AI gunnery mod

[Stonehouse]

Based on Halfoat's work here Fewer icons & slight AI gunnery change - Mods - IL-2 Sturmovik Forum (il2sturmovik.com) I've been mucking about with the AI aircraft text files. Specifically, I used what Halfoat did as a base and expanded it to cover all the fighter/heavy fighter aircraft to change the ranges the different skill levels engage at. Halfoat's premise was that the original values are such that the Novices begin to fire only at close range and Aces begin to fire at long range (but very accurately) and this seemed backwards to what you read in biographies and general references. ie Novice pilots tended to misjudge range and open fire too far away and break off too soon scared of collision, aces fire at close range and kill with short bursts efficiently. 

 

I did some investigation on the accuracy of bomber gunners which might be useful to others. See pdf AI gunner info and IL2 gunner error spreadsheet. Assuming I haven't cocked things up it seems like generally the relative velocity differential between the gunner and attacking aircraft is the biggest factor in inducing error in the gunners aim. I used this info to make a custom version of the BlindGunners mod as it helped me tweak values. Note that attacking from dead astern with little G loading (attacker not jinking) with little relative velocity difference is as expected not a good idea if you look at the errors at range.

 

Formula behind the pdf is:

error = tan( random number * ( AddCoef + (ProjSpeedCoef * speed diff) ) * ( (currrent overload * OverloadCoef) + CoefLow/Med/Hi/Ace) )  * distance to target

 

See turretcontrollerai.txt for explanation of parameters. Comments re constants below from Blindgunner thread

 

Note:

random number is a number between -1.0 and 1.0

 

stock skill coefficients are:

CoefLow        = 10.0f;
CoefMed        = 4.0f;
CoefHigh    = 2.0f;
CoefAce        = 1.0f;

So references to Skillx1.5 etc in the pdf means 1.5 or whatever times values above.

 

Constants:

 

OverloadCoef = 0.05f; 

// overload conversion rate to aiming units - believe Overload is the G loading on the gunner

//Comment: => this seems to be the effect on G forces, so how much the accuracy is affected by G-forces.
//High value: gunner is very inaccurate when plane is maneuvering.
//low value: gunner does not care about G forces at all

 

ProjSpeedCoef = 0.2f;

// coefficient of conversion of the speed difference of the object on which the turret and targets stand, in the aiming plane, to the error angles
//comment: setting this from 0.2 to 0.08 seemed to have quite an impact on accuracy
//high value: speed difference causes high inaccuracy
//low value: gunner will always be a sniper and has perfect lead to target

 

AddCoef = 0.05f; // default aiming error rate

 

So, following on from Halfoat I played around with the values testing things in AQMB and altered the AI text files to use these values below. The min is the range the AI stops firing, and the max is the max range they will start to fire at. At present these are as below:

Skill	Min	Max	Equivalent in yards
Novice	100	800	874
Normal	70	600	656
High	50	500	546
Ace	30	400	437

 

Stock values

Skill	Min	Max	Equivalent in yards
Novice	50	400	437
Normal	50	500	546
High	50	600	656
Ace	50	800	874

 

Using these values, you will see Novices spray and pray, and Aces tend to hold fire and shoot short bursts that kill things. Other skills are in between. This definitely does make a difference and getting bounced by an Ace is usually lethal and you often don't know what hit you. Get attacked by a novice and you might get peppered a little but unless you are unlucky you are still in the fight. Remember that there is usually a rate of closure involved. So, an ace will be quite close by the time of the end of their short burst. Only difference to Halfoat's work at this point is that I use slightly different values and expanded things to cover all the fighter type a/c.

 

The other parameter I came across and theorized about was "HitRateToDodge". Stock values are Novice = 2 ranging up to Ace = 5. I do not know but at this time believe it is the number of hits received before jinking/breaking off an attack. I think it is meant to be related to morale and tenacity. A novice will take a couple of hits and dodge, but an Ace will persevere until taking 5 hits. Tests in AQMB seem to confirm this theory.

I've been trying this value modded to 1 for all skills thinking that it might stop AI being so nonresponsive under attack at times. So far it seems to help but it isn't conclusive. 

Anyway, for whatever it's worth I thought to put the mod out here for people to play around with and see their feedback while also thanking and crediting Halfoat for the original work. 

Regardless I will continue to test this with the group I fly with. So, there may be updates or I may take this mod down if it turns out to be detrimental. JSGME ready.

 

The accuracy and responsiveness of Gunners were also updated using concepts from Rieper420's Blind Gunners mod. This was to balance the changes to fighter gunnery.

 

AI gunnery.zip

 

6.002 compatible

 

AI gunnery pilot despawn delay.zip

6.002 compatible

 

 

Reference info

Il2 gunner error.zip

 

turrets.zip

 

Changelog

15 Aug 2025 - AI Gunnery mod - add the two new Yaks.

30 May 2025 - AI Gunnery mod resync to 6.001 stock and add I-153 and Lagg3 Series 4

25 Apr 2025 - AI Gunnery mod resync to 5.507 stock. Include changes from 26 Mar beta as well as in response to feedback further increase in penalty for turret gunners firing under G, further reduction in accuracy for low and normal skilled gunners (most commonly encountered), reduction in G limit for WW1 gunners to 2Gs as well as reduction in accuracy for low and normal skilled WW1 gunners. AI Gunnery pilot despawn delay mod resync to 5.507 stock to include new 1916/17 crew models for German aircraft.

24 Mar 2025 - created reference info section, moved il2 gunner error.zip to it and added turrets.zip. Latter is pdf listing turrets by aircraft, position, weapon, number of guns in the position, rate of fire, number of rounds in a 1 sec burst, ammo count details, number of 1 sec bursts to exhaust ammo supply at the position. Note reloading a magazine from reserve takes about 8 secs approx. 

12 Feb 2025 - AI Gunnery mod. Resync LaGG3 series 29 AI file.

19 Dec 2024 - AI Gunnery mod. Resync unique skins files, add LMG 08/15 (WW1 German aircraft MG) weapon definition to mod and alter the fire rate to 450rpm.

20 Nov 2024 - AI Gunnery mod: Resync changed aircraft AI folder files like Tempest and Ta152, add new WW1 aircraft, add new WW1 aircraft turrets, add new WW1 AI gunners. Fix oversight in RE8 gunner definitions.  Merge the Unique Skins mod into AI Gunnery to avoid a file level clash for the Ju52 going forward. AI Gunnery pilot despawn delay mod: Add Airco DH2 pilot

21 Jun 2024 - AI Gunnery mod: Add Ta-152-H1. AI Gunnery pilot despawn delay mod: Add Ta-152-H1

26 Apr 2024 - AI Gunnery mod:AI Gunnery resync'd to 5.203b. Changes were for IL2/41/42/43 weapon usage. Believe this is the salvo changes discussed in the 5.203b changelog,

24 Feb 2024 - AI Gunnery mod fix small error I didn't see before uploading. Few people who downloaded please do so again. Sorry

24 Feb 2024 - AI Gunnery mod updated to match 5.202, added new aircraft. Added 2nd pilot position for Li2 and CG4A - note that I don't own these aircraft and was done via guesstimation and testing by a friend who does own these aircraft. So they may not be quite as polished as for the aircraft I own.

17 Feb 2024 - AI Gunnery mod updated for WW2 so that the 3x normal range concept is no longer used with gunners. Gunners will now fire out to their weapon range per their skill based situational awareness. This required a complete rebalance of gunner skill-based errors and responsiveness. Some weapons had their burst length updated to better suit their ammo supply. Fighters had their skill based max open fire range altered as a result of the gunner changes to arrive at a reasonably balanced fighter v bomber situation without altering the fighter v fighter situation as best I could. Novices now open fire at 600m, normal 550m, high 500m and ace 480m. Situational awareness values were also altered a little to fit the new environment. Increased the height advantage normal, high and ace fighters will try to gain before attacking bombers. I tried very hard to alter the fighter side of the mod as little as possible. I expect feedback will probably drive further tweaking and refinement. Please note that you will find bombers much better at defense now and you should use real life tactics - beam attacks, attacking with altitude difference, head on attacks, going for the out of position bomber etc. Novice AI tend to get punished quite strongly by their insistence on co-alt 6 o'clock slow overtake attacks. Don't copy them. WW1 side wasn't changed.

13 Jan 2024 - AI Gunnery mod updated with new aircraft, WW2 gunners rebalanced after 5.102 changes, Bf110E2 in stock does not have any restrictions on the gunners aim to represent the tail assembly blocking the field of fire. The Bf110G2 does. Believing that this is an oversight in the E2's case I have copied the G2s restrictions to the E2 as I feel the two versions of the aircraft are very close in terms of the turret layout and weapon mount position and the tail plane construction. WW1 both fighters and gunners rebalanced - this was fairly extensive. I ended up allowing aces to fire from further out at the same range as high and lots of gunner tweaks to get something workable. Fe2b reclassified as a fighter (stock = light bomber). Now includes the 2nd pilot position and credit for the initial concept and work on the A20 navigator position and Mosquito navigator position should go to Murleen. Murleen intended these positions to be able to fire flares and use binoculars, but I haven't been able to get this aspect completely sorted out. Some positions (eg Mossi) are a bit quirky, and you may find that the pilot 3d shape is not always visible. This seems to be due to the 3d shape of the 2nd pilot/nav clipping the pilot 3d shape. I have tried a few of these positions in MP using a friend as a guinea pig and they seem to work ok although again some have quirks. Aircraft covered are Ju52, C47, A20, Mosquito, Ju88c6. 

21 Dec 2023 - AI Gunnery pilot despawn delay updated with new aircraft.

2 Dec 2023 - AI Gunnery mod: Add custom Fe2B front gunner bot. Slightly tweak rear WW1 gunners to improve their accuracy. No WW2 related changes. Note that there is a known issue with the Fe2b gunner in that if the gunner is killed a new gunner will spawn and begin to fire. This is a game bug relating to the two separate guns (and therefore two gunner bots) for the same position and is not caused by this mod. Essentially, you have to kill both bots to silence the Fe2b guns. The bug has been reported to the dev team.

30 Oct 2023 - AI Gunnery mod:  correct what I believe is a typo/bug in the Me410 turret definition. Stock points at BULLET_GER_7.92x57_APsmk.bin for the targeting script. I believe it should reference BULLET_GER_13x64_AP.bin

26 Oct 2023 - AI Gunnery mod: Add Hanriot HD 1, La5F ser38 and resync La5 ser 8.

22 Sept 2023 - AI Gunnery mod: Add Albatross DII, Nieuport 17 and Li2. Move the npc crew of flyables into this mod from AI Gunnery despawn delay for better demarkation in the mods. Remove Ju88 fwd and lower turret definition as 5.106 fixed the bot missing the 1944 uniform. AI Gunnery despawn delay mod: Some bot definitions moved from this mod to AI Gunnery.

27 July 2023 - AI Gunnery mod: add Nieuport 11 and Halberstadt DII. AI Gunnery despawn delay mod: Add 1916 Fokker E3 pilot bot.

30 June 2023 - AI Gunnery mod: resync to base game files. AI Gunnery despawn delay mod: No changes.

27 April 2023 - AI Gunnery mod: add BF109G6/AS and Schuckert D4. AI Gunnery despawn delay mod: No changes. Accidentally removed AI Gunner Info.zip and had to readd it to this post.

7 April 2023 - AI Gunnery mod: resync to game files. Hotfix 5.101b altered some priority type values for WW1 aircraft and the fuel consumption rate for Spitfire XIV and XIVe. AI Gunnery despawn delay mod: Resync to games files. Bug fix to Pe2 bot definition in hotfix.

30 March 2023 - AI Gunnery mod: resync to game files, include Spitfire XIVe, update to cater for Pe2 crew changes. Ju88 fwd and bttm gunner 1944 version still not in stock so I left these in the mod. AI Gunnery despawn delay mod: Resync to game files. update to cater for Pe2 crew changes.

22 December 2022 - AI Gunnery mod: Added Sopwith Snipe, removed Ju88 top turret (31 Oct addition) as it now includes a 1944 gunner model. Oddly fwd and bot gunner still not fixed so left those in the mod. AI Gunnery despawn delay mod: Removed a temp folder I had accidentally left in, but existing users do not need to download again as it was only small and didn't impact anything.

 

28 November 2022 - Added an optional mod that delays despawn on death of AI pilots to 2 mins in line with AI gunners and other crew. It is cosmetic only in that you will no longer see the dead pilot despawn after 15 secs leaving the cockpit empty and instead they will despawn after 2 mins in line with the gunner behaviour. I believe this adds to immersion and is likely a nice addition for people creating videos etc. It has no effect on the actual gunnery of the AI pilots and gunners.

23 November 2022 - Only a small cosmetic change, added C47 crew to have a longer despawn on death delay for better immersion now it is flyable. No other apparent issues post 5.003

31 October 2022 - Quite a few changes made:

• Updated text above to be more in keeping with current state
• On discussion with Avimimus the burst length fired by a gunner is now dependent on the position's total ammo supply. This involves categorising the 60 odd turrets into long, medium, short and brief burst length and then assigning a burst length and time between bursts to each category. This keeps turrets firing longer and tends to help bombers survive longer as previously it was likely all ammo would be expended early in an engagement. Many thanks to Avimimus for the suggestions, info and help.
• The above point changed the dynamic between fighters and bombers. Consequently, gunner error settings and skill were reworked and tested to achieve a what I feel is a reasonable representative balance of WW2 fighter/bomber combat
• During testing it became plain that turrets armed with light caliber 7.62/7.9 mm weapons could almost be ignored and aircraft armed with such could be attacked without any consideration of return fire. Accordingly, turrets armed with light caliber weapons were split out into their own separate burst categories and assigned different error and skill values to raise their threat level to try to keep attackers honest. The intent being to force particularly players to treat the target bomber as a danger. Testing shows that the result was that the lighter armed bombers while not likely to kill you or destroy your aircraft will cause enough damage to force an attacker to break off if they attack using no brainer tactics.
• WW1 fighters and bombers reworked. The previous release just wasn't good enough. Fighter ranges were increased to better facilitate dogfights. Novices are 400m, Normal are 300m, High are 250m and Ace are 200m. This required the bombers to change as IL2GB allows turrets with sufficient ammo to engage at 3x normal range, which meant fighters were engaged and shot down before they could fire. By changing the WW1 bot definitions, I switched off the 3x normal range ability for WW1 bombers forcing them to wait until normal range. Additionally, I modified the error and skill values for WW1 gunners. I believe the results are a better representation of the close-range style of WW1 combat.
• Corrected the Ju88A4 turret definitions so the 1944 crew member will appear. Current stock definitions don't include the 1944 crew, so this is a temporary fix until it gets patched.

12 October 2022 - Review post 5.002 and adjustment to address concerns about frequency of head on attacks

21 September 2022 - Added the Hurricane which I somehow left out of the last version. Added WW1 fighters.

08 September 2022 - Updates due to 5.001 patch. Vulnerable gunners' portion of the mod no longer required as stock life values have been reduced to levels lower than the vulnerable gunner's mod. Due to this I returned the turret life values to stock meaning I could completely retire the vulnerable gunner's mod.

28 August 2022 - Discovered some differences in files particularly around gunner bots. Updated to bring mod into sync with current game version

4 August 2022 - added Me410, corrected files to match current game version. 

17 June 2022 - quick health check, added B26 gunners and also included 2nd pilots in flyable aircraft in an attempt to keep them around a bit longer when they get killed.

1 June 2022 did a quick health check of mod post patch and added Mosquito.

28 May 2022 removed AI dogfight gunnery as I noticed people downloading it instead of AI Gunnery. Fixed thread title.

26 May 2022 added AI Gunnery mod which is intended to cover both fighter and bomber air to air gunnery replacing AI dogfight gunnery

18 May 2022 new version of AI dogfight gunnery. Increased multiplier for head on attacks.

17 May 2022 new version of AI dogfight gunnery. Tweaked ranges, multiplier for head on attacks, angles

13 May 2022 removed 7zip files replaced zip files, no changes to mod content or pdf. Added Excel file behind pdf for those interested.

7 May 2022 added explanation of min & max and stock values, added further explanation of formula values

 

 

 

 

 

 

 

 

Edited August 15 by Stonehouse
see changelog

[Vishnu]

I’m assuming it’s GSME ready?

[Vishnu]

Oddly, I can't seem to download this file.   My screen just flashes, and I have to exit everything and reboot my browser.   Weird.

[Stonehouse]

Yes JSGME ready, sorry forgot to say. Strange......downloading works ok for me. Just verified it.

[Oyster_KAI]

Great, this makes me want to play this game again...

[Vishnu]

Got it working a few days ago.   AI definitely  seems “smarter”   

[l_commando]

From the tests that I've conducted in QMB, Ace-level AI are more accurate and less "trigger happy" but there might be an over-correction somewhere in the values as they seem less aggressive and hardly ever attempt deflection shots

[Stonehouse]

There are values which might be to do with angle off shooting. In Halfoat's thread I did mention them as it seems that angle off shooting becomes less prevalent as you go from novice to ace. Halfoat theorised that:

"However I assumed there was a hidden calculation or process that meant snap shots just don't succeed much at all. Therefore all the greater angle off shots are essentially considered wasteful for ALL skill levels. So ace skill in game comes from better flying parameters narrow angles off rather than shooting skill in a deflection sense."

 

In other words Halfoat's idea was that aces were more likely to wait until they got a good shot with little angle off and get a kill through superior maneuvering (there is definite differences in between the skills in flying ability). Novices would go for impossible crossing shots and miss.

 

If I get some time this week I will try some experiments to see if changes to these values makes a difference to what you've reported without making something silly happen.

 

@l_commando - just want to be sure I understand you........do you mean that the AI do not maneuver aggressively or that you feel they should be shooting and are not?

 

Edited May 16, 2022 by Stonehouse

[l_commando]

23 hours ago, Stonehouse said:

@l_commando - just want to be sure I understand you........do you mean that the AI do not maneuver aggressively or that you feel they should be shooting and are not?

 

 

They are holding their fire even when they're in close and should, at least in theory, have a good angle. This often results in them breaking off without firing a shot or enemy aircraft forcing them to overshoot. Now when they do open fire they hit what they're aiming at and cause a good bit of damage. However, this only seems to happen when an enemy aircraft is flying straight and level.

[Stonehouse]

3 hours ago, l_commando said:

 

They are holding their fire even when they're in close and should, at least in theory, have a good angle. This often results in them breaking off without firing a shot or enemy aircraft forcing them to overshoot. Now when they do open fire they hit what they're aiming at and cause a good bit of damage. However, this only seems to happen when an enemy aircraft is flying straight and level.

 

Just aces yeah? Probably will use tacview to try to get an accurate read on distance and angle off. 

[Stonehouse]

Been looking at what I_commando reported today. I've come to the conclusion that the 4 angle settings relate to the angle to open fire and stop firing at min and max range. So for example for a stock Novice I believe at 50m the min angle to open fire was 14 deg angle off and firing would stop when more than 28 deg angle off. At 400m it was 7 deg and 14 deg. 

 

When I modified the ranges in the first version I didn't change these values but thinking about it I believe I should have. I've assumed that the min fire, min open and max fire and max open are two points on a line. Likewise I have assumed that min fire, min stop and max fire, max stop are two points on another line and that going closer means large angles and moving away means smaller. You can see this in the table below So simple trig gives me the equation for both these lines and I can work out the min open, min stop, max open and max stop at any range. I did have to fudge one as it ended up being a -ve angle for the stop angle. Open must be less than stop for things to work. Obviously my assumption about the points fitting a line could be wrong.

 

Additionally while looking around I came across another parameter dealing with a range multiplier to use for head on attacks. The stock value was 1. I changed this to 1.5 as I had reduced the max fire distance for higher skills and making it 1.5 will push it out some. I also altered the max fire range for High and Ace slightly.

 

The results seem better and while I haven't had much time to test things as the editing and checking for typos took up a fair bit of time. I used P51Ds v's FW190A6s in a 2v2 quick mission all aces. 50% fuel. Generally I started seeing deflection shots at times again and a more shooting at very close range. 

 

Updated zip in the first post. See how you go.

 

	Minfire	MinOpen	MinStop	Deg diff	Maxfire	MaxOpen	MaxStop	Deg diff	Comment
Novice	50	14	28	14	400	7	14	7	Stock
Novice	100	13	26	13	800	1	2	1	Mod
Normal	50	11	19.2	8.2	500	5	8.5	3.5	Stock
Normal	70	10.7	18.57	7.87	600	3.67	5.512	1.842	Mod
High	50	8.5	12.7	4.2	600	3	4.5	1.5	Stock
High	50	8.5	12.7	4.2	500	4	5.99	1.99	Mod
Ace	50	5.5	6.8	1.3	800	1.5	1.8	0.3	Stock
Ace	30	5.61	6.93	1.32	400	4.17	5.13	0.96	Mod

Edited May 17, 2022 by Stonehouse

[l_commando]

I should have stated earlier that I only tested Ace level AI because that's what I have it set to in QMB by default and I did not get the chance to really mess around with the other levels. I assumed that the same issues would've been present there as well. After running some tests of the new version in 4v4 Ace-level match ups, there has been some definite improvement in deflection shooting. Enemies and friendlies both do it more often and appear to be more accurate with it. They also seem to fire in shorter bursts instead of burning through ammunition not hitting anything, although that did happen once.

 

Still, I would like to see AI attempt more deflection shooting since so many dogfights in the game devolve into turning fights. Ideally, I'd like to see higher-level AI attempt quick "snap shots" when an enemy passes in front of them, but I do not know to what extent that is possible. On the topic of head-on attacks, I have noticed the AI making more head-on attacks than usual but not firing often, or doing so when they are almost past each other.

[Stonehouse]

I can muck about some more with the angle values to try to improve attempts at crossing shots but from what I saw when trying extreme values to try to figure out what they did, it can cause stupid things to happen. eg Ace AI in a lag pursuit turning fight firing constantly (and I mean emptying their ammo) behind the target when obviously there is no chance to hit. It's possible that there is nothing accessible to tweak this sort of thing but it's worth a try. On the head on side, perhaps I need to increase the multiplier value further. I know with the AI gunners there is a value where the AI waits x secs (like 0.5 or something) as if they are firming their aim. Possibly there is something like that for pilots. If so then moving the head on multiplier further out will help high level pilots. You may unfortunately see occasional novices firing at 1600m plus as a side effect though fyi.

Edited May 18, 2022 by Stonehouse

[Stonehouse]

Apologies to people who have just downloaded yesterdays version but I did increase the head on attack multiplier and it did make a definite difference. The AI pilot must need a certain amount of time to make the attack and so pushing out the range a little more helps. Aces now have 800m to decide to attack which is what they had in the stock game. This may lead to normal AI and high being more deadly than expected in head on attacks but I think this is a compromise we'll have to live with. The definition in the game of "head on" might also be looser than I first thought as with the increase I did see some crossing shots being made on the front quarter of the target. Hopefully I might be able to take the word "draft" out of the thread title soon.

 

Ace has 800m to decide to make a head on attack

High has 1000m 

Normal has 1200m

Novice has 1600m

 

I'm assuming for now that gunnery error logic (which I haven't found for fighter pilots) will protect us from Novices but Normal and High might be something to watch out for. Ace is the same range for head on attacks as stock with the new multiplier value. 

Edited May 18, 2022 by Stonehouse

[Skycat1969]

There are two zips with the same name "AI dogfight gunnery." I've downloaded both and I can't see where either is more recent than what I downloaded this afternoon?

 

Edit: I figured it out. The bottom one has a more recent cplaneai.txt (5/17/2022 at 8:58 PM).

Edited May 18, 2022 by Skycat1969

[Stonehouse]

57 minutes ago, Skycat1969 said:

There are two zips with the same name "AI dogfight gunnery." I've downloaded both and I can't see where either is more recent than what I downloaded this afternoon?

 

Edit: I figured it out. The bottom one has a more recent cplaneai.txt (5/17/2022 at 8:58 PM).

Yes sorry. Was multitasking (which my wife tells me I can't do, turns out she's right it seems) and managed to delete the pdf instead of the mod zip file. All fixed now and you got the right one despite my goof. ?

Edited May 18, 2022 by Stonehouse

[Oyster_KAI]

Has anyone tested in career and looked at longer term data?

[Stonehouse]

Not career as such but the group I fly with use PWCG for our co-op campaigns and we started using a version back in Feb. As it didn't seem to be causing huge issues I felt encouraged to put up the first draft version at the end of April. Don't have stats though, just a general observation that it didn't cause huge issues like the AI force pool being depleted. Getting bounced by Aces in PWCG with this mod is a "WTH?" type moment. If you don't see them coming you tend to only know about it when the bullets arrive. Also not sure if the mod is responsible but with the last two versions I've been seeing P51s doing a lot more vertical fighting than before in the ace level QMB test missions.

 

It would be interesting hear feedback if someone wanted to go through a SP career with it.  

[Skycat1969]

I'm using it in career mode. I've flown 2 or 3 Kuban missions in Bf 109 G-4 and if I've noticed any affect by the mod, it could be that the enemy AI are less aggressive. That could be caused by official changes to the AI though. Or it's just my perception. I'll see if the newest changes seem to make a noticeable difference.

[Stonehouse]

Interesting. I don't see any drop in aggression in QMB with all pilots set to the same skill on both sides. ie 2 ace v's 2 ace etc. I use a head on engagement at co alt starting 2000m apart for testing.

 

Skill level is linked to search ability for air gunners and also attack and engage distance for fighter pilots. eg for a novice engage distance is 1500, for an Ace engage distance is 3000m

 

Could you be encountering mainly novice AI enemy pilots?

 

Are the AI getting "command lock" do you think? There have been instances in the bug thread reported where wingman just flew blindly about after being ordered to engage. I know this for sure as I posted a bug report last year about it. Even caught it on tacview as well as the track. I have seen others report the same thing.

 

I think I saw a recent discussion about waypoint configuration too - on some settings AI don't react to things I understand. Hard to tell this though unless you can edit the mission.

 

(edit) - guess also - in case terminology is getting in the way - what do you mean by less aggressive? Enemy is not maneuvering? Not shooting when in a position to do so? Not reacting when getting hit?

Edited May 18, 2022 by Stonehouse

[Mtnbiker1998]

Haven't been too inclined to mess with my game lately (doing a lot of coop) but this looks like a really excellent project!! Super well documented, and AI gunnery could certainly use some work. Maybe if the mod matures enough you can post over in the suggestions thread and get this into Vanilla! Wouldn't be the first time it's happened, though those mods were a bit different in scope..

[l_commando]

9 hours ago, Stonehouse said:

(edit) - guess also - in case terminology is getting in the way - what do you mean by less aggressive? Enemy is not maneuvering? Not shooting when in a position to do so? Not reacting when getting hit?

From the tests I conducted yesterday (I haven't tried out the latest version you just posted yet), I did notice some of the "command lock" you mentioned but really only when my flight was in pursuit position. My flight didn't seem to be in much of a hurry to engage the enemy then, but thankfully head to head engagements were a different story.

 

As for the other potential issues you mentioned:

 

- Both my flight and the enemy's seemed to maneuver as normal (command lock aside).

- While significant improvements were made in shooting and ammo conservation, as I stated, they often did not shoot when they had an opportunity to do so. That said, however, it's possible that they were out of ammunition or were suffering from G-effects. Or I could be biased in that I can often be much more aggressive than my wingmen (and a better shot most days)

- Situational awareness also seemed normal or, if anything, slightly improved. But there were instances of aircraft staying on the tail of an enemy while getting shot at.

[Stonehouse]

I see some odd things out of the commands while testing this. I need to repeat some of the tests as stock to compare.

 

I wonder at times how "attack nearest enemy" differs from "patrol for air targets" when the enemy aircraft are right in the AI's face. To me attack nearest enemy often ends up with 2 aircraft ganging up on one and even sometimes friendly fire due to shooting over the shoulder of a friendly. Patrol or patrol for air targets ( I switch off any ground units while testing) seems to allocate friendlies better so it is 1 aircraft per enemy.

 

I see what I'm going to call "tiredness" creep in. I tried a 6 v 6 ace test using the usual head on, co alt, 2000m engagement and me as leader and 7th friendly aircraft sitting off to the side to observe. Due to more care with firing and "skillful" maneuvering the fight goes on for a long time.  I have seen larger fights go on over 15 mins. Somewhere around the 10 or more min mark some of the AI pilots seem to lose interest and start flying off. I don't think it is they are out of ammo. Possibly it is G effects. Anyway I give orders to send them back to the fight and you will see just a little wing waggle and they ignore you. This is what I'm meaning by command lock (although the bug I reported last year they just ignored me from the start).

 

I will then try a series of different attack commands - sometimes this works sometimes it doesn't. Sometimes I find that telling them to cover me will get them to turn to my general direction and then usually an attack order puts them back into the fight.

 

They seem to ignore me more the further away I am so perhaps a command radius thing?

 

I've started moving not so far away and using turn autopilot to keep me in the area before using external camera to check things. That seems to help. I feel that part of the AI definitely needs revisiting when the devs have time.  

 

Bug post for reference 

 

Edited May 18, 2022 by Stonehouse

[Skycat1969]

As I said I don't know what is causing what, since the default AI was also recently tweaked. I play in VR and I also perceive time slows down a bit as soon as the battle is joined -- that could be CPU related. I'm not getting stuttering or judders, just a languid feeling and a decrease in the engine pitch audio. So what I'm experiencing might be caused by the new clouds or other technical factors causing the game to throttle AI computations. It did seem to me though that the enemies who interdicted last night didn't put up much of a real fight, they just swarmed around and got massacred. One of my guys received some damage. None of the AI in my schwarm were credited with kills. So it seems that besides the four kills I was credited with it was the 'other' Luftwaffe flights in area who did all the shooting.

 

Sorry that wasn't a very scientific analysis. I'll fly a mission tonight and watch for a behavior change with the new version of your mod.

[Stonehouse]

I think the slow down is because there are too many objects in play. I see something similar at times in my offline PWCG campaigns when there is a lot of action happening, it's like time is running at 90% normal speed. I don't use VR.

[Skycat1969]

The current version of the mod seems to work pretty well. I've had some decent (but not overly threatening) fights in the Kuban career. Flying the P-47 career in Bodenplatte though I got shot down by a Fw 190A who got behind me; then I replayed the mission and I was damaged by a Fw 190A I was chasing when he did an Immelman and reversed onto me.

Edited May 21, 2022 by Skycat1969

[pepper9881]

In case of fighters vs fighters for most aces the strategy of opening fire from a minimum distance is fair.
But is it so fair for attacking a formation of bombers? Especially when attacking on a catch-up course. In this case, the real ace will most likely also try to open fire from a greater distance avoiding entering the zone of effective fire of the gunners. Don't you think that with a maximum opening distance of 400m when attacking a formation of bombers, aces-bots will be less effective and will suffer greater damage than high or normal?

[Stonehouse]

No disagreeing per se, although by 1944 Luftwaffe tactics were generally to do head on attacks if they could rather than attacking from the rear of a bomber box as the AI do as doing so was very likely to get you shot down. There is also convergence to consider. If you fire outside your convergence range then likely you will fail to concentrate fire to a lethal extent so I'm not so sure that real life aces would fire from further away to try to avoid defensive fire. I kind of think they would try to use superior maneuverability and speed to get them to their desired range without getting disabled.

 

Anyway unfortunately for this mod it's a compromise. With what we have you can't do both as far as I can tell. Dev's could probably but all we can do is adjust parameters that are available. The same concern did occur to me as well. I haven't had time to do a lot of tests of a flight of fighters against a flight of bombers but the few times I did seemed to result in some casualties for the fighters and significant losses for the bombers so it doesn't seem to be a large issue. My few tests have been allied fighters attacking Luftwaffe He111s.  Even if the other way around and it was Luftwaffe fighters against Allied bombers I think it would be much the same as none of the bombers in the game are heavy bombers  nor do the AI bombers fly in correct defensive formations.

 

I will carry out more tests though to try to confirm it is not a problem. Mind you - there is the Blind Gunners mod so you can tweak the other side of the equation and in fact that is probably a better approach. Come up with a version of the Blind Gunners mod that is compatible with this mod. 

Edited May 21, 2022 by Stonehouse

[Vishnu]

This mod is awesome! Just like all your others.  Thanks Stonehouse!

Edited May 21, 2022 by Vishnu

[Stonehouse]

Thanks Vishnu.

 

A few initial tests of this mod against bombers using the version of Blind Gunners from the tail end of rieper_420's thread trying to address concerns about fighter losses due to this mod shows I possibly have the opposite problem. Admittedly an extreme case but 6 Ace Tempests v's 6 Ace He111s = a few slightly dinged up Tempests but nothing that couldn't patched and 6 exploded, burnt and otherwise destroyed He111s.

 

I'll have to do a lot more tests particularly with less well armed fighters but for now gunners really are blind lol.

[Stonehouse]

Done a few more test runs today of ace fighters v's ace bombers. It seems the B25 is the possible fly in the ointment. I assume it must be the better turrets and firing arcs but casualties in the fighters is much higher when attacking a flight of B25s. That's even when I set up Tempests against them just to compare to the test run against the He111s.

Possibly that is ok, not sure.

 

I think the right way to go is possibly a custom Blind gunners set up to suit this dogfight gunnery mod and in the end present it as a fighters and bombers gunnery mod. Going to take a bit of fiddling to get it a draft version up and rolling.

[pepper9881]

Very interesting! Have you tried "high" or "normal" fighters for comparison?

[Stonehouse]

Bit of an update, it's proving to be tricky to get the ai gunners compatible with these dogfight changes. Working on it and think eventually it will get there but definitely going to take some time as it has to fit in with real life stuff.

Some interesting pics I came across which I believe have been posted at places in the forums previously. Unfortunately it doesn't really indicate crew training and experience but you'd kind of assume that the 8th Air Force would base average results off experienced crews not green troops. For reference 600 yds is about 548m. The most accurate position was ball gunner and 15 ft spread is about 4.6m or about half the wing span of an average fighter worth of error. Note these are ground tests so it should be reasonable to assume that in the air it would be worse shooting at a maneuvering attacker even if the bomber was stable itself. I assume B25 gunners would be roughly the same standard as B17 or B24 crews.

 

From what I have read so far Luftwaffe bomber defensive arcs and number of guns were generally not as good as Allied equivalents. By 1944 Luftwaffe gunners were probably either new guys or old sweats who had somehow survived to that point I'm guessing. They didn't rotate crews in a tour fashion as far as I know and they just served until dead or disabled or captured. I know I can separate gunnery by nation or even aircraft but don't know if I want or need to at present as it means a lot more testing. Hoping that the guns and the arcs will take care of it.

 

The other general point I see is that largely bomber guns were meant to stop the bomber from being shot down rather than to shoot down fighters. They did this by use of defensive formations generating so much cross fire that attackers had to maneuver too much to get off good shots or it was perceived as too dangerous and so the attackers stayed too far away for their fire to be effective. So in my testing I am going to count a damaged fighter that stops attacking as equal to a kill from the bomber gunnery part of the mod's viewpoint in regard to tweaking gunners effectiveness.

 

The biggest issue really is that experienced pilots attacked differently than novices in real life and this is not reflected in the AI in game which pretty much always does a stern attack and often singly. Plus I guess generally AI bombers don't hold a real formation and really a bomber that wasn't in a defensive formation was most likely dead if a group of fighters found it. 

 

So whatever I get to in the end will be definitely lots of compromises. Hopefully from a playability viewpoint it is acceptable and an improvement.

 

 

 

Edited May 24, 2022 by Stonehouse

[Stonehouse]

So here is the first beta version of the AI gunnery mod. It combines the AI dogfight gunnery mod, a customised version of the Blind Gunners mod and a customised version of the Vulnerable Gunners mod (thanks and credit to Rieper_420 for these two mods). As I said previously it's a collection of compromises.

 

Some basic comments about this mod to set expectations;

 

The dogfight aspect of the mod is as per the last version of the AI dogfight gunnery mod (now superseded by AI Gunnery). 

 

Generally, bombers as we see them in IL2 GB will lose a fight with an equal number of fighters. Skill level of the bomber crews and fighter pilots do influence this but still the expectation is that bombers will get shot down and it’s just how many fighters are disabled or destroyed in the process. This is unescorted bombers. Why take this attitude?

• The bomber AI do not hold a defensive formation and there are not enough bombers usually to stop an attack anyway due to limits on bomber numbers.
• To offset this, I believe the stock game code allows the bomber’s gunners fire out to 3x their normal engagement range as long as their position’s ammo supply exceeds a given value. I experimented with turning off this ability by setting the required ammo very high and the bombers were decimated because they did not fire until the attacker was already firing. Even when the fighters were aces with an open fire range of 400m it tended to work out this way. It was worse with veteran fighters or cannon armed novices. Particularly this was evident with bombers with limited firing arcs as the poor formations used did not allow any cohesive defense. As a result, I have left this feature at stock values. Additionally, aircraft with limited ammo for the gunner were penalised unrealistically if it was changed.
• Because of the way bombers throw themselves around the effects of G loading on the gunners are set to be of limited impact. The stock cut off for being able to fire is 5G as a result. I actually have been in a BAC Strikemaster for a mock dogfight and experienced around 4Gs without a G suit. I don’t believe many people would be able to manage and effectively aim a heavy machine gun in an unpowered mount at more than two or three Gs especially if the aircraft they are on is changing direction a lot. But because the bombers don’t fly as they should I think the impact of G loading has been minimised. I’ve ended up leaving this at the stock value.
• Fighters have only one attack plan for aircraft. So I can’t have a different range to start firing if the target is a bomber with defensive guns to the range for a target which is a fighter or a transport. This single fact means that you have to fudge the way bomber defensive guns and AI skills work to avoid unrealistic situations. E.g, in the stock game novices attack a bomber at 400m and an ace attacks a bomber at 800m. In the stock game novices attack a fighter at 400m and an ace attacks a fighter at 800m. From every biography or reference I’ve ever read this is the opposite of real life for fighter v’s fighter and also fighter v’s bomber. Novices generally tended to misjudge range and fire too far away. Aces fly better and sneakier and will choose the range to best get a result.

 

Added the download to first post. Constructive feedback more than welcome but it is going to be trade offs all across the board so people may need to accept that not all situations can be catered for. If the average is an improvement then I figure it is a success. 

 

Pretty obviously don't use Blind Gunners or Vulnerable Gunners and AI gunnery as AI gunnery combines all 3.

Edited May 28, 2022 by Stonehouse

[Stonehouse]

I noticed there was confusion about whether people should download the AI Dogfight Gunnery mod or the AI Gunnery mod. As AI Gunnery includes AI dogfight gunnery but adds the tweak for bombers gunners to match the changes in fighter gunnery and therefore provides a better solution I decided it was best to remove the AI dogfight gunnery mod from the first post.

 

If you previously downloaded the AI dogfight gunnery mod you really should download the AI gunnery mod now and use this instead.

[l_commando]

I finally got around to playing the latest version of the mod and in the QM's I've tested so far, all seems fairly well. The gunnery changes are definitely a welcome addition and add to bomber intercepts. However, I would suggest looking into and possibly tweaking the P-38 values specifically as, for whatever reason, veteran and ace-level AI in that aircraft seem particularly unaggressive and hesitant to shoot, and fairly inaccurate when they do shoot. All other aircraft I've flown with and against so far (mainly German and Soviet) seem to be alright. 

[Stonehouse]

1 hour ago, l_commando said:

I finally got around to playing the latest version of the mod and in the QM's I've tested so far, all seems fairly well. The gunnery changes are definitely a welcome addition and add to bomber intercepts. However, I would suggest looking into and possibly tweaking the P-38 values specifically as, for whatever reason, veteran and ace-level AI in that aircraft seem particularly unaggressive and hesitant to shoot, and fairly inaccurate when they do shoot. All other aircraft I've flown with and against so far (mainly German and Soviet) seem to be alright. 

 

Ok not sure how much I can alter unaggressive behaviour but will look into it. 

 

Did a quick health check post patch and didn't see any issues. Added Mosquito and uploaded the new version to first post

[Stonehouse]

Did some tests with the P38. I'm not sure if it is just that it's a bigger aircraft than say a Spitfire or 109 and so less nimble or if it is something with the AI and how it handles dogfighting. They seem to generally get stuck in but I see things like they don't want to continue dives too far.

 

I have not so far seen the AI pop air brakes in a dive in a dogfight, doesn't mean they can't just I haven't seen it at a time I think using it would allow a dive to continue to prosecute an attack and instead I see the AI usually rollout of the dive and gain height instead allowing the enemy to escape. It's possible that if it is true that the AI doesn't use the dive recovery airbrakes that the P38J25 is limited to the combat moves of the earlier less successful models and this is what you are reporting.

 

Anyway will do some more tests and do some more looking at what the AI does 

Edited June 3, 2022 by Stonehouse

[RedeyeStorm]

I would not be surpris3d if they don’t use them. For example the AI doesn’t feather props in twin engined aircraft that have the ability to do so.

[Stonehouse]

Well I've done some more testing on the P38. They definitely do engage seemingly with as much intention as other aircraft but they are less nimble so aircraft like 109s seem to out maneuver them in turning fights and in the vertical the P38 pilots seem to break off dives (I've still not seen them use their dive brakes and commit to a full diving attack) and so enemy aircraft run out from under them while the P38 dives half heartedly rolls out and climbs in a turn usually. They don't even go into a full dive using the brakes to save themselves when defensive. I would have thought that diving with two engines they would outrun a lot of things.

1v1 usually ends up in a turning fight that goes on for a very long time and often the enemy will nibble away the P38 tail plane controls and that ends up with the P38 getting shot down. It goes a bit better with higher numbers and 4v4 higher the P38s get kills but often end up on the losing side. It's not conclusive but I think what I_commando reports is the result of the AI not fully using the aircraft's abilities.

 

So P38s (in current AI hands) probably should be regarded more like a Mosquito - a fighter bomber - good fast strike ability and range, ok against bombers, average at air to air and outclassed by a pure fighter aircraft from the game viewpoint. If the devs ever get time to rewrite the AI then I would think the P38 would need special handling so the AI pilots fully use the aircraft's abilities.

 

Just a PS - there is nothing I've seen in the files that I can tweak to resolve this. I looked at the Ju87 parameters and I couldn't see anything dive brake/dive recovery system specific. So I suspect that it is part of the dive bombing logic. I wouldn't be surprised if P38s used their brakes when dive bombing but it isn't part of air to air combat so P38s don't use them there. Ju87 logic predated P38s by a long way - part of Stalingrad if I remember right - so the Ju87 dive brake logic when dive bombing would have been baked in from the start. P38s are from Bodenplatte so strapped on with minimal AI changes for their specific use case is my thinking.

Edited June 4, 2022 by Stonehouse