Gunner Performance Measured

[Charon]

I've seen a lot of opinion-based discussion of gunners, but not enough data quantifying their performance. So let's fix that.

 

Test Procedure, Stage 1:

 

Two identical stock Halberstadt CL.IIs. One is 'player', the other is 'Ace'. Both follow a high-priority waypoint at the same speed and parallel courses until the waypoint is reached, then simultaneously switch to a medium priority waypoint. A message appears on screen at this time, and the gunners engage each other.

 

I start this mission by immediately turning on AI autopilot (full autopilot, not the auto-level one). I then switch to the gunner's seat. I do not take control of the gun until prompted to engage. After doing so, I wait for the AI to fire at least two bursts at me (to give him a sporting chance). I then try to shoot him down.

 

During the test, I discovered that sometimes one plane would wind up slightly ahead of the other (nondeterminism in mission start?) Waiting for the AI gunner to shoot first also gave him a chance to get back in position if he drifted too far back (because shooting over the wing at 10 o'clock is harder than shooting backwards at 8 o'clock).

 

Realism is full except that autopilot is enabled, technochat is enabled, warmed-up engine is enabled, and spectators are enabled (for damage assessment).

 

Outcome:

 

100m: I ran 5 trials at 100m. In 5 trials, I shot down the enemy Halberstadt 5 times, four via pilot kills and once by fire. In the first trial the AI managed to land a few hits on my right wing. In the four other trials I was completely undamaged.

 

200m: 5 more trials. I shot down the enemy three times by killing the pilot, twice by setting him on fire, and once by both setting him on fire and killing the pilot. I was completely undamaged in all five trials.

 

400m: 5 more trials. I shot him down twice via pilot kill and once via fire. Two other trials were draws when we outran my allotted test course, although in one of these I did at least manage to set his engine smoking. I received superficial damage in one trial. 

 

Media:

 

This video shows two representative trials, at 100m and 400m. Note how even at 400m, even my shots that miss are passing within a few meters of the target, while his aim is wildly off.
 

Spoiler

 

 

 

 

 

 

Here are two cropped screenshots showing the AI's accuracy (in false color, to make the tracers more visible). In the first shot, he's off by 6.3 degrees. In the second shot (at 400m) he's missing by a remarkable 13.5 degrees.

 

 

In this shot, his fire can be seen to miss by 100m. These planes are traveling at 150kph (==40m/s), and with 400m between them it takes half a second for the bullet to cross this distance. Thus the bullet is passing 80m behind the point where I was when he fired it.

 

Not only is his lead too large, but he's leading in the entirely wrong direction!

 

Tracks:

 

Attached is a zip file containing both tracks used above, as well as all three variants of the mission, if you'd like to give this a try yourself.

gunner-testing.zip

Edited November 26, 2023 by Charon

[Stonehouse]

Had a break from digging retaining wall footings in the back yard and I tried your 100m mission using no mods, but I stayed in the pilot seat and let the gunners do their thing. Best I could see both in my track and tac view essentially both gunners missed despite several hundred rounds fired by both - I did notice that in my case the enemy Halberstadt drifted ahead after about 1-2 mins even though I was on full auto pilot. Should that have happened with your priority settings? Neither aircraft maneuvered at all otherwise as expected.

 

Repeated it with AI Gunnery, definitely got hits and hit this time but less than expected considering the range and skill. Enemy again drifted forward. As I mentioned in the mod discussion thread where I have a beta version up for testing, I did wonder if I needed to tweak the WW1 gunners on two seaters at least. Seems like this confirms I do. Very useful set of test missions - thank you !!.

 

FYI range is a final multiplier for working out gunner error according to the comments in the game files. So especially for WW1 the 100m test is quite sufficient for your test mission as you would expect twice and 4 times as much error at the other ranges. ie if you aren't hitting at 100m you definitely will not at 200 or 400m.

 

I saw a post by LukeFF that possibly there is a bug in that the AI level for the gunner in a player Fe2b was not Ace as it was meant to be. Maybe it is more widespread than that. So perhaps I will wait until that is confirmed before doing much more tweaking. Good thing really as I don't have much spare time at present.

 

Edited November 26, 2023 by Stonehouse

[Charon]

Thanks for testing it @Stonehouse

 

2 hours ago, Stonehouse said:

I did notice that in my case the enemy Halberstadt drifted ahead after about 1-2 mins even though I was on full auto pilot. Should that have happened with your priority settings?

 

Ideally no, but I'm not sure how to get rid of it. There's a small offset that puts the AI slightly ahead; this is because otherwise they seem to freak out and maneuver, throwing the whole thing off.

 

I would be careful with using these for tuning, though. This is of course a very artificial setup.

 

What I really want to see with gunners is that player decisions should matter. My general feeling right now is that gunner performance is too 'flat'; if you present gunners with a hard problem they sometimes get hits, and if you present gunners with an easy problem they sometimes get hits, how you behave doesn't seem to matter a whole lot to how they do.

 

This is an extreme example that only illustrates that gunners wildly under-perform humans on the easiest of shots, but if you tune your settings for this test you may find that they out-perform humans when maneuvering or against rapidly moving targets.

 

2 hours ago, Stonehouse said:

FYI range is a final multiplier for working out gunner error according to the comments in the game files. So especially for WW1 the 100m test is quite sufficient for your test mission as you would expect twice and 4 times as much error at the other ranges. ie if you aren't hitting at 100m you definitely will not at 200 or 400m.

 

Interesting. Is this angular error or absolute error?

 

If the former, it seems like a bug and would explain the wild shots I often see. If they're 1 degree off at 100m they shouldn't be 4 degrees off at 400m.

 

Actually in practice I often find that range is inversely related to angular error. The only really wild shots I take are at 25-50m range when I'm presented with a low-probability snap-shot. At 400m my angular error is much smaller (but my rate of fire goes way down) because I don't bother shooting unless I'm presented with a straightforward shot.

Edited November 26, 2023 by Charon

[Charon]

3 hours ago, Stonehouse said:

Should that have happened with your priority settings? Neither aircraft maneuvered at all otherwise as expected.

 

I'm attaching revised versions of the test mission that ameliorate this flaw; also included are 25m and 50m versions.

 

Even at 50m, the stock Ace gunner still manages to completely miss my plane often; at 25m he's able to hit (but not generally to deliver the fatal blow before I kill both him and his pilot).

gunnerduels.zip

[Stonehouse]

3 hours ago, Charon said:

Interesting. Is this angular error or absolute error?

Could be entirely wrong but I think a perfect shot is calculated, dispersion for the weapon is somehow applied and then an error is calculated for each of the x,y and z dimension. The error is then used to determine where the shot actually goes. So like an offset in 3 dimensions to the dispersed perfect shot location in 3 dimensions.

 

Note according to the comments in the turretcontrollerai.txt (this is the one and only bot file for air gunners) there is a random value in the formula. This random value is between -1 and 1. 

 

Formula mentioned in the comments is:

 

// formula for calculating the aiming error zone

// calculation formula
// axis_error (meters) = tan(
// random_number * (AddCoef + ProjSpeedCoef * speed_difference_in_area_of_aim (in meters per second) )
// * (current_overload * OverloadCoef + CoefLow/Med/Hi/Ace)
// )
// * distance_to_target;
// random_number is a number between -1.0 and 1.0
// tan is the tangent. depending on the tangent of the error angle and the distance to the target, the resulting zone is obtained
 

Stock values:

AddCoef (default aiming error factor) is 0.02,

ProjSpeedCoef (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) is 0.05, 

OverloadCoeff (coefficient for converting G overload into aiming units) is 0.05

current overload is I think the difference between 1G and the G the gunner is experiencing. ie 1 if the gunner is experiencing 2G

CoefLow is 10, CoefMed is 4, CoefHi is 2, CoeffAce is 1

 

I believe speed difference in area of aim is the vector component for the axis in question of the difference in gunner aircraft vector and target aircraft vector. In your test case until the AI starts to move forward the difference would be about 0 m/s I feel.

 

Not sure if tan is calculated in degrees or radians. I tend to use this as a qualitive value rather than an actual one for comparing one set of values to another. It makes quite a difference to the error value. eg if the value of all the bits in the brackets is 0.2 then at the max limits of the random range of -1 and 1 at 100m degrees gives you +/- 0.349m radians give you +/- 20.271m in your 100m test case. Note than tan 0 (deg or rad) gives you zero. The two largest variable factors in generating error are the random value and the speed difference. 

 

No idea how dispersion factors in so I tend to ignore it and do testing across values using the same aircraft so it is not a material factor in my qualitive comparisons.

 

Also note that comments are in Russian so the translations are me using google, deepl yandex etc and therefore I could easily misinterpret things.

Edited November 26, 2023 by Stonehouse
clarify

[Charon]

Oh! In that case I think we have an actual bug! Not just that the coefficients are too large, but the formula here seems incorrect!

 

Consider: tan() has singularities at ±π/2. What happens when it's argument passes that singularity? Well, the result starts returning to 0 again. Past a certain point, the average error saturates (and in fact starts improving very slightly as G and relative speed increase).

 

How quickly does this occur? Let's restrict ourselves to the Ace gunner in level flight: current_overload=0 and CoefAce=1. Then the argument reduced to r ( 0.02 + 0.05 * v). Solving for 0.02 + 0.05 * v >= π/2, we find that for a projected relative velocity above 31m/s (111km/h), the ace AI has a chance to reach this singularity. At that point the maximum possible error is infinite, and they start trying to paint the sky. (For 'low' AI, this occurs at just 3.1m/s !).

 

I think you're wrong about this error being computed for three axes, by the way, for the simple reason that when the argument to tan() approaches π/2, the error box would grow infinite on all dimensions. But I've never seen a gunner actually shooting on the opposite heading of the target, so I think this error-box is probably computed on two dimensions, in the plane normal to the target vector (and indeed, `dist * tan(...)` only makes sense if this is so).

 

(This also raises a wild guess about the accuracy of the Me 410's gunners... Is the error being computed as usual, but then projected into a plane normal to the barbette axis? This would cause that large error box to collapse into a line, and might explain its unique behavior).

 

----

 

What we really care about is angular error, so let's compute that. The distance factors cancel out and we get:

 

Ae(r_x, r_y) := angular_error = arctan(sqrt(tan(r_x * (AC  + PSC * v))^2 + tan(r_y*(AC + PSC * v))^2 ))

 

as a function of two random numbers (one on each axis).

The average magnitude of this error is given by

 

1/4 * \int_-1^1 \int_-1^1 Ae(r_x, r_y) dr_x dr_y

 

We can integrate this numerically:

 

import numpy as np
import matplotlib.pyplot as plt

import matplotlib as mpl
import math


N = 200
M = 500

SKILL=1 # 1=ace
AC=0.02
PSC=0.05

VMAX=100

rs = np.linspace(-1,1, num=M)
rxs=np.tile(rs, (M, 1))
rys=np.tile(rs, (M, 1)).transpose()

# arctan(sqrt(tan(r_x * (AC  + PSC * v))^2 + tan(r_y*(AC + PSC * v))^2 ))

vs = np.linspace(0,VMAX, num=N)
m_aerrs = [np.trapz(np.trapz(360/(2 * np.pi) * np.arctan(np.sqrt(np.tan(rxs * (AC + PSC * v) * SKILL) ** 2 + np.tan(rys * (AC + PSC * v) * SKILL) ** 2)), rs, axis=1), rs) / 4 for v in vs]

plt.plot(vs, m_aerrs)
plt.axis([0, VMAX,0,90])
plt.show()

 

This graph shows the average angular error, as a function of speed_difference, assuming SKILL=1 and g_overload=0:

 

x axis: speed_difference_in_area_of_aim (m/s)

y_axis: average angular error (degrees)

 

 

(Although I suspect in practice the way it works is that the game chooses *two* points and draws a line between them, which has the effect of reducing the average error somewhat compared to what this shows.)

 

This saturates quickly even for ace gunners and anything beyond 20m/s is very wild; for SKILL=10 anything beyond 2m/s is equally wild.

 

----

 

What's really out of place here is the speed_difference_in_area_of_aim argument. The comment says this is in m/s. Tangent of a distance, rather than a unitless number? That sounds quite strange.

 

Envision this... a plane off at 500m crossing your field of view at 30m/s (a pretty good clip). But at that distance, it's only moving across your vision at 3.4 degrees per second... It's angular movement isn't that fast, and anyone ought to be able to put shots within maybe 5 degrees of it just by leading a little.

 

But the game's computation doesn't appear to base angular error on angular speed, it appears to use absolute speed, and so the mean angular error for such a shot winds up being about 55 degrees, even though the target isn't tracking across view so quickly!

 

What should probably be computed instead is this:
 

 // formula for calculating the aiming error zone

// calculation formula
// axis_error (meters) = tan(
// random_number * (AddCoef + ProjSpeedCoef * speed_difference_in_area_of_aim (in meters per second / distance_to_target) )
// * (current_overload * OverloadCoef + CoefLow/Med/Hi/Ace)
// )
// * distance_to_target;
// random_number is a number between -1.0 and 1.0
// tan is the tangent. depending on the tangent of the error angle and the distance to the target, the resulting zone is obtained

 

This makes angular error proportional to angular speed, so a plane that flies rapidly across the gunner's field of view will have a large angular error (a snap shot), while one that traverses slowly at long range will have a small angular error (but still be hard to hit, because that long range implies a lot of empty sky).

 

The coefficients would all need to be tuned, of course, otherwise they'd all just turn into snipers, but it would prevent the current overreaction to moving targets.

 

@LukeFF, do you mind sharing this with your engineers?

 

----

 

The other thing that seems suspicious here is CoeffOverload. Am I reading that right? Does that an ace gunner need to experience 20G before he degrades to the level of a 'high' gunner? That doesn't seem right!

 

Edited November 27, 2023 by Charon

[Stonehouse]

17 hours ago, Stonehouse said:

axis_error (meters)

This line in the comments giving the formula is what made me think it was computed for each axis, plus would you not need a 3-dimensional reference to the aim point x,y,z co-ordinates to place the shot? 

The formula gives a result in meters - I would have thought this isn't enough to give the co-ords for the point the projectile passes through. If the aimpoint is say the x,y,z co-ords of the centre of the aircraft then working out an x,y,z offset for the shot from the aim point would give you the location of the point the projectile passes quite easily. I assumed speed difference was that axis's component of the resultant velocity vector of the two aircraft.  ie taking the difference in velocity vector of the two aircraft would give a resultant vector. eg 3 m/s in the x axis, 2 in the y kind of thing. The formula doesn't involve the angular displacement between target and gunner. Just the distance relative to a point. I don't know what that point is. I would guess it was the aircraft x,y,z location in 3 d space relative to a known reference. In DCS the reference point is as I recall always the 0,0,0 point of the map where z=0 is sea level. The location of the aircraft in DCS is the centre point if I remember correctly.  I would think the IL2 GB team would have a similar system as there isn't too many ways to do it.

 

Without knowing the logic around this calc and definitely how it is handled then it is all guesswork from our point of view. 

 

I've no idea how the formula was derived and no knowledge of the assumptions behind it. It may be a simplification/approximation of the actual complex math for reasons of performance - especially if this calc is done for each round. Which considering damage is by individual round it quite possibly is at this level.

 

I did see that at times you get undefined results when certain values are entered. A lot would depend on how such results are treated downstream in the code. For example, there may be a max error result defined somewhere which is applied in such a situation rather than let such a result flow through. I know personally from a programming viewpoint I would be trapping and handling such a result somehow as allowing such a value to flow through code is highly likely to cause issues.

 

39 minutes ago, Charon said:

The other thing that seems suspicious here is CoeffOverload. Am I reading that right? Does that an ace gunner need to experience 20G before he degrades to the level of a 'high' gunner? That doesn't seem right!

 

The issue is sidestepped as there is a skill independent cut out value of G at which point all gunners cease fire. Stock is 5G I believe

Edited November 27, 2023 by Stonehouse

[LukeFF]

1 hour ago, Charon said:

@LukeFF, do you mind sharing this with your engineers?

 

Sure thing.

[Charon]

I ran some tests with a fighter traversing from one side to another, and was surprised by the results.

 

In the case of a MiG-3 crossing behind a player bf 110E, the rear gunner actually performed quite well. In all 5 trials, he landed hits on the wing, tail, or fuselage of the Mig. Not enough to bring it down, but consistent hits. When I tried this myself, it was quite hard. I was able to get hits in all five trials, but I was never confident in my shooting. I felt like the AI matched or exceeded me here.

 

But then I ran a test with a Bf 109 traversing behind a player A-20; a similar set-up (in fact, the 109 actually passes a little closer to the a-20). But the results couldn't be more different: here the A-20 gunner opens fire at very long range (1km!), reloads, and waves his gun around quite a bit. He struggled to land any hit (I think he might have gotten one or two hits in five trials) and did no serious damage. But when I tried this, I was able to shoot the 109 down four times out of five (PK, fire, damaged, PK, PK).

 

So for some reason, the 110E gunner is seemingly performing on-par with a player, and the A-20 gunner is performing much worse!

 

Some representative video:
 

Spoiler

 

Attached are tracks showing AI performance:

 

ai-tracking-shot.zip

[Stonehouse]

8 hours ago, Charon said:

here the A-20 gunner opens fire at very long range (1km!)

 

So, this is just background commentary, there is a thing with gunners - which I think is to compensate for lack of real defensive formations and bomber numbers in game and to act as a balance mechanism for playability - if they have sufficient ammo, they will fire out to 3x the normal range. You should be able to notice that on initial contact there is a lot of quite long-range fire from the AI that will taper off over time. This is because their ammo supply has dwindled down below the value required for 3x normal range shooting. I did experiment ages ago when first starting the AI gunnery mod with turning off this ability but pretty much with this off if the bomber can hit the fighter, then the fighter can hit the bomber........with a much greater and overwhelming density of fire. Especially if the fighter is armed with cannon. 

 

You would never want to fly a WW2 bomber with AI gunners if this 3x ability was turned off and the defensive formations and bigger numbers of aircraft were not implemented in game properly. This is even with the fighter attack ranges as per AI Gunnery. If using stock with this 3x ability off, it is much worse as Aces will engage at 800m and low skill fighters at 400m (WW2) but even using AI Gunnery where Aces wait until 400m the first decent burst of fire from the fighter usually damages the bomber to point it is destroyed or crippled and then can be destroyed at leisure. So, for WW2 I left it on in the end.

 

However, I did actually turn off the 3x range in AI gunnery for WW1 gunners as they were able to kill fighters at long range due to this ability when running the prototype mod at the time due to the range for WW1 fighters to engage being quite short compared to WW2. So, for WW1 I ended up with reduced ranges for WW1 fighters and whatever the normal engagement range is for the bombers. From memory I think this may be 600m but it isn't totally clear from the comments in the files anyway.

 

Balancing things to end up with something playable is very difficult. If you look through the AI gunnery thread you will see some of the discussions about this.

 

Setting up different burst lengths for gunners according to ammo supply did help quite a bit. Stock has only 1 burst length ranging from 1-1.5 secs. Some gunners use up all their ammo with 1 or 2 bursts in stock. At some point in the future, I may also split burst length by temperature too, but this is much more work. For instance, when I am the Halberstadt gunner I tended to overheat the gun quite often until I figured out a suitable burst length. I'm not sure if the AI respects this temp limit and if they don't, that may explain some of the pauses in firing you get with AI gunners.

 

The various adjustments I've made within the mod was really to have the gunners fire in a more controlled manner and not be that accurate until the fighter is fairly/believably close or if it flies with little vector difference (eg crawling up the bomber's 6 at nearly the same speed as the bomber). I was also able to make AI fighters more sensitive to getting hit as well so they tend to jink away when hit, this spoils their attack and lets the bomber continue on. This way the bombers can keep up a sustained fire for longer and have a chance to achieve their goal - which is to discourage the fighters but not necessarily kill them and deliver their bombs on target. eg if a bomber damages the fighters to the point that they must RTB the bomber has won even if no kills are scored. However, if the fighters get in close enough to fire properly the bomber is going to get shot down. It's fairly inevitable with late war fighters with 20 or 30mm guns.  Some of the other changes I made have seemed to make the AI fighters attack less predictably as well, so more vertical slashing/head-on etc attacks happen than in stock and not all attacks are from 6 o'clock although quite a few still are. At lot of the changes were arrived at via long periods of experimentation rather than knowing exactly the effect of changes to each parameter. Poking around in other words.

 

The 3x range setting is in turretcontrollerai.txt and is EnoughAmmoCount. The stock value is 150 rounds. 

 

Anyway, hopefully someone of the dev team has time to look at the AI side eventually.

 

 

Edited November 30, 2023 by Stonehouse

[Stonehouse]

On 11/30/2023 at 3:18 PM, Charon said:

I ran some tests with a fighter traversing from one side to another, and was surprised by the results

 

Just wanted to say that there are sector limits that place restrictions on where the AI can fire. I'm not sure how to interpret them but they are as below. I assume the A20s are to stop the gunner shooting through their own tail, Ditto the 110g2. I do also note that there are also aim sectors in the turret definition. I believe (but cannot be sure) these are defining the field of fire for the gunner's position while the sector limits define no fire zones. Some aircraft have much better fields of fire than others as well as multiple gunners overlapping the same field of fire, and this has a big impact of defensive performance (eg B25)

 

A20 top turret

SectorLimits = -45.0, -25.0, -10.0,  +90.0
SectorLimits = -25.0, -17.0, -6.0,  +90.0
SectorLimits = -17.0, +17.0, -2.0,   +90.0
SectorLimits = +17.0, +25.0, -6.0,  +90.0
SectorLimits = +25.0, +45.0, -10.0,  +90.0

 

110e doesn't seem to have any (the single line is commented out). 

 

110g2

SectorLimits = -180.0, -35.0,  -5.0,  +90.0
SectorLimits = -35.0, -25.0,  -20.0,  +90.0
SectorLimits = -25.0, -17.0,  -25.0,  +90.0
SectorLimits = -17.0,  -10.0,  -29.0,  +90.0

 

As you can see, they are quite different. I don't know how each value and signage applies and how they should be read to define the zone. You could probably make guesses based on gunner position and aircraft but only the devs know for sure. I don't know if there is a delay to represent picking up the target again for the gunner AI when the target crosses into and out of a no fire sector. If there was, then a large no fire sector could have a larger than expected reduction on opportunity to fire. A human gunner would only have the limit of their reflexes and judgement.

 

Another factor is rounds in magazine and ROF and burst length. Stock burst length applied to all gunners is 1 to 1.5 secs (I assume a random value between these two limits is selected for each firing attempt), the ROF of the M2 is 850 RPM. The A20 top gunner has 30 rounds in the mag. 850 RPM is about 15 rounds per sec. So pretty much the gunner has to reload at best every 2nd burst fired (which incurs a reload delay at least and possibly forces the gunner bot to retravel the aiming process again depending on how the logic is coded - only the devs could say) and at worst every 1.5 bursts. The 110e gunner has 75 rounds per mag and ROF is 1000 RPM. This is 16 rounds per sec. Roughly 2 to 2.5 the number of bursts available before reloading.

 

As an aside it is possible that the single uniform burst length range is what triggers the AI gunners overheating their guns. While it is just my opinion, I would think a gunner trained with a particular weapon would know the burst length to use to avoid overheating and even in combat muscle memory developed with practice would tend to make them try to keep to this unless they panicked or got desperate. This isn't represented in game. Remember there is only a single set of gunner bot parameters (turretcontrollerai.txt) and it is only the turret definition and weapon definition differences that dictate the size of the opportunity to fire, the dispersion and overheating and how reloading plays a part. Only the skill-based error comes from the bot file. I've come to believe that by basing burst length on ammo supply in AI Gunnery I sidestep some of the overheating and certainly some of the reload frequency issue and so the gunner fires longer and with more success than just due to the changes in error parameters I've made.

 

In any case while I could easily be wrong, I would have assumed however that the 110e and the 110g2 would have the same sector limits as essentially the two aircraft are identical from the gunner's viewpoint.

 

@Charon perhaps retry your experiment with the 110g2? I believe you will see different results. The A20 is best at offset rear arc targets, the turrets don't really allow overlapping fire. Dead astern co-alt or very slightly higher is pretty much a blind spot.

 

@LukeFF perhaps the 110e should have sector limits matching the 110g2?

Edited December 15, 2023 by Stonehouse

[LukeFF]

3 hours ago, Stonehouse said:

perhaps the 110e should have sector limits matching the 110g2?

 

My guess is that there is a good reason why they are different - probably because the machine guns are different. 

[Stonehouse]

5 minutes ago, LukeFF said:

 

My guess is that there is a good reason why they are different - probably because the machine guns are different. 

I was just thinking that if the sector limits are there to stop the AI shooting through parts of their own plane, then the 110e2 and 110g2 have exactly the same tail plane configuration so you would expect the same sector limits since the turret location is identical and this would logically be independent of the gun used?

 

The comment in the turret def is

// ограничения сектора обстрела, что бы ИИ не пытался наводится в указанные убласти

 

which google gives me as 

 

// restrictions on the firing sector, so that the AI does not try to aim at the specified areas

Edited December 15, 2023 by Stonehouse

[LukeFF]

Hmm, I'll have to ask about that, then.

[Yogiflight]

22 hours ago, Stonehouse said:

perhaps the 110e should have sector limits matching the 110g2?

 

18 hours ago, Stonehouse said:

I was just thinking that if the sector limits are there to stop the AI shooting through parts of their own plane, then the 110e2 and 110g2 have exactly the same tail plane configuration so you would expect the same sector limits since the turret location is identical and this would logically be independent of the gun used?

The Bf 110 E2 gunner can shoot a little outside of the vertical stabilizers, because he opens the rear window for shooting with the machinegun. That way he is able to turn the machinegun a little more to the sides than in the G2 with its closed rear window. You can see that in outside view (although it is not animated, that the E2 gunner has to move more to the side with the gun as well)

[Stonehouse]

5 hours ago, Yogiflight said:

The Bf 110 E2 gunner can shoot a little outside of the vertical stabilizers, because he opens the rear window for shooting with the machinegun. That way he is able to turn the machinegun a little more to the sides than in the G2 with its closed rear window. You can see that in outside view (although it is not animated, that the E2 gunner has to move more to the side with the gun as well)

They still would have certain zones within their firing arc where they couldn't shoot though? It's not like they are above the tail plane assembly or a tail barbette type thing. Currently the 110E2 has no sector limits at all.