AI Flight Model Fidelity

[ericinexile]

I've searched (without luck) all over the DCS website for a "Developer's Note" which discussed the improved flight model fidelity for AI planes and helos. My recollection is that the note/FAQ/whatever which I read years before the DCS release promised a fidelity in DCS AI which would be comparable to Lock-On human controlled aircraft. Does anyone else recall that note?

 

Smokin' Hole

[ericinexile]

Disregard. Found it.

 

https://www.digitalcombatsimulator.com/index.php?end_pos=950&scr=default&lang=en#p3

 

Now, does it apply to AI helos?

 

Disregard, found that too.

Edited May 26, 2009 by ericinexile

[ericinexile]

Your intuition serves you well, kemosabi.

 

Well, when I'm outclimbed by a Hip or chased down by my wingman when I'm cruising at 300kps makes me wonder...

 

Looks like we have three flight models:

 

1) AFM: DCS Black Shark and Lock-On Su25(t)

2) SFM: Other Lock-On human controlled planes and DCS AI fixed wing

3) AI FM: DCS AI helos

Edited May 26, 2009 by ericinexile

[Yo-Yo]

Your intuition serves you well, kemosabi.

 

Well, when I'm outclimbed by a Hip or chased down by my wingman when I'm cruising at 300kps makes me wonder...

 

Looks like we have three flight models:

 

1) AFM: DCS Black Shark and Lock-On Su25(t)

2) SFM: Other Lock-On human controlled planes and DCS AI fixed wing

3) AI FM: DCS AI helos

 

 

Please do not forget that cargo helos in BS fly without any cargo so is it a miracle that Mi-8 with the same powerplant as Ka-50 overclimbs it when it carries all its payloads and full fuel? Mi-26 as far as I remember can do the same because it's EMPTY.

The models of HAI were tuned for main energetic parameters very close to documentation as well as for weight/balance. For example take look at Chinook or AH-64 landing. You can get detailed info about HAI FM on our site.

 

You are quite right about the FM types except AFM. There are two AFMs now - AFM for Su-25/25T and AFM-M :) for Ka-50. AFM-M has thermodynamics model for engine and highly detailed systems models that allows to consider it the second generation. Looking forward I can say that A-10C in its turn has some improvements in comparison with AFM-M.

[amalahama]

Some time ago I built a little test mission with four AI Su-27 and strong turbulence. AI Sukhois looked like they went over railways. Perfect formation and landing. I could see some ugly behaviour when they flew very close to the ground, in a terrain-following path (sharp, unrealistic pitch angles following mountain slopes). It looks like Lockon AI a lot. I don't see any difference.

 

Regards!!

[Yo-Yo]

Some time ago I built a little test mission with four AI Su-27 and strong turbulence. AI Sukhois looked like they went over railways. Perfect formation and landing. I could see some ugly behaviour when they flew very close to the ground, in a terrain-following path (sharp, unrealistic pitch angles following mountain slopes). It looks like Lockon AI a lot. I don't see any difference.

 

Regards!!

 

 

Yes, you are right - SFM traditionnaly uses its own athmosphere. New turbulence affects only flyable Ka-50 and HAI. We plan to replace SFM by downgraded AFM but I really do not know when it will be... :)

 

LockOn AI is burned out from the project... try to force BS fixed wing fly high and fast when it is heavy weighted... :) look at the crosswind landing.

Of course there are some issues in AI behavior we already fixed in the patch.

[ericinexile]

Your intuition serves you well, kemosabi...."

 

Just so I don't look like a schizophrenic talking to himself, that was in reply to a post that was later removed by the poster.

[amalahama]

Yes, you are right - SFM traditionnaly uses its own athmosphere. New turbulence affects only flyable Ka-50 and HAI. We plan to replace SFM by downgraded AFM but I really do not know when it will be...

 

it sounds nice! So many things, so little time...

 

Regards!!

[159th_Viper]

Just so I don't look like a schizophrenic talking to himself, that was in reply to a post that was later removed by the poster.

 

Guilty as Charged :)

 

I only picked up on your edit after my post, and deemed my answer thus superflous and vaporized it.........

 

On the odd occasion when I do delve into the Singleplayer environs I usually end up Scuttling my Wingie/Order RTB........A Habit picked up after also Wondering too many times...

[ericinexile]

Guilty as Charged :)

 

I only picked up on your edit after my post, and deemed my answer thus superflous and vaporized it.........

 

On the odd occasion when I do delve into the Singleplayer environs I usually end up Scuttling my Wingie/Order RTB........A Habit picked up after also Wondering too many times...

 

I'm with you there brother! After abandoning a buddy in the jungle on a Hash Run (no time to explain) I earned the moniker "Judas". Good thing DCS can't court-marshal.

[ericinexile]

I'm sorry ED but I just don't see it. I set up a simple mission with a 16 m/s crosswind and watched a Su25t takeoff and land. Its flightpath was completely unrealistic and seemed quite similar to Lock On. I'm not saying that the TO and Landing were scripted, only that they appeared to occur with no consideration for atmospherics or aerodynamics except for a crab angle on final. In other words, they were nothing like what I would expect an aircraft flying with a Standard Flight Model to look like. Also, aside from flaps and speedbrakes, there was no control surface movement (which admittedly has nothing to do with the flight model but does help make A/I flight "seem" more real.)

 

None of this is important now but it will be as more planes are added to the DCS environment.

Edited May 29, 2009 by ericinexile

[Distiler]

I tested it also and there is no strange movement due turbulences (at max 60x0.1). Only costant yaw opossing wind direction, but the Su-25T is as stable as a rock.

 

PD: ok, somewhat got it with wind at 40 m/s and turbulence at 60x0.1

Edited May 29, 2009 by Distiler

[Yo-Yo]

First of all - the new model of turbulence is for AFM and HAI FM. As SFM historically uses the older model we could not implement the new model into it. The "turbulence" you can experience flying old SFM is not due to the athmospere wind vector stochastic additions but directly stochastic process added to the movement equations FOR FLYABLE aircarft. That's the point.

I really have no idea why we refused to "noise" AI. Maybe it caused problems at early stages of control equations development and was switched off. And nobody remembered to switch it back. I think it could be main the reason that the old way of "turbulence" does not support correlation for the planes in tight formation and it was rejected because of ridiculous behavior of the plane in the formation.

Surely the model is SFM. Try AIs in the same flight with different weights and look at the different AoAs or even at the problems they have holding tight formation, try dogfight for aircraft of the same type but with different weights. There are many signs that it is really SFM :).

But in generally SFM is not the holy cow we want to keep. If we have spare time of course we will try to get rid of it and make something like HAI FM.

[GGTharos]

^^^^

 

And I'll attest to this, having tried these things out before. Yo-Yo isn't fibbing.

[ericinexile]

 

...But in generally SFM is not the holy cow we want to keep...

 

Sacred Cow? Holy Grail? (((insert smile here--doesn't seem to work with Safari)))

 

I believe you, Yo-Yo, regarding improvements in some performance criteria such as wieght vs AOA vs turn rate. Nonetheless, the statement: 'For “Black Shark”, the same Standard Flight Model (SFM) will be used for AI-controlled aircraft that was used for player-controlled aircraft in “Lock On”' gave me the impression that an AI DCS Su27 would be identical in every performance aspect as the LockOn Su27 flown by a human player. That might be the case but it sure doesn't seem that way from my layman's POV. Perhaps my feelings are based on that fact that the AI flys so flawlessly and with such a complete lack of any random movement that it just makes AI planes appear hopelessly unreal.

 

Now I have to go look up "stochastic" (((another smile)))

 

Smokin' Hole

[Yo-Yo]

Sacred Cow? Holy Grail? (((insert smile here--doesn't seem to work with Safari)))

 

I believe you, Yo-Yo, regarding improvements in some performance criteria such as wieght vs AOA vs turn rate. Nonetheless, the statement: 'For “Black Shark”, the same Standard Flight Model (SFM) will be used for AI-controlled aircraft that was used for player-controlled aircraft in “Lock On”' gave me the impression that an AI DCS Su27 would be identical in every performance aspect as the LockOn Su27 flown by a human player. That might be the case but it sure doesn't seem that way from my layman's POV. Perhaps my feelings are based on that fact that the AI flys so flawlessly and with such a complete lack of any random movement that it just makes AI planes appear hopelessly unreal.

 

Now I have to go look up "stochastic" (((another smile)))

 

Smokin' Hole

 

No matter what kind of cow you have if it provides milk. :)

 

Of course the style of flying AI controlling by the pure mathematics differs a bit from the style human flues. It does not perform wobbling stick movement to feel the plane reaction.

But in your suggestion that AI has "THE THIRD MODEL" is very far from the real things. Please think about the price for the new model development and for the modifying the existent.

 

All in all you can believe in SpecialFM for AI or you can not believe - it's your right. Some people believe in little green humanoids... :)

[amalahama]

Of course the style of flying AI controlling by the pure mathematics differs a bit from the style human flues. It does not perform wobbling stick movement to feel the plane reaction.

 

Well not so much. As you know, in aeronautical control system design, pilot is inserted inside the control loop like a PID controller with fixed coefficients, in order to help to avoid Pilot-induced-oscillations. However, in DCS it looks like AI pilots knows perfectly how much amount of stick they have to push to perfectly fit to the wanted trajectory. It doesn't look like a feedback system but like a open-loop one. But I tested it some time ago, I will test again with over-weighted aircraft to be sure...

 

Regards!!

Edited June 1, 2009 by amalahama

[Yo-Yo]

Well not so much. As you know, in aeronautical control system design, pilot is inserted inside the control loop like a PID controller with fixed coefficients, in order to help to avoid Pilot-induced-oscillations. However, in DCS it looks like AI pilots knows perfectly how much amount of stick they have to push to perfectly fit to the wanted trajectory. It doesn't look like a feedback system but like a open-loop one. But I tested it some time ago, I will test again with over-weighted aircraft to be sure...

 

Regards!!

 

 

:) :)

 

I can not understand your statement... what signes of feedback systems presence do you know? Oscillations at transients? But the system tuned in a right way has aperiodic transient responce.

Moreover as you can see sometimes AI do have oscillations in bank most of all.

And, by the way, how can you design open loop system to aim the moving target, for example?

[ericinexile]

I want to steer this thread away from issues of turbulence and wind-correction. What I'd rather like to know about A/I flight modeling is how A/I aircraft perform when compared to their human flown counterparts. Do two identical planes/helicopters, one human-flown the other A/I, share turn-rates, climb-rates, max/min speeds, sink-rates, etc for a given loadout? If they do then the appearance of flying on rails or lack of turbulence reaction is irrelevant to me; although I would like to see control surface movement (heck, even Falcon 4 did that).

 

Smokin' Hole

[amalahama]

:) :)

 

I can not understand your statement... what signes of feedback systems presence do you know? Oscillations at transients? But the system tuned in a right way has aperiodic transient responce.

Moreover as you can see sometimes AI do have oscillations in bank most of all.

And, by the way, how can you design open loop system to aim the moving target, for example?

 

Mmmm... Ok maybe I'm misunderstanding AI behaviour. My theory is that a high level AI builds trajectories for each AI aircraft in real time, having into account his mission, combat envelope and aircraft characteristic, and then, there is a more low level AI, a individual controller linked to SFM in each aircraft, which try to keep the trajectory that the high-level AI have said to be done.

 

In this case, you can use an open loop control if you know perfectly how aircraft is going to act and you know without any doubts the trajectory you have to do.

 

It's something similar to FD-AP link; FD says AP what trajectory have to follow and AP try to keep in track all the time. AP is a fedback controller this time obviously.

 

If you are using feedback controller, maybe the point is you are using too responsive a controllers that it looks unrealistic, because pilots aren't.

 

Regards

[Yo-Yo]

I want to steer this thread away from issues of turbulence and wind-correction. What I'd rather like to know about A/I flight modeling is how A/I aircraft perform when compared to their human flown counterparts. Do two identical planes/helicopters, one human-flown the other A/I, share turn-rates, climb-rates, max/min speeds, sink-rates, etc for a given loadout? If they do then the appearance of flying on rails or lack of turbulence reaction is irrelevant to me; although I would like to see control surface movement (heck, even Falcon 4 did that).

 

Smokin' Hole

 

 

OK. What I like to say in this thread is what the equations fixed wing AI is driven are the same (generally it is the common code) for AI and human SFM. For example, if you are dogfighting in Su-27 vs Su-27 AI the final of the fight depends not only on your skill but on fuel loading. Of course you can not test it :) so you have to believe me. I spent many hours fighting with new bots when we were teaching them to fight.

 

When A-10 is ready its AI "alter ego" will have the same trajectory parameters as AFM-M A-10 because the main AD data, engine thrust, etc will be taken from the AFM.

 

For helicopters AFM and reduced AFM-AI energetic parameters are practically the same regardless of simplified engine and rotor model. The differences can be in the short-period movement, but as you can see the AFM-AI looks like natural.

[ericinexile]

^^^^

Of course I believe you. ED hasn't lied to me yet as far as I know. All of this stuff is complete magic to me anyway. I am amazed and astounded at the level of detail even in areas where few will notice or care. Yesterday my wingy flew under me while maneuvering to rejoin and my radar altimeter jumped! That happens to me all the time flying over the ocean. A completely unnecessary detail when all that needed to be modeled was the difference between helo and terrain with a simple algorithm correcting for any non-level attitude. But nooo, ED had to do it the hard way. It's an amazing attention to detail that sometimes leaves me shaking my head in wonder. As I said, LOMAC/DCS is an incomprehensible bit of magic running on my PC that in many ways, most ways actually, exceeds the realism of $20M simulators I'm forced to fly annually. It is probably in my wonderment that can be sourced my occasional doubt. Certainly the devs must be cheating?!

 

Smokin' Hole