AI "overperformed"

[Raven (Elysian Angel)]

1 minute ago, Silver_Dragon said:

it's very strange that these problems with the AI continue to exist

Yes, it is. But to paraphrase ED, "we will continue to have to be patient" 

[Pikey]

@Lidozin I'm not picking on you, i'm defintiely picking on your argument, its got more reversals than an IL2 Sopwith Camel dogfight. Here's your 3 page argument condensed down and fall apart, summarized, with quotes since you like empirical data.

Step 1. Claim the FM is fine.
Step 2. Be told its not applied consistently and thus it is in doubt.
Step 3. Argue that you were only talking about Combat routines and not non combat routines like land/takeoff, follow/escort
Step 4. Show that by verifying non combat routines like climb performance, it proves the AI is following physical models and limits all the time in combat!

Am I  the only one here noticing this?

You can say the FM observes the rules sometimes: Accepted. You cannot say that because it uses physics soemtimes that it always does.

You know, and everyone knows here that the software chooses when to use flight models, but the key to knowing that its using a flight model is to wether the moment to moment decisions are natural, or its a repeat of a sequenced set of events sewn together to look like its real. And that is what you miss. WHich is why I say you dont play the game enough to notice. Those loops. I see the WW2 planes also using them. Its just canned sequential responses, not a real FM and its not staying within physics between these sequences. They can do it forever.

You can be 100% correct, 1% of the time. But you can't use the example of being right once as evidence that you are always right!

The point is that AI strings together canned tracks and puts them together. You need to look at the enitre picture holistically. Its software, its simplified, its designed to work well enough for casual scrutiny, but when you put the whole picture together, it collapses, along with your argument that the AI observes physics.

Where is ED's GFM they talked about? The one that should react properly to physics, they said. By your reckoning, we dont even need it!
 

On 7/10/2025 at 8:11 AM, Pikey said:

Ive got some news for you. You built your premise on an assumption, that the physics model is being used all the time.

See those routines AI performs... See how they follow patterns...see how they snap together so neatly... or maybe you aren't looking.

Let me tell you something you apparently don't know or have ever seen. it will come as a shock.

 

Ai doesn't use physical models during all aspects of flight. 

On 7/10/2025 at 7:53 PM, Lidozin said:

The original discussion wasn’t about general AI behavior across all mission stages. It was specifically about how the AI performs in a dogfight, and whether the AI's flight model during combat is based on real aerodynamic parameters that correspond to those of the real aircraft.

On 7/13/2025 at 12:06 PM, Lidozin said:

I’m not suggesting that the AI behaves perfectly in every respect — only that, in this specific context, its energy performance in sustained climb matches both the manual and computed data to within a few percent. That’s not “superhuman” — that’s simply a correct implementation of aerodynamic tables.

The formation example is a common misunderstanding:

On 7/12/2025 at 4:16 PM, Lidozin said:

The confidence in how the simulation behaves stems from well-established knowledge of the trajectory model in use. 

[..snip..]

It’s a modest investment of effort, but it yields clear data: either the simulation behaves as predicted by the aerodynamic tables, or it does not — and in either case, we move from speculation to grounded evaluation. The test can also be easily shared and repeated by others, allowing for open verification.

 

On 7/13/2025 at 1:20 PM, Lidozin said:

That said, I’d like to remind everyone that the original goal of this analysis was not to examine AI behavior in terms of tactics or input realism, but simply to test the claim that the AI “doesn’t obey physics, or has physical performance beyond what a player-controlled aircraft can achieve.” The flight test results suggest otherwise. Let’s avoid shifting the discussion away from that specific and measurable question.

I can get you your empirical data that AI doesnt observe physics, but its more fun listening to the various ways you avoid finding that important.

[Bowie]

Am just starting in w/ the A-4E/F-5E v. MiG-15bis/MiG-21bis - Vietnam, off the short fields and On Yankee Station, on the Marianas II map.

Coming from '44 Normandy and the P-51D/P-47D v. Bf-109G6/Fw-190-A8, generally set All flights(8) to 1(Ace)/4(Veteran)/3(Trained), with no nuggets.

Aces - Are nearly unbeatable, w/o luck and an energy advantage.

Veteran - Are tough, dogged and relentless, but beatable.

Trained - Will keep you sharp, but can be beaten with basic ACM skills.

Rookie - 1G strafing targets.

 

Gulf of Tonkin, 1967? MiG-15bis...?

Veteran - Are Terminators.

Pure assassins, that will fly up behind a hard maneuvering, high skill (Ace/Vet) pilot in an ACM loaded A-4E/F-5E, and drop'em like a 1G strafing target.

Forget about the Aces.

 

Have dropped the NVA skill-set to Veteran/Rookie/Trained/Rookie per division(4), and still need numerical superiority to break even.

Much less run the table, or maybe with a coupla losses, like the '44 Normandy scenario.

 

They need at least a two performance/skill level shift down.

 

Bowie

 

[Lidozin]

On 7/15/2025 at 3:05 PM, Pikey said:

@Lidozin I'm not picking on you, i'm defintiely picking on your argument, its got more reversals than an IL2 Sopwith Camel dogfight. Here's your 3 page argument condensed down and fall apart, summarized, with quotes since you like empirical data.

Step 1. Claim the FM is fine.
Step 2. Be told its not applied consistently and thus it is in doubt.
Step 3. Argue that you were only talking about Combat routines and not non combat routines like land/takeoff, follow/escort
Step 4. Show that by verifying non combat routines like climb performance, it proves the AI is following physical models and limits all the time in combat!

Am I  the only one here noticing this?

You can say the FM observes the rules sometimes: Accepted. You cannot say that because it uses physics soemtimes that it always does.

You know, and everyone knows here that the software chooses when to use flight models, but the key to knowing that its using a flight model is to wether the moment to moment decisions are natural, or its a repeat of a sequenced set of events sewn together to look like its real. And that is what you miss. WHich is why I say you dont play the game enough to notice. Those loops. I see the WW2 planes also using them. Its just canned sequential responses, not a real FM and its not staying within physics between these sequences. They can do it forever.

You can be 100% correct, 1% of the time. But you can't use the example of being right once as evidence that you are always right!

The point is that AI strings together canned tracks and puts them together. You need to look at the enitre picture holistically. Its software, its simplified, its designed to work well enough for casual scrutiny, but when you put the whole picture together, it collapses, along with your argument that the AI observes physics.

Where is ED's GFM they talked about? The one that should react properly to physics, they said. By your reckoning, we dont even need it!
 

I can get you your empirical data that AI doesnt observe physics, but its more fun listening to the various ways you avoid finding that important.

Most of the frustration and speculation expressed in this thread seems to stem from combat-related behavior. I haven’t come across many complaints about AI taxiing, takeoff, or landing. And to be honest, those phases don’t particularly interest me either, since I primarily view the AI as a sparring partner in aerial combat.

Now, climb performance represents a critical component of combat behavior ( it’s energy gain at 1g, and while it doesn’t occur in isolation that often, it fully defines acceleration in level flight and in shallow dives) both of which are common in real engagements.

What I’ve shown is that in this regime, the AI follows the physics defined in its data tables and behaves exactly as the real aircraft would according to flight test documentation. That alone should dispel many doubts.

What do we observe more often in dogfights? Sustained or transient turning flight with increased load factors, where energy is either lost or traded in ways governed by well-known aerodynamic relationships. I also showed that the aerodynamic data used for the FM (lift, drag, thrust) supports correct energy behavior in those turning regimes. So far, everything lines up.

However, to completely rule out the suspicion that the AI is “cheating” in these cases, the next logical step would be to analyze a 1v1 fight recording where both the player and AI aircraft are of the same type. Specifically, you’d export the time history of TAS, altitude, and G-load for both. Using known energy equations, one can compute the specific excess power and compare it to the observed load factor.

If the AI is cheating — by bypassing the FM or using hidden scripts — it would become immediately obvious. Either its energy behavior would be physically implausible, or you'd see clear discontinuities or artifacts in the data.

I believe this type of empirical comparison would cut through all the theoretical debate and provide developers with a solid foundation for any investigation or follow-up.

 

[Raven (Elysian Angel)]

1 hour ago, Lidozin said:

I haven’t come across many complaints about AI taxiing, takeoff, or landing.

That's you telling people how new you are to DCS and/or the forum without telling you're new 

Ah the entertainment value of this thread is priceless. Please keep going! 

[diego999]

1 hour ago, Raven (Elysian Angel) said:

That's you telling people how new you are to DCS and/or the forum without telling you're new 

Ah the entertainment value of this thread is priceless. Please keep going! 

 

Just last night my AI wingman crashed into me while I was lining up for landing.

In the same sortie, two C-130 decided to block each other at the end of the runway, effectively ruining the entire mission.

Does this count as a complaint? Or do I need to fill a form?

Edited Wednesday at 04:16 PM by diego999

[MAXsenna]

 
Just last night my AI wingman crashed into me while I was lining up for landing.
In the same sortie, two C-130 decided to block each other at the end of the runway, effectively ruining the entire mission.
Does this count as a complaint? Or do I need to fill a form?

Form is attached in the OP.

Sent from my SM-A536B using Tapatalk

[Pikey]

I'm going to play with this sarcastically. I'm going to parody the ideas in Comic Sans. Once more, its not personal, I'm sure I'd like you as a person, just the ideas don't sit well with me and its got to a point where its a bit funny now. Heres the sracasm highlights.

We see an SFM table in the coremods of the plane lua. This means that the AI uses this when calculating its dogfight state table. We know that the existence of the SFM means we only need to look at the actual model, not how its executed or used in the software. The fact that the table exists means DCS uses it. It doesnt matter that the C code is encrypted and unreadable, its only the LUA. And anyway, I can read the program, I know things for sure. Just use a stopwatch.

We can rule out misuse of the SFM data because that doesnt make sense. Why would a game need to force the AI to do something at all, when it flies realistically? Stupid.

There is no evidence of the AI not using the SFM data. Apart from the times when it doesnt, like when its following the player in line abreast perfectly and can turn at the same time and aero brake stop 100kts in 1 second. But this doesnt actually matter because we know because of the SFM data that the AI doesnt behave like that during dogfights. SFM = dogfight and climb. Everything else can be scripted, thats ok, but we know for sure its not breaking this rule.

We can prove this by looking at when the AI isn't doing something stupid, like climbing. If we examine the climbing then we know for sure this must apply to the dogfight data.
When we look at tacview, we can see the moment to moment forces and speeds and alpha recorded for the AI. It's OK to see things here that are within the SFM data. Look, the plane is going at 500kts straight up. That's perfectly OK, so the entire fight must be fine here. I don't need to see the plane doign these 500kt climbs again and again, because its just flying perfectly, there is no issue over time with the energy state.

We havent seen any evidence of the AI using scripted behaviour. At least not between 3.30pm and 3.45pm in the afternoon of June 12th 2025. The AI would tell us he is using scripted behaviour through the comms menu.

We haven't seen the Scripts folder of the DCS application where the AI routines are kept. At least, the ones that apply to normal flight. The dogfight ones got moved some years ago to protect people keep on arguing about this non existent problem because they are just bad fliers and need to stop wasting their time looking for excuses.

We know the DCS AI is very good so the routines can be shared from MiG pilots to the Luftwaffe, so they can use boom and zoom too. Its a special trick, it might look the same, but actually each plane type, jet or prop can use the identical vertical manouvers and energy. But the SFM is what decides how it really is different.

We also know that ED eventually gave up their special Flight model in development after realising it was pointless. It was pointless because the AI already was perfect. Why develop something to make it different when it is already the best? 

Also we know a lot of things about planes and so we've marked ourself as the solution very early on in this thread. Quite simply, they dont understand about SFM, its only for aeronautical engineers and high IQ. So being the solution saves time and is more effective in conversations.

 

Edited Wednesday at 04:22 PM by Pikey
typo

[Pikey]

14 minutes ago, diego999 said:

 

Just last night my AI wingman crashed into me while I was lining up for landing.

In the same sortie, two C-130 decided to block each other at the end of the runway, effectively ruining the entire mission.

Does this count as a complaint? Or do I need to fill a form?

Yes, please fill out this accident form:

I would definitely recommend this AI to a friend or family. (Tick a number below)
10   10   10   10   10   10   10   10   10   10 

[MAXsenna]

@Pikey Thanks, I had to switch to Web to get the correct font!  

[Lidozin]

2 hours ago, Pikey said:

I'm going to play with this sarcastically. I'm going to parody the ideas in Comic Sans. Once more, its not personal, I'm sure I'd like you as a person, just the ideas don't sit well with me and its got to a point where its a bit funny now. Heres the sracasm highlights.

We see an SFM table in the coremods of the plane lua. This means that the AI uses this when calculating its dogfight state table. We know that the existence of the SFM means we only need to look at the actual model, not how its executed or used in the software. The fact that the table exists means DCS uses it. It doesnt matter that the C code is encrypted and unreadable, its only the LUA. And anyway, I can read the program, I know things for sure. Just use a stopwatch.

We can rule out misuse of the SFM data because that doesnt make sense. Why would a game need to force the AI to do something at all, when it flies realistically? Stupid.

There is no evidence of the AI not using the SFM data. Apart from the times when it doesnt, like when its following the player in line abreast perfectly and can turn at the same time and aero brake stop 100kts in 1 second. But this doesnt actually matter because we know because of the SFM data that the AI doesnt behave like that during dogfights. SFM = dogfight and climb. Everything else can be scripted, thats ok, but we know for sure its not breaking this rule.

We can prove this by looking at when the AI isn't doing something stupid, like climbing. If we examine the climbing then we know for sure this must apply to the dogfight data.
When we look at tacview, we can see the moment to moment forces and speeds and alpha recorded for the AI. It's OK to see things here that are within the SFM data. Look, the plane is going at 500kts straight up. That's perfectly OK, so the entire fight must be fine here. I don't need to see the plane doign these 500kt climbs again and again, because its just flying perfectly, there is no issue over time with the energy state.

We havent seen any evidence of the AI using scripted behaviour. At least not between 3.30pm and 3.45pm in the afternoon of June 12th 2025. The AI would tell us he is using scripted behaviour through the comms menu.

We haven't seen the Scripts folder of the DCS application where the AI routines are kept. At least, the ones that apply to normal flight. The dogfight ones got moved some years ago to protect people keep on arguing about this non existent problem because they are just bad fliers and need to stop wasting their time looking for excuses.

We know the DCS AI is very good so the routines can be shared from MiG pilots to the Luftwaffe, so they can use boom and zoom too. Its a special trick, it might look the same, but actually each plane type, jet or prop can use the identical vertical manouvers and energy. But the SFM is what decides how it really is different.

We also know that ED eventually gave up their special Flight model in development after realising it was pointless. It was pointless because the AI already was perfect. Why develop something to make it different when it is already the best? 

Also we know a lot of things about planes and so we've marked ourself as the solution very early on in this thread. Quite simply, they dont understand about SFM, its only for aeronautical engineers and high IQ. So being the solution saves time and is more effective in conversations.

 

Thank you for the entertaining interlude — sincerely. It's always refreshing to see people stay engaged, even in parody. However, as enjoyable as it was to read, I couldn’t find in it anything resembling a technical counterpoint to the tested climb profile or the data comparison with the real-world reference chart.

Regarding the 500 knots straight up — if that was a serious remark, I would kindly ask for clarification. A well-trimmed MiG-15 starting from 950+ km/h (which is about 510 knots) absolutely can convert kinetic energy into altitude for a short time — that's basic energy conservation and directly tied to its dynamic ceiling. There's nothing unnatural about it unless you're claiming the AI sustains it indefinitely, which is easily testable in TacView or even with a stopwatch and status bar, as already demonstrated.

You’re very welcome to propose a reproducible test that demonstrates any claimed violation of physics. If it's testable, measurable, and repeatable — I'm all ears.

Otherwise, I’d suggest we let the data speak. Because ten minutes of quiet measurement saves hours of speculative back-and-forth.

[Pikey]

On 7/10/2025 at 7:53 PM, Lidozin said:

If there’s a mismatch in energy performance, turn rate, climb, etc., under those circumstances — that’s something worth looking into. But if the concerns are about form-up logic, taxiing behavior, or scripted transitions, those are separate layers of the simulation, and not what’s being discussed when we refer to the AI using a physics-based trajectory model during combat.

You already said you wont accept AI doing unphysical things as evidence it does unphysical things, based on when in the game it does them - "Layers" you call them. You then accepted climbing outside of combat as evidence it supports that AI doesnt cheat in dogfights. Meanwhile you conveniently ignore that AI picks when and how it observes the SFM according to no logic that you see, it is in fact, at code level for which you continue to assume, is bound in physics. 

Simply put, you've made an assertion and said 'prove me wrong'. Which is shifting the burden of proof and a logical fallacy. In fact its a well known situation called Hitchens razor in that What can be asserted without evidence can also be dismissed without evidence. You disagree because you think you've supplied evidence that the SFM is working because you observed it at some point(s) doign so. Unfortunately it doesnt work that way. By the same token, God exists because you saw a miracle.

This is not solvable empirically, it only needs reasonable doubt about how the AI does work. And that reasonable doubt can come from observational data, not just empirical data. The reaosonable doubt is, because I saw a plane float, I know AI is not confined to physics. Its software and has no regard for Physics and chooses when and where to use it as a convention that helps the game play elements of the game. This is all that is needed to dismiss your assertion, no matter what you want to throw at it. The burden of proof falls upon YOU to prove that the AI always uses Physics. Not sometimes, but always. Until then I shall continue to spectate, just in case you manage to satisfy my curiosity. Because I am open to all things, even if you have code level access.

Continue with your argument. WHy dont you try a counter hypothesis approach where you formulate a hypothesis that proves your own is wrong, seek to eliminate that instead, like an actual scientist.

[Temetre]

Am 15.7.2025 um 13:58 schrieb Silver_Dragon:

I still think most AI aircraft continue to use SFM, with the corresponding problems it may have, such as cases where the AI inexplicably runs out of fuel when we still have fuel left, as if they were always running the engine at 100% or using afterburner, burning fuel rapidly.

We've talked about GFM... but has anyone confirmed that any current DCS World aircraft has it implemented? We haven't heard from anyone about it for years.

IIRC the current stance is 'GFM is gonna come in the future', and thats about all we know. Its not in the game and who knows when if ever itll be a thing.

Am 15.7.2025 um 13:58 schrieb Silver_Dragon:

As for the MiG-15Bis, I'm afraid it's an inherent problem with the old code, because it's very strange that these problems with the AI continue to exist, and that, given everything that's been said, it hasn't been minimally resolved.

I suspect its so broken that ED wants to just redo a lot of stuff with the GFM, and therefor doesnt bother with bandaids? Theyve done some improvements for Mig-29, but not much more.

Cant say I like that logic; considering how far away the GFM seems and how damaging these issues are to the experience, bandaids could go a long way. See the Mig-15 mod, a single person fixing a lot of stuff. 

[Raven (Elysian Angel)]

41 minutes ago, Temetre said:

Cant say I like that logic; considering how far away the GFM seems

Neither do I: we were supposed to get the GFM in 2023.

42 minutes ago, Temetre said:

See the Mig-15 mod, a single person fixing a lot of stuff. 

Exactly! If we would have the same thing for the MiG-21  and the F-5 (those 3 are the worst offenders), it would go a long way with keeping people patient.

That being said, the helicopter AI FM also needs "some" work (*cough*).

[MAXsenna]

1 hour ago, Raven (Elysian Angel) said:

That being said, the helicopter AI FM also needs "some" work (*cough*).

Are you referring to those invisible rocket boosters?  

[Raven (Elysian Angel)]

1 hour ago, MAXsenna said:

Are you referring to those invisible rocket boosters?  

Exactly those yes, and the handbrakes too!

[MAXsenna]

Exactly those yes, and the handbrakes too!

Ah, yeah! Forgot about those!

Sent from my SM-A536B using Tapatalk

[Dragon1-1]

On 7/15/2025 at 1:58 PM, Silver_Dragon said:

I still think most AI aircraft continue to use SFM, with the corresponding problems it may have, such as cases where the AI inexplicably runs out of fuel when we still have fuel left, as if they were always running the engine at 100% or using afterburner, burning fuel rapidly.

Everything so far had been SFM, GFM has not been released yet.

Fuel issues is mostly because AI sucks at managing their fuel. They use AB willy-nilly, fly at full power as opposed to efficient cruise, and don't respect any sort of bingo, so they continue wasting gas until they fall out of the sky.

Edited Thursday at 11:30 PM by Dragon1-1

[Bowie]

Interesting Workaround:

Started w/ Curly's "MIG-15bis Accurate SFM Mod" (a big thanks)

And then added:

263    M_empty                     = 3991 ,    -- [MiG-17] (+238 kg)  -- with pilot and nose load, kg -- 3753 ,    -- with pilot and nose load, kg ,

264    M_nominal                  = 5340 ,    -- [MiG-17] (+296 kg)  -- kg -- 5044 ,    -- kg

To reflect the A/C as a MiG-17, as is being utilized.

Superman - loses his cape.

At the lower Vet/Rookie/Trained/Rookie skill set, their performance is similar to that of the A-4E/F-5E, as wing-loadings and T/W ratios' would predict.

So, with parity, they are now more vulnerable to the 1960's missile technology, AIM-9B.

Will be upping the skill set back to 1/5/3.

 

Bowie

Edited 18 hours ago by Bowie