Confused by auto pilot ATT

But you said your self the course is drifting, didn't you? From OP:

 

In that respect ATT and HÖJD should act the same. They don't drift from course. My comment was about that part of your OP, not the returning to course (as I said I didn't get working either).

 

Ah OK, I see what you mean. Sorry!

Yes, I'm not seeing the drifting I was talking about any more. I wonder if it was in a specific mission or set of circumstances.

Anyway, yes, you're right, that doesn't seem to be an issue.

Ah OK, I see what you mean. Sorry!

 

Yes, I'm not seeing the drifting I was talking about any more. I wonder if it was in a specific mission or set of circumstances.

 

Anyway, yes, you're right, that doesn't seem to be an issue.

I'm running a test for the return-to-course issue. Surprisingly enough it does seem to work! ...sort of...

 

Mind you, it is HIGLY sensitive on controller input and INCREDIBLY slow!

 

I attached the (very simple) mission I used, a start-in-air flight across the widest part of the Black Sea with x-tank and no wind, start with active pause. This is how I went about:

I opened DXTweak (controller management software) on my second monitor and made sure my stick and pedals was thoroughly centered (with an input value close to 32768 out of 65536).

Turned off my TrackIR.

Loaded the mission (in active pause) and activated HÖJD and noted indicated speed (about 850km/h).

Deactivated active pause and got a pitch jump, it did return to set pitch and altitude within a few seconds though. Heading was not affected. the steering dot stayed inside the FPV circle (not perfectly centered though, but steady).

 

Made sure to stabilize the speed at the original noted above and noted throttle input value in DXTweak for reference if repeating the test later. Kept it flying this way to make sure everything was stabilized. Adjusted seat height and zoom to bring the HUD to fill my main screen to really see even tiny changes.

 

Yanked the stick carefully to one side, the plane initiated a turn away from course, then returned the stick to center. All with an eye on DXTweak. The displacement was about 250-300 of the input value (less than 0.5% of 65536 or 1% of 32768 ). If the return-to-course would not work, the plane would settle at the new course it had at the moment I returned the stick to center, but it didn't, it returned slowly to previously established course towards the WP! I repeated this test several times with stick deflections of the same size but in both directions with the same result, slow return to course after recenter. It didn't return to EXACTLY the same course, but very close. With the history of Heatblure being very thorough about system modeling, I would not be surprised if they added some sort of small system errors causing this.

 

Next thing I tested was to keep the stick at said 250-300 deflection. At first the return-to-course didn't seem to work, but I let it go on. S..l..o..w..l..y.. the turn rate decreased, almost settled on a new course, then S..l..o..w..l..y..started to return to set course!

After a loooong time the course settled almost towards the WP. It was more off in relation to the WP than the previous tests, but I guess the return-to-cource isn't actually returning to the exact same track that should have been if no stick deflection, but rather just return to the same direction. In this case the return-to-cource was so slow that the plane shifted its position sideways quite much in relation to said track that it ended up quite far away but parallel to it.

 

My conclusion is that return-to-cource actually do work, but is WHAY to slow to make it practically usable. Remember that the deflections I used in this test was VERY small and performed very carefully, barely touched the stick. Imagine what bigger deflections would do.

 

The altitude hold on the other hand is fast enough to be usable. I get the impression that the issue with slow return-to-cource might be just a wrong order of magnitude of the scaling constant for the corrective signal (Δ-course) or the steering input for the control surfaces.

Course hold test No wind.miz

Edited January 13, 2019 by Holton181

:worthy:

 

Btw, how slow is slow, minutes?

:worthy:

 

 

 

Btw, how slow is slow, minutes?

Actually didn't time it, should have done when you mention it. For the small yanks where I returned to center (after 1-2s deflected, relative small change in course) about 10-15s, the ones were I kept the stick deflected maybe 10-15min, maybe more.

Actually that sounds quite reasonable. I mean if you keep the stick deflected you are basically fighting the autopilot. Also, 10-15 seconds is not so bad, isn't it?

I'm running a test for the return-to-course issue. Surprisingly enough it does seem to work! ...sort of...

 

Mind you, it is HIGLY sensitive on controller input and INCREDIBLY slow!

 

I attached the (very simple) mission I used, a start-in-air flight across the widest part of the Black Sea with x-tank and no wind, start with active pause. This is how I went about:

I opened DXTweak (controller management software) on my second monitor and made sure my stick and pedals was thoroughly centered (with an input value close to 32768 out of 65536).

Turned off my TrackIR.

Loaded the mission (in active pause) and activated HÖJD and noted indicated speed (about 850km/h).

Deactivated active pause and got a pitch jump, it did return to set pitch and altitude within a few seconds though. Heading was not affected. the steering dot stayed inside the FPV circle (not perfectly centered though, but steady).

 

Made sure to stabilize the speed at the original noted above and noted throttle input value in DXTweak for reference if repeating the test later. Kept it flying this way to make sure everything was stabilized. Adjusted seat height and zoom to bring the HUD to fill my main screen to really see even tiny changes.

 

Yanked the stick carefully to one side, the plane initiated a turn away from course, then returned the stick to center. All with an eye on DXTweak. The displacement was about 250-300 of the input value (less than 0.5% of 65536 or 1% of 32768 ). If the return-to-course would not work, the plane would settle at the new course it had at the moment I returned the stick to center, but it didn't, it returned slowly to previously established course towards the WP! I repeated this test several times with stick deflections of the same size but in both directions with the same result, slow return to course after recenter. It didn't return to EXACTLY the same course, but very close. With the history of Heatblure being very thorough about system modeling, I would not be surprised if they added some sort of small system errors causing this.

 

Next thing I tested was to keep the stick at said 250-300 deflection. At first the return-to-course didn't seem to work, but I let it go on. S..l..o..w..l..y.. the turn rate decreased, almost settled on a new course, then S..l..o..w..l..y..started to return to set course!

After a loooong time the course settled almost towards the WP. It was more off in relation to the WP than the previous tests, but I guess the return-to-cource isn't actually returning to the exact same track that should have been if no stick deflection, but rather just return to the same direction. In this case the return-to-cource was so slow that the plane shifted its position sideways quite much in relation to said track that it ended up quite far away but parallel to it.

 

My conclusion is that return-to-cource actually do work, but is WHAY to slow to make it practically usable. Remember that the deflections I used in this test was VERY small and performed very carefully, barely touched the stick. Imagine what bigger deflections would do.

 

The altitude hold on the other hand is fast enough to be usable. I get the impression that the issue with slow return-to-cource might be just a wrong order of magnitude of the scaling constant for the corrective signal (Δ-course) or the steering input for the control surfaces.

Interesting, but it just doesn't work for me.

I've tried setting ATT on your mission, make <7 degree slight (few degrees) course change, stick recentre and nothing happens. New course sticks.

I have also tried making larger course changes but of course, that doesn't work either.

I've waited up to 10 minutes, but nothing.

Actually that sounds quite reasonable. I mean if you keep the stick deflected you are basically fighting the autopilot. Also, 10-15 seconds is not so bad, isn't it?

No, you're not fighting the autopilot in roll terms. Both ATT and HOJD hold your roll if its between 7 and 66 degrees. So, ATT holds roll and pitch, whereas HOJD just holds roll as it is holding your altitude.

Actually that sounds quite reasonable. I mean if you keep the stick deflected you are basically fighting the autopilot. Also, 10-15 seconds is not so bad, isn't it?

Due to how small input deflections I did, it could almost represent a changed side wind and in my opinion the AP should be able to adjust for that within seconds, like the altitude hold. The size of the deflections where also small enough to represent 4-5LSB change on a 10bit controller or 2LSB on a 9bit controller (my Saitek Evo). Really not much and in the order of noisy potentiometers.

To even make the pitch start to change while in HÖJD I had to deflect the stick about 3300 compared to immediately bank when deflecting it sideways, and in the above tests it was side deflections of about only 300, an order of ten lower than making the pitch change.

I just tested to deflect the stick both in roll and pitch in the order of 5 degrees (rough estimate) while in HÖJD and kept it there. It levelled off pretty quickly in both axis and returned to set altitude, but the bank took several minutes to really stop the turn and I can't yet see any tendencies towards restoring the course. Still running and waiting.

I let the above test run for quite some time, but it never returned. It was just a "quick" test with arbitrary selected stick deflections (2200 in pitch, 900 in roll). The cat yanked the stick before I got the time to re-center the stick (I wasn't sitting at the controls at the time).

Conclusion: amount of stick deflection do seem to have an effect on the ability to return on course even if the plane has settled. For it to return the stick has to be deflected very lightly or completely centered.

Secondary conclusion: Never bring your cat to the lab.

I rerun this latest test (without the long wait), and as soon the plane had settled on the new course I re-centered the stick. Then it returned to set course. Good.

 

I have a theory:

The implementation of the scaling constant for the corrective signal (Δ-course) or the steering input for the control surfaces is related to the angular deflection from set course for the bank, and deflection from set altitude for the pitch. This we know.

But to RETURN to said course or altitude one needs to make a correction that is double the deflection (and opposite). If the correction is the same size as the deflection we will only settle the attitude.

For the altitude hold this seems to be correct, for a stick deflection lower than a particular value. If above this value the AP can't even halt the angular pitch change. But for the course it seems the correction is the same as the deflection, explaining why it can level the plane in bank and stop the turn, and also why this is a relatively slow process, but it can't make it return since the deflection and correction is of the same size but opposite so no resultant signal to stear it back on course.

The slow process might be a way to limit oscillations, and would not be experienced as slow if the correction signal was correctly double the deflection.

 

As a side note, the manual for the real aircraft only mention "the control surfaces" when talking about returning on course, while the Heatblur manual only mention the "rudder servo" for the same task.

 

Edit:

I realized I used "stick deflection" when I actually only meant either angular deflection from set course or deflection from set altitude. Corrected, but there is still one "stick deflection"stick present and is meant that way.

Edited January 13, 2019 by Holton181

I rerun this latest test (without the long wait), and as soon the plane had settled on the new course I re-centered the stick. Then it returned to set course. Good.

 

 

I have a theory:

The implementation of the scaling constant for the corrective signal (Δ-course) or the steering input for the control surfaces is related to the angular deflection from set course for the bank, and deflection from set altitude for the pitch. This we know.

But to RETURN to said course or altitude one needs to make a correction that is double the stick deflection (and opposite). If the correction is the same size as the stick deflection we will only settle the attitude.

For the altitude hold this seems to be correct, for a stick deflection lower than a particular value. If above this value the AP can't even halt the angular pitch change. But for the course it seems the correction is the same as the stick deflection, explaining why it can level the plane in bank and stop the turn, and also why this is a relatively slow process, but it can't make it return since the stick deflection and correction is of the same size but opposite so no resultant signal to stear it back on course.

The slow process might be a way to limit oscillations, and would not be experienced as slow if the correction signal was correctly double the stick deflection.

 

 

As a side note, the manual for the real aircraft only mention "the control surfaces" when talking about returning on course, while the Heatblur manual only mention the "rudder servo" for the same task.

I don't fully follow what you are saying, but all I can say is that no return to course happens for me whether tiny deflection, small defection, large deflection, re-centring, auto centring. etc. Nothing brings it back on course. No matter how long I wait!

I don't fully follow what you are saying, but all I can say is that no return to course happens for me whether tiny deflection, small defection, large deflection, re-centring, auto centring. etc. Nothing brings it back on course. No matter how long I wait!

 

I see you missed my edit. But it doesn't change the issue you are experiencing. I might figure out a better way to explain, but have to take my little kid to bed right now.

I don't fully follow what you are saying, but all I can say is that no return to course happens for me whether tiny deflection, small defection, large deflection, re-centring, auto centring. etc. Nothing brings it back on course. No matter how long I wait!

I'm back.

As I said you missed my edit I did at the same time you posted the above.

The reason why you don't seem to be able to make the return-to-course to work whatever you try might be due to your controller. I don't know what you have, but if it doesn't have high resolution and/or have noisy, maybe worn potentiometers and maybe in addition have a centering mechanism that isn't efficient enough to bring it to dead center (and not just to appear to be centered), it would for sure be difficult to get it working. In all my tests all stick deflection has been itsy bitsy tiny even for a high res, noise free controller with a very good centering mechanism. Other than that I have no clue.

 

Regarding my theory I realized now when trying to find a better explanation for you, I was a bit too fast sharing it. It is only partly explaining what is happening. The rest I haven't figured out yet. I remember from the University that the subject of "automatic control engineering" (what this thread is all about) was considered among the most difficult subjects to wrap your head around. So don't bother to much about my theory for now. Let's just stop at the conclusion it doesn't seem to work as intended.

But I would like to add a thought I forgot to share previously. In my first rest regarding the return-to-course functionality I said it really did return even if not centering the stick. It really did. Then in the second test, where I deflected the stick a bit more and both in pitch and bank, it returned to altitude but not to course. Now why this inconsistency? I really have no clue!

I really wish @RagnarDA, @Cobra847 or anyone else from the team could chime in on this already!

I'm back.

As I said you missed my edit I did at the same time you posted the above.

The reason why you don't seem to be able to make the return-to-course to work whatever you try might be due to your controller. I don't know what you have, but if it doesn't have high resolution and/or have noisy, maybe worn potentiometers and maybe in addition have a centering mechanism that isn't efficient enough to bring it to dead center (and not just to appear to be centered), it would for sure be difficult to get it working. In all my tests all stick deflection has been itsy bitsy tiny even for a high res, noise free controller with a very good centering mechanism. Other than that I have no clue.

 

Regarding my theory I realized now when trying to find a better explanation for you, I was a bit too fast sharing it. It is only partly explaining what is happening. The rest I haven't figured out yet. I remember from the University that the subject of "automatic control engineering" (what this thread is all about) was considered among the most difficult subjects to wrap your head around. So don't bother to much about my theory for now. Let's just stop at the conclusion it doesn't seem to work as intended.

But I would like to add a thought I forgot to share previously. In my first rest regarding the return-to-course functionality I said it really did return even if not centering the stick. It really did. Then in the second test, where I deflected the stick a bit more and both in pitch and bank, it returned to altitude but not to course. Now why this inconsistency? I really have no clue!

Thanks again for that.

As you can see from my signature, I have a TM HOTAS Warthog, and it works perfectly. There is no problem with my device in centring terms or any other. In any case, with movement less than 7 degrees, the device self levels in the Viggen.

If we put the science to one side, the fact is that this is not working at all for me, and I'm struggling to see how you are getting this drift back on course over a long period of time.

As I said before, I have tried everything to get to work with ATT and it just doesn't. Small (<7 degree which when stick is released auto-level) drifts don't revert, nor do any trim drifts, larger movement, etc. Nothing. I have made 1-2 degree drifts and still no revert.

I am mystified by how you are achieving this. Can you post a video?

Thanks again for that.

As you can see from my signature, I have a TM HOTAS Warthog, and it works perfectly. There is no problem with my device in centring terms or any other. In any case, with movement less than 7 degrees, the device self levels in the Viggen.

If we put the science to one side, the fact is that this is not working at all for me, and I'm struggling to see how you are getting this drift back on course over a long period of time.

As I said before, I have tried everything to get to work with ATT and it just doesn't. Small (<7 degree which when stick is released auto-level) drifts don't revert, nor do any trim drifts, larger movement, etc. Nothing. I have made 1-2 degree drifts and still no revert.

I am mystified by how you are achieving this. Can you post a video?

 

Thanks again for that.

As you can see from my signature, I have a TM HOTAS Warthog, and it works perfectly. There is no problem with my device in centring terms or any other. In any case, with movement less than 7 degrees, the device self levels in the Viggen.

If we put the science to one side, the fact is that this is not working at all for me, and I'm struggling to see how you are getting this drift back on course over a long period of time.

As I said before, I have tried everything to get to work with ATT and it just doesn't. Small (<7 degree which when stick is released auto-level) drifts don't revert, nor do any trim drifts, larger movement, etc. Nothing. I have made 1-2 degree drifts and still no revert.

I am mystified by how you are achieving this. Can you post a video?

 

I'm on tapatalk, don't see any signatures. Sorry for that.

Googled a bit for TM HOTAS Warthog and it seems to have excellent 16bit resolution. So not the problem. It has Hall sensors, so noise and worn sensors goes out the window. Can't tell about centering, but I trust you it's not the issue. The only thing I can think of is that I don't look at bank angles per se, but on input values from the stick (DXTweak). I make sure they are as small as possible (300 out of 65536) after I first stabilize. Do you have an extension? As I said earlier I have a basically full length helicopter cyclic (without centering spring but with friction to hold it in place), much easier for tiny stick deflections, and I basically only tapp lightly on it to get the 300 input, can't realy see it on the stick or on the controller helper (RCtrl+Enter). If you don't already have it, or something similar, I recommend you search for a copy of DXTweak try it out.

I have never made a gaming video and have no such software for it, sorry.

I'm on tapatalk, don't see any signatures. Sorry for that.

Googled a bit for TM HOTAS Warthog and it seems to have excellent 16bit resolution. So not the problem. It has Hall sensors, so noise and worn sensors goes out the window. Can't tell about centering, but I trust you it's not the issue. The only thing I can think of is that I don't look at bank angles per se, but on input values from the stick (DXTweak). I make sure they are as small as possible (300 out of 65536) after I first stabilize. Do you have an extension? As I said earlier I have a basically full length helicopter cyclic (without centering spring but with friction to hold it in place), much easier for tiny stick deflections, and I basically only tapp lightly on it to get the 300 input, can't realy see it on the stick or on the controller helper (RCtrl+Enter). If you don't already have it, or something similar, I recommend you search for a copy of DXTweak try it out.

I have never made a gaming video and have no such software for it, sorry.

If you are running Windows10 then you have the software. It's just win+g when in game.

EDIT: Just saw your sig. Don't think it works on Win 7.

Correct

I really wish @RagnarDA, @Cobra847 or anyone else from the team could chime in on this already!

+1

+1

 

I've said that here and on other threads where there are issues and we don't get any help.

I appreciate that the guys are busy, but I see replies from Dagnar on other threads in the last few days.

I don't think it's too much to ask really for some help with this and other issues.