Latest Asobo update has radically changed the a/p

[robert young 2,897]

It could be either my imagination or something peculiar to my MSFS installation, but I don't think so. Every aircraft I've tried so far, default or modded, starts coming out of a heading-select turn way, way too early. Example: You are flying due North and you select due East, and switch on altitude hold then heading. Before the update, most aircraft banked at 20-30 degrees then would roll out of the turn approximately 5-10 degrees before the selected heading, then settle on the heading within around 5-8 seconds max.

After the update every aircraft I've tried starts rolling fine, but then begins rolling to level far too early, resulting in the last 10 degrees capture of the heading taking ages. Now if I am wrong on this I apologise, but anyone interested in this might like to test it for themselves. The other odd thing is that so far I cannot over-ride these settings at all, whatever I tweak. This has implications for ILS capture too, as the roll for heading correcting is so slow now that I keep missing the localiser unless I capture the LOC from a very shallow angle.

[jpe828 624]

Robert, your observations are correct. The "smoothing" of the AP they have done has rendered it rather unrealistic. If the airplane is following a GPS course it seems to have very slow, but long course corrections too. And certainly, on the approach it impacts the Localizer capture and following the LOC. RNAV too. 

[robert young 2,897]

The default DA62 is taking more than 30 seconds to capture and arrive at an autopilot heading from only ten degrees to go. Same with the default C172 G1000. Same with the Caravan. They are all incredibly sluggish to capture a heading or ILS localiser.

[jpe828 624]

Yep, seems like they overcompensated. And that is a bummer it is not easily tweaked yet. I know other on the main forum are not happy about this part of the update. 

[Claudius_ 326]

Confirmed, I tried with the XCub A/P and the reaction is very slow when turning the heading.

[robert young 2,897]

28 minutes ago, jpe828 said:

Robert, your observations are correct. The "smoothing" of the AP they have done has rendered it rather unrealistic. If the airplane is following a GPS course it seems to have very slow, but long course corrections too. And certainly, on the approach it impacts the Localizer capture and following the LOC. RNAV too. 

I wouldn't call it "smoothing" - if that is the term they used. I would use a rather more rude description. This is not just a slight error in judgement. It is an unbelievable example of complete misunderstanding of how autopilots should work. The odd thing is that in any case, the ai.cfg and systems.cfg files allowed you to tweak whatever heading/capture settings you wished. Now it looks like this "smoothing" fiasco has actually disabled ANY meaningful tweaking at all.

I'll take a stress pill and try to calm down before further comment! Except to say the DA62 mod is now completely scuppered and will not appear until or unless this major example of "progress" can be overcome.

[marsman2020 1,310]

They need to approach these problems as if they are simulating something real or engaging in engineering and not as game designers.  I'm still not seeing a lot of evidence of that though.  Go talk to Garmin about how an AP works, it is Garmin's name on the panel of all these aircraft. 

Heck they could just go watch P. Gatcomb's IFR videos on YouTube which showed the prior working version of the sim....working.  He was doing pattern turns and holds with HDG mode.  Totally impossible now with this new AP behavior. 

Edited November 26, 2020 by marsman2020

[Donstim 109]

Did you watch the Developer Q&A?  Seb characterized the autopilot makeover as a major improvement. He described a scenario where people would try to activate the autopilot with a high climb rate and set a heading change of 180 degrees and autopilot would freak out. He says it now handles this smoothly, but may need some "tweaking" for localizer capture on some airplanes.

I have to kind of laugh at this because of all the updates, this one impacts me (for the worse) the most, and I had expected this be a fairly minor update in terms of airplane changes.  I thought it was just supposed to be a North America scenery update with some CPU optimization. I am surprised they did not find it to be unsatisfactory in their own testing, but surely they must be aware of it by now. i just hope we don't have to wait another month for it to be fixed. 

[robert young 2,897]

Asobo need some proper pilots and people who know what they are doing. Every update like this goes completely over the top. Smoothing means making something less abrupt. This ain't smoothing. It's strangling!

[PuffeB 27]

Yes - this change of the AP behaviour is extremely annoying. 

[marsman2020 1,310]

It doesn't need 'smoothing', what it needs is logic based on how an actual A/P would work...

-Increase control deflection to achieve a specified max roll rate (can put some constraints on this part to make it 'smooth' yet prompt)

-Roll to a specified max bank angle

-Stop roll (can put some constraints on this part to make it 'smooth' yet prompt)

-Watch heading rate of change

-At desired heading minus some number, repeat the process to roll back to level

They implemented PID loops for all the A/P axes, but an aircraft isn't a thermal chamber controller, and a PID loop is not how to do roll control.  It needs to stay in the roll until the end and then roll out....not roll out slowly based on proportional error. PID is fundamentally the wrong algorithm for a roll autopilot.....  Maybe OK for pitch.

Edited November 26, 2020 by marsman2020

[Glenn Fitzpatrick 3,798]

Why do I get the impression this was done to help out incompetent pilots ?

I am probably biased as I spent a lot of time back in the IL2 days 20 years ago arguing with people claiming the Spitfire should far be easier to fly as "real pilots in the Battle of Britain only had 8 hours training" and hence any competent gamer should be able to handle one after a few hours practice  (hint, the truth seems to be some RAF pilots had as little as 8 hours CONVERSION training and hence 8 hours on TYPE before being sent to a combat squadron. Presumably they would also have had 500 or more hours in everything from Gloster Gladiators to Furys and Defiants or even Hurricanes before being allowed anywhere near a Spitfire) usually also quoting pilots from the time saying they were "easy and a pleasure to fly" failing to note the comments were part of a comparison to something else that tried to kill you every time your attention wandered.

 

 

Edited November 26, 2020 by Glenn Fitzpatrick

[Jetman67 517]

Such a pity this has been changed for the worse,let's hope this is fixed asap

[odourboy 1,289]

Thanks for brining this up. I'd wondered what happened.

[robert young 2,897]

38 minutes ago, marsman2020 said:

It doesn't need 'smoothing', what it needs is logic based on how an actual A/P would work...

-Increase control deflection to achieve a specified max roll rate (can put some constraints on this part to make it 'smooth' yet prompt)

-Roll to a specified max bank angle

-Stop roll (can put some constraints on this part to make it 'smooth' yet prompt)

-Watch heading rate of change

-At desired heading minus some number, repeat the process to roll back to level

They implemented PID loops for all the A/P axes, but an aircraft isn't a thermal chamber controller, and a PID loop is not how to do roll control.  It needs to stay in the roll until the end and then roll out....not roll out slowly based on proportional error. PID is fundamentally the wrong algorithm for a roll autopilot.....  Maybe OK for pitch.

Ideally yes. I hate to compare with FSX as people get very annoyed, but for all its faults FSX had a decent autopilot, and if it didn't work it was probably due to a faulty flight model. PIDs are not the ideal, but they can be made to work. In two decades I never had the slightest problem with the FSX autopilot. It was very tuneable, and completely reliable for ILS, Nav and VS/heading. The only time it went wrong was actually when some devs tried to get too clever and assign all a/p functions to custom coding, which some did successfully but others made it worse than just sticking to the default variables.

As you say, capturing a heading should be very easy if people would stop tinkering (badly) with it. Command a roll of 20-30 degrees. Ten degrees before capture, gently roll back to wings level. The integral and derivative PID params are there for slight corrections. With Asobo's version you have to fight each variable just to get a simple banked turn. They really need to learn to keep things simple, but every time they update they make something worse than it was before. ARGH!!!!