOG 19:47:34.185] Packing Frontiersman LV Debris for orbit

[LOG 19:47:34.229] Packing Frontiersman LV Debris for orbit

[LOG 19:49:36.685] KW2mengineMaverickV collided into Zn123222333 - relative velocity: 158.2474 - no impact momentum (no RB)

This is the entirety of the flight record, from launch to crash. There are no error entries concerning the loss of control.

I'll try rebuilding my vessel to increase the torque and see if that has any effect.

[Deathsoul097]

In which node on the tree do you unlock the auto-land and landing guidance?

[BARCLONE]

Just tried increasing the torque in my little Stayputnik core to a full 20 units, and still the vessel went full back-flip (although at a much higher altitude). The behavior is the same -- at some point in the ascent, the rocket looses a lock on its heading and starts deviating from prograde, MJ appears to try correcting, but the rocket over-reacts and MJ can't keep up with what it's doing. MJ appears to be causing the back-flip as a result of over-correction (not able to quit correcting for the initial deviation).

[AbeS]

Does it flip when you fly it manually?

[BARCLONE]

Does it flip when you fly it manually?

Of course! That's why I use MechJeb!

Seriously, I'm a lousy pilot when trying to fly manually. If I didn't have MJ, I couldn't make orbit 5 times out of 10. I know what to do, but my finger dexterity just can't keep up.

[AbeS]

hmm, because it could be a FAR related problem. It would be a MJ related problem if the craft could be flown without flipping manually

[Starwaster]

Just tried increasing the torque in my little Stayputnik core to a full 20 units, and still the vessel went full back-flip (although at a much higher altitude). The behavior is the same -- at some point in the ascent, the rocket looses a lock on its heading and starts deviating from prograde, MJ appears to try correcting, but the rocket over-reacts and MJ can't keep up with what it's doing. MJ appears to be causing the back-flip as a result of over-correction (not able to quit correcting for the initial deviation).

hmm, because it could be a FAR related problem. It would be a MJ related problem if the craft could be flown without flipping manually

I have to concur, that really sounds like you're using FAR, and if you're going too fast when you try turning (or MJ tries turning) then flipping is the usual result. A gravity turn should start when you're still going slow (Ferram said 100m/s I think?)

Since you're using MJ, I'd see how high you are when you reach that speed and set your gravity turn to start at that altitude. You also want your rocket to have a starting TWR of about 1.2.

If you're not using FAR and the above isn't applicable, try setting an acceleration throttle limit of 21 m/s

(just FYI, flipping isn't something a real rocket would do.... if it turned that violently at supersonic speeds it would likely disintegrate....)

[Deathsoul097]

So nobody knows where the auto-land and landing guidance is? I am only asking because that is the only reason I want mechjeb. (My best "Pinpoint landing" near another ship, was nearly 5Km away from the target. I can land. Just not accurately.

[CleverLikeMe]

So nobody knows where the auto-land and landing guidance is? I am only asking because that is the only reason I want mechjeb. (My best "Pinpoint landing" near another ship, was nearly 5Km away from the target. I can land. Just not accurately.

Well, if you're asking about the modules within mechjeb, they are there. If you are playing career, you'll need to unlock a specific tech for it to show up, though. For which one, I would google mechjeb career tech tree or some such and that should get you a reference.

[Virtualgenius]

If you go to this post http://forum.kerbalspaceprogram.com/threads/12384-PART-0-23-Anatid-Robotics-MuMech-MechJeb-Autopilot-v2-1-1/page403 post number # 4029 and edit your ar202 case and replace whats there with the post info it will unlock everything at the start of the tree or you can selectively change the pieces you want to unlock earlier

[DevL]

The ones with two crew manifest icons check navyfish's folder there should be a toolbar.dll in there that shouldn't

Doh! That was of course it. Thanks!

[sebi.zzr]

its under unmaned tech along with other modules that dont need crew

i forgot to add quote,answer is for post #4634 @deathsoul097

Edited December 31, 2013 by sebi.zzr

[sarbian]

Gaius : I'm sure you can apreciate that I could read it the wrong way.

If nearly forgot about that problem. I'll look into it while I investgate the other undocking problem that crept in .23

Galane : I'd rather force the ship to choose a new control part than display a message. Something must be broken on .23 on undocking.

BARCLONE : Can you post in the CreW Manisfest thread ? I'll look into it as sooner as I sobber up tomorrow.

I would welcome some feedback on the Tf auto tuning in dev 150+. How does your small / large ship turn with it. If it does not work well please post a screenshot with the attitude adjustment window opened so I can see the new info I added.

The formula need some tweaking but It's nearly there. I may try to use something less linear.

Edited December 31, 2013 by sarbian

[Galane]

With my Duna kethane scanning satellites, it wasn't undocking, it was decoupling. They didn't have root parts. In .22 Mechjeb was able to control them. I did nothing with them after updating to .23 and MechJeb couldn't control them until I set control from here to their OKTO2 cores.

So the not having any part set to control from must have happened in .22 but it having an effect on MechJeb's ability to control the satellites didn't crop up until after updating to .23 as well as updating to some .23 compatible version of MechJeb.

I know a way to further test this. Tomorrow evening (most likely) I'll launch one of my Armed Guard missiles, making sure the root part is not on the Kinetic Energy Transfer Module, and see if the KETM is uncontrollable by MechJeb.

Something I noticed tonight in the aftermath of yet another failed Eve landing experiment is that on a ship with an unmanned 2 man lander can and reaction wheels, but having lost the RGU in the crash, MechJeb can still control the orientation of the craft (I hyperedited the wreckage back to LKO).

Can't control the throttle or landing gear or anything else that requires a Kerbal onboard or a part capable of remote guidance, but SASS can still point the ship this way and that, including using force roll - but no manual control of orientation or anything else is possible and MJ functions such as landing guidance won't do anything with it.

Dunno if that should be called a bug or a feature. ;-) With no Kerbal and no part capable of remote guidance, I'd think a ship should be completely unresponsive.

[sarbian]

Undocking and decoupling is the same thing. You had 1 ship and you now have 2. It did not happen in .22 because they changed the code for .23 ...

[rosenkranz]

I don't have any screen shots but the Tf-autotune does need some ... tuning

So far that I've seen a Tf of .3 works for a fairly wide range of craft. Smaller than 5t and the Tf needs to go down (.15 works for small stuff). For bigger stuff, i'm not sure. It really depends on how much reaction torque you've got on the craft. At a Tf of 2 (per the autotune) a 175t craft is so sluggish it won't even try a gravity turn, it just goes up.

When I get home I'll try making a standard 100t craft and try it at different Tf values and see what it does. Anyone have a good idea what a reasonable amount of torque for that would be? Assuming a mk1-2 pod (15 torque) plus two SAS units (20 torque each) for a total of 55 torque sound right, or am I over killing the torque?

Second question would be this: What kind of metric should I be looking at to grade performance? Just general wobble? Does MJ have any way to know when it has over corrected? Tracking that for debugging purposes would be cool.

[sarbian]

The windows added a MOI / Torque Ratio. That's the relevant stat since it show how fast you ship will turn.

In #150 Tf is that ratio divided by 20 and clamped between 0.04 and 1. It's most likely too linear and I welcome any idea on other formula.

55 for 100t sounds about right, maybe a bit large. Keep in mind that mass is not the best stat to track. MoI take mass and its distribution around the center of mass into account.

Useful metric are

- control wobble with kill rot

- wobble on a 180° turn

- speed of a 180° turn

[BARCLONE]

I have to concur, that really sounds like you're using FAR, and if you're going too fast when you try turning (or MJ tries turning) then flipping is the usual result. A gravity turn should start when you're still going slow (Ferram said 100m/s I think?)

Since you're using MJ, I'd see how high you are when you reach that speed and set your gravity turn to start at that altitude. You also want your rocket to have a starting TWR of about 1.2.

If you're not using FAR and the above isn't applicable, try setting an acceleration throttle limit of 21 m/s

(just FYI, flipping isn't something a real rocket would do.... if it turned that violently at supersonic speeds it would likely disintegrate....)

I am using FAR, the gravity turn is set to begin at about 2-3 Km, the curve is around 80%, the altitude end is around 125 Km, and the acceleration limit is set at 15. I tried going lower on the limiter, but it looked like it tried to hover after completing the flip.

EDIT: Just a few minutes ago, I set up a clean sandbox to try something:

Turn start = 1 Km

Turn end = 150 Km

Turn shape = 85%

Acc limit to 15

The rocket went to nearly 2500 m before beginning the turn, with a V of about 90 m/s. At 4700 m, MJ started reversing the turn and the rocket began doing the back-flip at a V of 163.5 m/s.

Edited December 31, 2013 by BARCLONE