Problems with MechJeb and Jools moons

[Tokay Gris]

I build this here lander.

It looks a bit complicated but is not, really.

The basic plan is to first land on Tylo. For this, the lab and the science module (from top to bottom) are decoupled (to stay with the cruiser).

The rest then lands on Tylo. For ascent, the two lateral pylons are decoupled and stay on the surface. Then, the remaining vessel again docks to the science module, is transferred to the next moon (Val, Bob or Pol), decouples from the lab module, lands, lifts up again, docks to the lab again (which is with the cruiser), take data out of the stuff (thermo, Goo, the works), which are then cleared with the lab, transfer to next moon, rinse and repeat.

That is the theory.

Practical tests had these problems:

Landing on Tylo using MechJeb worked, but was off about 10 km.

Ascent worked up until only about 0.5 m/s left on the maneuver node, which then shifts to an angle that does not help anymore but keeps burning anyway (MechJeb that is).

Stopping this by hand (orbit was achieved, although not an exactly circular one). Then, the rendezvous autopilot has the same problems. Burns the nodes up until about 0.6, the node marker then shifts to a kind of perpendicular position, MechJeb keeps burning, but in a direction that does not help any.

Of course, I can do all that by hand, but since this mission profile includes quite a few rendezvous and docking maneuvers, I'd like to really use MechJeb.

So what is my problem here?

Is it MechJeb? I had some trouble at other bodies as well.

Or is it the design? The command capsule below the tank?

The "obvious" problems I checked. I did change the controlling part to one of the big docking ports. The one on top as well as the one facing forward in between. And back again to the command capsule. Power is there. Torque is there. No clipping (I can see).

Anyone has an idea? I kind of like my design and would hate to redo it. But if all else fails, I will. Too many rendezvous and docking maneuvers, as well as too many landings to be done on this mission. And landings that should be fairly accurate. (Rendezvous on the ground with methane rigs is planned.)

Other mods are: KAS (not in this design), S.C.A.N.sat (also not in this design), Kethane (not in this design), Alarm clock and Hyperedit.

The only mod part in this design is actually Mechjeb.

Edited May 10, 2014 by Tokay Gris

[Tokay Gris]

To illustrate the profile some more.

This is what lands on Tylo:

And this is what lands on Vall, Bop and Pol:

[Tokay Gris]

Here is the .craft file.

With MechJeb on it.

http://www.filedropper.com/tylolandermk1f

[Dkmdlb]

It's typical for the maneuver node marker to move significantly during the last few m/s of a burn. It sounds like that is what's happening. Is Mechjeb trying to follow the drifting blue marker on the navball? If yes, it's normal. If you don't like it, just cancel the maneuver.

Same with rendezvous autopilot. Half a meter per second shouldn't make that much difference. Just cancel the manuever if you don't like it. Mechjeb isn't very efficient when it comes to rendezvous anyway.

Also landing autopilot isn't that great at precision landing either.

[linkxsc]

Do you mean it was 10km from where you wanted to land. Or like the landing control stopped working at 10km and just dropped you.

Also the "angle that doesnt help anymore" mechjeb isnt trying to correct for plane is it?

[Tokay Gris]

It's typical for the maneuver node marker to move significantly during the last few m/s of a burn. It sounds like that is what's happening. Is Mechjeb trying to follow the drifting blue marker on the navball? If yes, it's normal. If you don't like it, just cancel the maneuver.

Thats what I did. Drifting of the blue marker is normal, but usually not in a perpendicular direction that does no longer reduce the nodes delta-V.

Same with rendezvous autopilot. Half a meter per second shouldn't make that much difference. Just cancel the manuever if you don't like it. Mechjeb isn't very efficient when it comes to rendezvous anyway.

So far, it worked pretty good with some exceptions. But never before was it THAT far off to the point where it didn't make rendezvous.

Also landing autopilot isn't that great at precision landing either.

Ususally for me it is about 10 metres off max.

To the point where I pay careful attention. More than once one craft landed on top of the other.

[Tokay Gris]

Do you mean it was 10km from where you wanted to land. Or like the landing control stopped working at 10km and just dropped you.

MechJeb landed just fine, just 10 km from the place where I wanted it to land. Which doesn't help if you want to make rendezvous on the ground with a kethane rig. Expecially if both don't have wheels.

Also the "angle that doesnt help anymore" mechjeb isnt trying to correct for plane is it?

MechJeb does this even on closing in on the rendezvous and just burns perpendicular. Kind of in a circle around the target.

Like this: target is in the east, MechJeb burns north.

[Claw]

Ascent worked up until only about 0.5 m/s left on the maneuver node, which then shifts to an angle that does not help anymore but keeps burning anyway (MechJeb that is).

I have been experiencing this more so in 0.23.5 than I did in 0.23.0. I'm not sure if they are tweaking the MJ code and it's a bit off, or if I just didn't run into it that much in 0.23.0. Although I've noticed the maneuver node steering also seems to run off a bit earlier than it 0.23.5.

In either case, I do like others have hinted at and cancel the autopilot when it's getting toward the end. That way I can keep the heading where it's at and complete the burn myself. It's not ideal if you want MJ to do it all, but it works okay for those times when the steering races away.

I'm sorry, I can't comment on the landing because I never use it. I know MJ has problems on atmospheric bodies, but that doesn't really help...

[Tokay Gris]

Just did a test on Mun.

Same problems there.

Where other vessels pinpointed their landings (10 m off target at most) and made rendezvous countless times, this vessel lands more than 10 km off target and rendezvous using MechJeb is impossible.

I haven't tried docking autopilot yet, though.

[Sirrobert]

While I almost never have problems with drifting node markers, I know the problem.

Mechjeb will always try to follow the marker, and the problem is that, if it drifts like that at the very last stages of the burn, Mechjeb will follow it. But it will keep burning the engines while chacing the marker. THAT is what's causing the problems.

The docking autopilot is extremely inefficient, if you know how to dock, don't use it

[Tokay Gris]

While I almost never have problems with drifting node markers, I know the problem.

Mechjeb will always try to follow the marker, and the problem is that, if it drifts like that at the very last stages of the burn, Mechjeb will follow it. But it will keep burning the engines while chacing the marker. THAT is what's causing the problems.

I know it does. But that is not the problem here. I am not even sure the problem is actually MechJeb. Because MechJeb just follows the marker.

The marker is the problem. But why it shifts in a direction that is actually useless, I don't understand.

The docking autopilot is extremely inefficient, if you know how to dock, don't use it

Not for me, it isn't. I can do it myself, but since I am using quite big ships with many parts right now, docking by hand and in the slideshow this produces is ... well, lets say: Boring.

[Tokay Gris]

I'm sorry, I can't comment on the landing because I never use it. I know MJ has problems on atmospheric bodies, but that doesn't really help...

Just tested another lander bound for Duna. With parachutes.

Landed with 6 metres off target.

So my guess is there is some basic problem with my design. But not a problem with MechJeb itself. A problem with the nodes.

I have so far no idea whatsoever why this particular design fails so much.

I also noticed that with this design, the landing predictions (made by MechJeb) wander off considerably during descent.

The deorbiting burn stops within 300 meters of predicted target, then it changes during descent.

Which might make sense on a body with atmosphere, but not on one without.

And like I just tested (but with another design), it works like a charm on Duna.

[Sirrobert]

I know it does. But that is not the problem here. I am not even sure the problem is actually MechJeb. Because MechJeb just follows the marker.

The marker is the problem. But why it shifts in a direction that is actually useless, I don't understand.

I see

Well I can try to awnser that one to.

What the marker actually tries to do, is point you in the direction you need to burn to get the projected orbit. So if you overshoot the manouvernode, it will autoadjust to make sure you still get the exact projected orbit if you follow the marker, and the exact amount of deltaV given.

So Mechjeb is probably a tiny bit off center with the marker, and thus the marker adjusts to make sure you still get the exact projected orbit.

Likewise if you burn to far, the marker will actually flip around and tell you to burn in the opposite direction

[Tokay Gris]

What the marker actually tries to do, is point you in the direction you need to burn to get the projected orbit. So if you overshoot the manouvernode, it will autoadjust to make sure you still get the exact projected orbit if you follow the marker, and the exact amount of deltaV given.

I know. But this is not what happens.

What happens is that the marker goes in a direction that is not helping anything at all.

Like this: I want to match velocities. MechJeb can make a node for this. Either MechJebs autopilot or I by hand execute that marker and usually this does the trick. Usually and with all my other vessels this works seamlessly.

Not so with this vessel. My guess is that the marker is actually correct but in the process of closing in some error creeps in. I looks like the mechanics of the game assume or detect some direction of movement with this vessel that is not exactly in the direction of the thrust applied. Which then, during the last few m/s totally throws off the marker. It wants to compensate (like you said) for something that is not there.

Which the looks like the thrust applied (in the direction of the blue marker) is not pro- or retrograde in relation to the target or the orbit, but perpendicular to this.

And any thrust applied perpendicular to the movement vector is not changing the movement vector.

(Or is my english at miss here? I mean to say "rechtwinklig" or "at a right angle". That is "perpendicular", isn't it?)

So Mechjeb is probably a tiny bit off center with the marker, and thus the marker adjusts to make sure you still get the exact projected orbit.

Likewise if you burn to far, the marker will actually flip around and tell you to burn in the opposite direction

The marker is the problem, as far as can I see. Not MechJeb, which just executes the node. I think.

(Which I really like. Executing nodes by hand is crude and usually - as you say - a bit off.)

[undercoveryankee]

I know. But this is not what happens.

What happens is that the marker goes in a direction that is not helping anything at all.

Like this: I want to match velocities. MechJeb can make a node for this. Either MechJebs autopilot or I by hand execute that marker and usually this does the trick. Usually and with all my other vessels this works seamlessly.

Not so with this vessel. My guess is that the marker is actually correct but in the process of closing in some error creeps in. I looks like the mechanics of the game assume or detect some direction of movement with this vessel that is not exactly in the direction of the thrust applied. Which then, during the last few m/s totally throws off the marker. It wants to compensate (like you said) for something that is not there.

Which the looks like the thrust applied (in the direction of the blue marker) is not pro- or retrograde in relation to the target or the orbit, but perpendicular to this.

And any thrust applied perpendicular to the movement vector is not changing the movement vector.

(Or is my english at miss here? I mean to say "rechtwinklig" or "at a right angle". That is "perpendicular", isn't it?)

The marker is the problem, as far as can I see. Not MechJeb, which just executes the node. I think.

(Which I really like. Executing nodes by hand is crude and usually - as you say - a bit off.)

What the node marker shows you is the difference between your current velocity and your original target velocity. If the burn is a few hundred meters per second, then a fraction of a degree of pointing error can leave you with a difference of several meters per second at right angles to the original burn.

Another source of error in executing a node is if your center of mass is off-axis. MJ has to apply some pitch or yaw input to keep you on the target orientation, and your engines gimbal and accelerate you to the side of where you're pointing.

If you're executing a node by hand and you see the marker go flying out to the side, you cut your engines and either decide that you're close enough or line up carefully with the marker before you burn again.

Mechjeb tries to keep the engines on at minimum thrust through much larger orientation corrections than I would if I were hand-flying. If you have a high thrust-to-weight ratio and a vehicle that's slow to turn, you can end up basically "orbiting" your target velocity.

Solutions: Decrease TWR and/or increase pitch and yaw authority. Put your "control from" part near the engines and the center of mass to reduce flex so the calculated velocity is stable. Consider locking gimbal on engines if you have enough other sources of control authority. Be prepared to abort execution of a node if you're no longer getting improvement.

[Tokay Gris]

What the node marker shows you is the difference between your current velocity and your original target velocity. If the burn is a few hundred meters per second, then a fraction of a degree of pointing error can leave you with a difference of several meters per second at right angles to the original burn.

This I know. But thats BEFORE I start the maneuver.

During execution it seems, the error creeps in. As if the engine was off angle. While it is not.

Another source of error in executing a node is if your center of mass is off-axis. MJ has to apply some pitch or yaw input to keep you on the target orientation, and your engines gimbal and accelerate you to the side of where you're pointing.

This vessel is pretty symmetric.

Mechjeb tries to keep the engines on at minimum thrust through much larger orientation corrections than I would if I were hand-flying. If you have a high thrust-to-weight ratio and a vehicle that's slow to turn, you can end up basically "orbiting" your target velocity.

I need the pretty high TWR to get off Tylo in the first place. Also, I have done this method with TWRs up to 30.

And the torque on this vessel is no problem. Plenty of it.

I am right now testing another design and a slight modification on the current design.

And again: Other designs so far worked without a hitch. MechJeb is not "it" as far as I can discern.

[Claw]

Hmm, something else that I just thought of. I don't know what the root part of your vessel is, but I'm wondering if the control reference is being transferred to MechJeb's position.

I had a symmetric 4x probe rocket that kept using one of the probe cores as the control reference and MJ had a heck of a time controlling it. MJ was always slightly off with it's pointing which got worse when near the end of a burn. I never landed with it, but when I took off the thing was a mess and I had to provide lots of input. Eventually I had to go back and put another probe core along the center of the ship in order for MJ to not be confused. I would guess the lander can is the root of your ship, but maybe something weird happens with the control reference when docking/undocking.

I'm not sure if "control from here" will fix that problem, but maybe give that a try. I have my doubts given how symmetric your craft looks, but might be worth a shot to maybe put the "control from here" on the docking port on top.

[Tokay Gris]

Hmm, something else that I just thought of. I don't know what the root part of your vessel is, but I'm wondering if the control reference is being transferred to MechJeb's position.

That I thought of. It isn't.

Trying right now a different design... I do have a suspicion.

Can't really believe that this specific part is the problem. But am testing anyway.

[Tokay Gris]

I just don't get it.

I am testing other designs right now and for some reason unknown to me, they are off by a very long shot.

I tested designs that worked before without a hitch, MechJeb did land them within 10 meters of intended target.

But it doesn't anymore.

Now it is off by 5 to 10 KILOMETERS.

And I just don't know why.

There is one thing I still have to test. Then I would say I nailed the problem. But so far I do not understand what is happening here.

[Tokay Gris]

Did nail the problem.

I don't really know why it is the way it is, but the solution is simple:

Don't use Hyperedit.

I tested landers that did work before, filled them up with hyper edit and they landed WAY off. They did land, however. Using MechJeb.

If I did not fill them up with hyper edit, they worked dandy. Just landed an almost empty fuel tanker that still had roughly 35 tons of mass on Mun. Difference from target: 24 meters.

So I assume that the problem is somewhere in the predictions that would have to compensate for the changed mass but don't. And somehow does not compensate for the sudden translocation to another body.

The good thing is that with this I am save. I just used hyperedit in the testing phase, not in the "working" save.

In the "working" save, I did everything the classical way. Bring vessels to orbit, bring tankers to orbit, refuel, the works.

So this particular problem - my guess is the rendezvous problem is the same - should not come up in the "working" save.

The bad thing is that testing is not really possible. I actually have to bring that lander to Jool and land. And then rendezvous.

Well, well.

[magnemoe]

Mechjeb has problems landing on Tylo, strange as the 0.18-0.21 versions did not but the newer versions miss target with hundreds of kilometers and often crash during final decent as it prefer to get sideway speed down to 0.01 m/s.

Yes it might be my landers, I learned in the early versions that an aspargus style lander was bad because the reduced TWR caused crashes, this might have improved with current version I can brake with the transfer stage, drop it and land successfully so I only drop tanks during landing.

[Tokay Gris]

Yes it might be my landers, I learned in the early versions that an aspargus style lander was bad because the reduced TWR caused crashes,

My experience is that asparagus landers cause problems because the changing mass and TWR cause a hell of a lot of trouble.

It kind of works if the TWR of the stages gets higher. Which means, if you through off empty tanks without engines it works.

[Claw]

Did nail the problem.

I don't really know why it is the way it is, but the solution is simple:

Don't use Hyperedit.

Glad you found it. It might be worth posting to the MechJeb guys and maybe they'll fix it up.

[Tokay Gris]

Glad you found it. It might be worth posting to the MechJeb guys and maybe they'll fix it up.

Thanks. I am not sure it is MechJeb. MechJeb calculates the nodes correctly (I guess), but I think it has something to do with the changed weight and therefore the nodes are wrong.

But I will mail them.

[Claw]

Thanks. I am not sure it is MechJeb. MechJeb calculates the nodes correctly (I guess), but I think it has something to do with the changed weight and therefore the nodes are wrong.

But I will mail them.

It might be that MJ is calculating wrong if it has old TWR and weights stuck in there somewhere. I'd say KSP might also be to blame, but KSP simply puts the nodes where you tell it to.