Jump to content

adjust data for AI WWII Airplanes to historically correct values


B4ron

Recommended Posts

Dear DCS WWII developers,

 

DCS-World WWII collection is one of the most realistic available flight simulators of WWII aircrafts. The player flight model is developed with a lot of effort to meet the historic role model. Annoyingly there is some incorrect data in the simplified AI flight model which lead to historical incorrect behaviour of AI aircrafts. There are a lot of discussions in the forum about AI flight model and I understand that the PFM cannot be used for AI in order of performance reasons.

 

What can be done IMO is to adjust the SFM to historically correct data. I used to adjust that data by myself but since update 2.5.6 for the stable version all aircraft.lua files are encrypted.

 

Following data affecting the flight performance of AI aircraft need to be corrected in the aircraft.lua files in order to refer to historical data:

 

1. Adjust weight settings:

 

Bf109K-4:

M_empty = 2947kg -- was 2900.65kg (calculated from M_nominal)

M_nominal = 3400kg -- was 3100kg (Report A/IV/294/44 app. 5026/28 of OBB Forschungsanstalt Oberammergau)

M_max = 3900kg -- was 3400kg (max. bomb load) => with 3400kg the plane will be overloaded with the 500kg bomb!

 

[ATTACH]239416[/ATTACH]

 

FW190A-8:

M_empty = 3640kg -- OK

M_nominal = 4300kg -- was 4273kg (document nJ.190.801.-051 of Focke-Wulf Flugzeugbau G.m.b.H)

M_max = 4851kg -- was 4840kg (max. bomb load)

M_fuel_max = 388kg -- was 409kg (main tank = total fuel 640l - rear tank 115l)

 

fw190-a8-climb-13nov43.thumb.jpg.aee405a6368b012098f53b73a5aeb348.jpg

 

FW190D-9:

M_empty = 3613kg -- was 3600 (calculated from M_nominal)

M_nominal = 4273kg -- OK (document FW5232 of Focke-Wulf Flugzeugbau G.m.b.H)

 

fw190d9-climbchart-flugmechanik-24-3-45.thumb.jpg.a8dd9d8dc1da8c7876d446a8019c8329.jpg

 

P-51D:

M_empty = 3576kg -- was 3549 (calculated from M_nominal)

M_nominal = 4593kg -- was 4000 (Flight Tests on the North American

P-51D Airplane, AAF No. 44-15342 + 60gal fuel capacity left)

M_max = 5260kg -- was 4800 (wikipedia.org)

 

P-51D_15342_Climb-Fig5.thumb.jpg.e1fede8fa4e8116b08409659af667064.jpg

 

SpitfireLFMkIX (Merlin 66):

M_empty = 3004kg -- was 2914.0 (calculated from M_nominal)

M_nominal = 3395kg -- was 3100 (spitfireperformance.com)

M_max = 3837kg -- was 3400 (max. bomb load)

 

bs543climb.thumb.jpg.380c5ed7bf726beb89e74c08261c6328.jpgab197.gif.1bd8887b14f72188fb5448d6704c67ce.gifab197wl.gif.2a6ee4a8cb669ce59ce0f567795686f4.gif

 

2. Adjust weapon effects

Weapon effects on AI airplanes are generally to low due to historical data.

For german weapons the following average number of hits have been reported to kill a plane:

 

MK108 30mm:

1-2 hits for single engine fighter aircrafts (british trials of German 30mm ammunition)

1-2 hits for twin engine medium bombers (british trials of German 30mm ammunition)

A very good summary with historical data for the test can be found on:

https://forum.il2sturmovik.com/topic/41665-3cm-mk108-historical-test-data-vs-ingame-test-data/

 

MG151 20mm:

4-5 hits for single engine fighter aircrafts

20-30 hits for heavy bombers (wikipedia.org / "Messerschmitt Bf 110: Die Rehabilitierung eines Flugzeugs" ISBN 978-3837022896)

 

To get a closer to real weapon effect the value "lifebar" should be adjusted to:

life = 9; -- lifebar -- was 18

 

This is still too low compared to historical weapons tests IMO but on about the same level like for player flight model.

 

3. Bf109K-4 MK108 rounds

MK108 motor cannon was equipped with 65 rounds (wikipedia.org + several more sources). The value "count" is missing in Bf109K-4.lua for MK108 => the count will be taken from weapons.lua which is set to 60. (This is true for MK108 wing mounted guns.)

 

There are more incorrect data to be found in the aircraft.lua files which however do not have affect to the AI flight behaviour according to my testing.

 

Thanks

B@ron

5026_28_DBSonder_MW_steig.thumb.jpg.ab18cab9d3df5bc81d9a795b6e2e0632.jpg


Edited by B@ron
Link to comment
Share on other sites

What can be done IMO is to adjust the SFM to historically correct data.
Yeahhhhhh, nope. Do you know something about IA coding? Despite the name IA is dumb, and compromises have to be done in order to make it both credible and challenging enough since a computer doesn't "think" by itself. For instance, one of those tools programmers do use is something called rubberband IA, which BTW I don't know if it's used in DCS but my guess is it is from those claims of invincibility some people do speak. Basically regular rules doesn't apply to AI and they get a "boost" in performance to cope with their inability to "think", "decide" or "reason", like a human would.

 

 

DCS IA is being revamped indeed, specially to tackle with WWII stuff, and a new DM is coming which I believe would affect IA either but for a better experience go multiplayer against humans. If that's not an option because you think you would loose every combat to humans, set IA to rookie :thumbup: .

 

 

 

S!

"I went into the British Army believing that if you want peace you must prepare for war. I believe now that if you prepare for war, you get war."

-- Major-General Frederick B. Maurice

Link to comment
Share on other sites

I don't want AI opponents to be easier to kill, that's not the point. I'm frequently winning air combats against AI planes on high or excelent skill level.

 

The point is that DCS is a flight simulator with a claim to be as close to historic realism as possible. This should also be true for AI planes since you are interacting with them. This involves to set up correct data, in order to be at least as close as possible to the historic role model.

 

1. weight settings in aircraft.lua files:

This is the settings that will be shown in mission editor. These weight settings influence the flight behaviour of both AI and player aircrafts. I tested it! In order to have a simulation closer to reality, those weight settings should be according to historical data.

 

2. weapons effects:

Historical tests and flight records show specific damage for certain weapon hits. If a AI P51 is still flying after 6 rounds of MK108 ammunition hits (there's different ones, I know) this can't be close to historic realism. Also if you consider AOA and approach speed (which by the way shouldn't have to have much effect on HE ammunition with a filling of 85g of TNT). 6 hits would have pretty shurely brought a B17 down! The current player DM is way more close to realism than the AI one. This can be adjusted with the "life" setting in the aircraft.lua.

 

3. amount of ammunition

This is just a bug in the Bf109K-4.lua file. The command "count = 65" for MK108 is missing. If you insert it, the setting will be correct. This is also valid for both AI and player aircraft.

 

I know that there's an improved DM supposed to come, but I doubt that this will solve all problems with AI.


Edited by B@ron
Link to comment
Share on other sites

I think what B@ron has to say sounds quite convincing. And it seems considerable effort has been spent to arrive at his conclusions.

Maybe ED devs can consider this information when looking at AI flight models. If possible even before the big bang AI FM revamp, since this seems to be mainly pointing at simple lua configurations.

Kein Anderer als ein Jäger spürt,

Den Kampf und Sieg so konzentriert.

 

Das macht uns glücklich, stolz und froh,

Der Jägerei ein Horrido!

Link to comment
Share on other sites

  • ED Team

These values are not for FM physics itself. They are potentially for using in rough estimations regarding their potential maneuvring ability.

The actual weights are the same human plane has, and they are calculated using mostly detailed weight and balance documents.

Ніщо так сильно не ранить мозок, як уламки скла від розбитих рожевих окулярів

There is nothing so hurtful for the brain as splinters of broken rose-coloured spectacles.

Ничто так сильно не ранит мозг, как осколки стекла от разбитых розовых очков (С) Me

Link to comment
Share on other sites

I don't want AI opponents to be easier to kill, that's not the point.
The point is, if you seek ultimate realism fight against humans only, mate.

 

S!

"I went into the British Army believing that if you want peace you must prepare for war. I believe now that if you prepare for war, you get war."

-- Major-General Frederick B. Maurice

Link to comment
Share on other sites

These values are not for FM physics itself. They are potentially for using in rough estimations regarding their potential maneuvring ability.

The actual weights are the same human plane has, and they are calculated using mostly detailed weight and balance documents.

 

They do definitely effect flight behaviour.

 

weight settings:

The total weight of AI and player aircrafts is calculated in the mission editor: M_empty+total fuel weight+total weight of ammunition (calculated from "round_mass" from weapons.lua)+weight of additional weapons (bombs and rockets). If you add an editional fuel tank or a bomb or reduce fuel, flight behaviour will change (especially the climb rate). Ergo also M_empty must have an effect! It can be tested by changing the value in stable version 2.5.5 or open beta (the aircraft.lua files have been encrypted or compiled in stable version 2.5.6 => this change has been taken back in open beta and hopefully will also be for stable version). If you set the value lower than your maximum thrust available, you'll even be able to climb vertically straight from the start (that's where those nonsense videos on youtube come from and likely the reason for encrypting the files). M_nominal and M_max are only values for comparison to the actual total weight IMO.

 

weapons effect:

If you set the value "life" from 18 to 9 you'll need 2 MK108 hits in average to get a P51 down (which is far closer to historic data and the player DM). If you set the value to 1, the AI aircraft will go straight back to the airfield for landing. You'll need only 1 MG131 hit to bring it down then. So this value will not effect weapons effects directly, but will set the "damage bar" for AI aircrafts to a lower initial value.

 

MK108 rounds for Bf109 motor cannon:

You can count the total ammunition by firing your whole ammo stock and count the number of tracer bullets. In standard setting every 3rd bullet is a tracer one. You'll also be able to see it on ammunition weight. So this setting definitely influences the ammo count for the player flight model.

Link to comment
Share on other sites

The point is, if you seek ultimate realism fight against humans only, mate.

 

S!

 

I agree. But for possibly historically based context I like the missions and campaigns that are also available as free user files. Why not doing those improvements (or bug fixes whatever you call it) when it's just simple settings?

Link to comment
Share on other sites

I agree. But for possibly historically based context I like the missions and campaigns that are also available as free user files. Why not doing those improvements (or bug fixes whatever you call it) when it's just simple settings?
You call programming a realistic and believable IA "simple settings"? Really? Ok…

 

 

S!

"I went into the British Army believing that if you want peace you must prepare for war. I believe now that if you prepare for war, you get war."

-- Major-General Frederick B. Maurice

Link to comment
Share on other sites

You call programming a realistic and believable IA "simple settings"? Really? Ok…

 

 

S!

 

The things OP is talking about are "seperated" and labelled lines in the lua. So yes, that makes it simple.

 

He didn't say about reprogram AI behaviour to be realistic.

Link to comment
Share on other sites

They do definitely effect flight behaviour.

 

So what you're saying there is that you know better than ED's main FM designer what these values are for and how they're used. OK...

The DCS Mi-8MTV2. The best aviational BBW experience you could ever dream of.

Link to comment
Share on other sites

So what you're saying there is that you know better than ED's main FM designer what these values are for and how they're used. OK...

 

No, that is not what he was saying.

Kein Anderer als ein Jäger spürt,

Den Kampf und Sieg so konzentriert.

 

Das macht uns glücklich, stolz und froh,

Der Jägerei ein Horrido!

Link to comment
Share on other sites

Bravo, B@ron. I have purchased all of the WWII modules and all but two of the rest. It would be very nice to have better AI behavior at high skill levels. Particularly with WWII era craft. There are rarely active WWII servers with a low enough ping for me to participate in PvP. Having sunk well over $1k into DCS, I find the situation somewhat depressing. The AI is either very stupid or slightly less stupid with the ability to bend physics.

 

I hope the AI sees some improvement soon.

Link to comment
Share on other sites

I'm not a programmer, but the AI flight models don't behave like they're in the same physical world. They never seem to stall, and every AI whether it is ww2 or modern in DCS does the same thing when in a dogfight: vertical loops. It's not enjoyable.

 

 

Banner EDForum2020.jpg

Link to comment
Share on other sites

I'm not a programmer, but the AI flight models don't behave like they're in the same physical world. They never seem to stall, and every AI whether it is ww2 or modern in DCS does the same thing when in a dogfight: vertical loops. It's not enjoyable.

As a tactic for AI, mostly vertical loops is kind of understandable. It's their behavior at the extremes that kills my immersion. It usually looks like they retain too much control at low speed, and have no torque effects to deal with (there's probably a much better way of putting this, but I only fly on the PC). You sometimes catch them in the act, but Tacview really makes it obvious. They're still beatable, but it's frustrating.

 

I find that for WWII, I'm best with the 109. I'll fly the F86F SoH - Dogfight mission borrowed from the F86 quick missions and change the aircraft to suit my mood. This places you head to head over Tunb Kochak. Since the P-47 came out, I've been in a WWII mood. So I switch between these modules, setting the AI to excellent. If I'm flying a 109, I only have trouble with the Spitfire. Even keeping up the energy state and knowing where it's going to be after my turn, that thing is a murder machine with excellent level AI. It's still dumber than a box of rocks... but it can turn so fast . The I16 is surprisingly good against AI opponents too. On the other end is the A8. I die to everything in that.

 

The Korean War era jets are probably the most frustrating though. Going against an excellent MiG15 in a F86 is just ugly. A MiG21 vs F5 fight is a close second. The "modern" modules are a bit different imo. The AI there is just plain dumb, and it's easy to exploit. Radar gun sights and predictable behavior make almost every conceivable matchup too easy. But I'm straying off topic there.

Link to comment
Share on other sites

As I see this seems to lead to nothing, I will upload the modded aircraft.lua files into the user files for download as soon as they are hopefully editable with the next stable update. I also did some modification on the aiming ability of AI opponents in order to behave more human like.

 

So everyone can decide whether to download or not...

Link to comment
Share on other sites

These values are not for FM physics itself. They are potentially for using in rough estimations regarding their potential maneuvring ability.

The actual weights are the same human plane has, and they are calculated using mostly detailed weight and balance documents.

 

Just one more question: Why did you encrypt or compile the data in these files then with update 2.5.6?

Link to comment
Share on other sites

Probably due to increased size of the files. They have lots of data for the new DM in now. Have been asking NL a couple times to see if we can get into them to find out whats in the new ammo belts but have been ignored so far.

9./JG27

 

"If you can't hit anything, it's because you suck. If you get shot down, it's because you suck. You and me, we know we suck, and that makes it ok." - Worst person in all of DCS

 

"In the end, which will never come, we will all be satisifed... we must fight them on forum, we will fight them on reddit..." - Dunravin

Link to comment
Share on other sites

Probably due to increased size of the files. They have lots of data for the new DM in now. Have been asking NL a couple times to see if we can get into them to find out whats in the new ammo belts but have been ignored so far.

 

As Nightstorm posted in this threat: https://forums.eagle.ru/showthread.php?t=274099&page=11, ED seems to have taken back the encryption of these files in open beta again. I hope this will also get into the stable version with the next update...

Link to comment
Share on other sites

I believe I checked them after reading that post but it was not the case (have been trying to figure out the new ammo belts). I’ll look again and double check.

9./JG27

 

"If you can't hit anything, it's because you suck. If you get shot down, it's because you suck. You and me, we know we suck, and that makes it ok." - Worst person in all of DCS

 

"In the end, which will never come, we will all be satisifed... we must fight them on forum, we will fight them on reddit..." - Dunravin

Link to comment
Share on other sites

I pitted 16, Blue Spitfire/P-47/P-51, against 16, Red 109/A-8/D-9, on Channel Map.

 

 

At 6500', 16500', and 26500'.

 

 

At all altitudes, the Blues wiped the floor with the Reds every time.

 

 

I don't know what this proves, but I was expecting a much closer match.

 

 

Reduce the Blue to 12, they still won easily.

 

 

Very entertaining watching.

 

 

..

I7 2600K @ 3.8, CoolerMaster 212X, EVGA GTX 1070 8gb. RAM 16gb Corsair, 1kw PSU. 2 x WD SSD. 1 x Samsung M2 NVMe. 3 x HDD. Saitek X-52. Saitek Pro Flight pedals. CH Flight Sim yoke. TrackIR 5. Win 10 Pro. IIyama 1080p. MSAA x 2, SSAA x 1.5. Settings High. Harrier/Spitfire/Beaufighter/The Channel, fanboy..





..

Link to comment
Share on other sites

  • 2 weeks later...
I'm not a programmer, but the AI flight models don't behave like they're in the same physical world. They never seem to stall, and every AI whether it is ww2 or modern in DCS does the same thing when in a dogfight: vertical loops. It's not enjoyable.

 

ED stated how they made this : their AI is "perfect" in terms of aircraft management, they pilot their plane in such a way that they induce the minimal drag possible, they retain the maximum speed possible, etc, etc....

Imho this results in UFO behaviour since NO ONE ever did that, ever.

They also have perfect knowledge of opponent energy state and slightly change their pattern accordingly to maximise energy difference gain.

Imho this is kind of a lazy "solution" around "IA will never be good anyway" (a poor argument I've read, why bother making IA then if that is your starting principle? Not saying ED has that principle, but that shouldn't be used as an argument, tbh...).

 

As well, I've yet to see team tactics from AI. That would be a good direction to look for improvement instead of making unrealistic perfect beings.

Whisper of old OFP & C6 forums, now Kalbuth.

Specs : i7 6700K / MSI 1070 / 32G RAM / SSD / Rift S / Virpil MongooseT50 / Virpil T50 CM2 Throttle / MFG Crosswind.

All but Viggen, Yak52 & F16

Link to comment
Share on other sites

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...