What has happened to our gunners???

[[CPT]Crunch]

Funny how they can still pull off the incredulous shots though, 90 degree crossing shot on an RTB A-20, he was low on the deck, I was slightly lower in a bomb laden Fw-190 outbound on a strike mission.  Put rounds in his number one and set in on fire, crossed slightly over his tail at close to 450 Kph still at a 90 degree angle, never intended to pursue or deviate course, he just happened to be crossing my gun sight.  Gunner pings me on the pass at over 300 mph, yeah right, how could it see the approach and acquire, let alone have the speed and trigger reflexes.  Needs a general overhaul, can't hit where it should be hitting, yet still hits the pure bullshit one in a million impossible shots.

 

Good job on the A-20 pilot, he got the fire out and made it home safe.

[IckyATLAS]

1 hour ago, [CPT]Crunch said:

Funny how they can still pull off the incredulous shots though, 90 degree crossing shot on an RTB A-20, he was low on the deck, I was slightly lower in a bomb laden Fw-190 outbound on a strike mission.  Put rounds in his number one and set in on fire, crossed slightly over his tail at close to 450 Kph still at a 90 degree angle, never intended to pursue or deviate course, he just happened to be crossing my gun sight.  Gunner pings me on the pass at over 300 mph, yeah right, how could it see the approach and acquire, let alone have the speed and trigger reflexes.  Needs a general overhaul, can't hit where it should be hitting, yet still hits the pure bullshit one in a million impossible shots.

 

Good job on the A-20 pilot, he got the fire out and made it home safe.

What was the AI level of the gunner or what has been coded by whoever made the mission. You can be High Medium or Low. At high it is laser beams guided by radars locked on you. Ideally for the AA you should be Low to be realistic. But also the preparedness and reaction time depends on another parameter that can also be defined.

If everything stands at the max say for a Vierling, and you are inside the kill zone then you cannot escape. 

[[DBS]TH0R]

Wish I had that A-20 gunner on at least one of my sorties... ?

 

7 hours ago, ShamrockOneFive said:

That's definitely problematic then.

 

It seems there's a narrow window between sharpshooter and spray and pray modes for the gunners.

 

As if the margins only got higher. Gunners are now next to useless in MP environment.

[oc2209]

17 hours ago, 6./ZG26_Custard said:

Currently, fighters can just sit on the six of bomber aircraft with impunity. There is more or less no threat. Some of our squad was flying on Flugpark, in Bristols on Friday and everyone's aircraft was shot down, by one fighter. Just sat on the tail with not a care in the world. 

 

I'm wondering if this is another online versus offline performance difference. You know, like pilots being killed from not-very-hard landings online.

 

I just did 4 quicks, ace enemy, 1 B-25, 3 Pe-2 (w/turret).

 

Attempt 1, versus Pe-2: I was hit in the engine (oil leak) at 320m. Subsequently killed outright at 30m (by the ventral gunner):

 

Spoiler

 

Attempt 2, versus B-25: destroyed bomber, but I was hit in the fuselage (to no effect) at 260m:

 

Spoiler

 

Attempt 3 versus Pe-2: destruction of target with no successful return fire.

 

Attempt 4 versus Pe-2: destruction of target with severe damage taken. Recording:

 

Spoiler

 

 

I thought the clip was worth sharing also because of the blind landing I survived with a wheel missing. No injuries despite the flip. Touchdown speed was 108 MPH.

 

In all of the above attacks, I made no special effort to approach the target evasively. So in other words, the laziest and most reckless approach possible.

 

I was critically damaged in 2/4 attempts, which I consider pretty decent performance for the AI.

Edited February 15, 2022 by oc2209

[Lusekofte]

I think ai aa,aaa and gunners is gonna be a constant frustration.

I like to fly IL 2, PE 2 and A 20's

I got shot down a lot when they where sniper. Now in MP it is damned hopeless

So aqmb and career is pretty much all I do.

There is not going to be a perfect setting for it in foreseeable future, I guess the coders are quite busy.

The fighter jockey are finally silenced. Now they have their fat targets, and improved killrate. 

I for one got more kills sitting in the turret myself, than my sniper gunner ever did.

[kendo]

I expect this will be tweaked again in the next update.

 

Getting AI to simulate human gunners realistically and miss the target relies on clever coding. They changed things to get away from the 'sniper AI' that everyone has suffered from and complained about, but obviously they moved too far the other way. 

 

The good news is it is being addressed. Hopefully next tweak will get it close to where we want it.   

[Zooropa_Fly]

On 1/26/2022 at 12:39 PM, ShamrockOneFive said:

I guarantee that it does. Setup a Stuka in QMB and run it on rookie and ace and notice how much more quickly the ace gunner can bullseye you.

 

I was wrong, however, that nothing had changed as someone noted above that G forces cause more issues for gunners. I suspect a lot of players don't give their gunners the ability to fire back and swing around in tight turns and high G evasive maneuvers.

 

I always thought that the ai skill level made no difference regarding player controlled planes.

But it certainly does to the gunner of a bot as you say.

[I./JG52_Woutwocampe]

As of now, the main threat of being at the 6 of a BF110 is not its gunner, but the flaps, as they lose them with little provocation, just like the Mustang.

[6./ZG26_Loke]

Quote:

Originally posted by Mahoney:

Saddling up to 2 seaters was suicidal in WWI; just about all the aces agreed attacking 2 seaters was really more dangerous than dogfighting other scouts, because mobile rear-facing guns made them so lethal. Planes like the FE2b and Bristol Fighter were dangerous enemies. 

Un-quote. 

 

This is far from the case in FC now, were you are just a flying target if you fly a 2-seater. 

 

[[CPT]Crunch]

I'd be happy if the individual gunners don't turn on or track until a plane enters one of their cones of fire all based on historical aircraft gun position cones of fire, none of the AI gunners are alerted or go activate to an enemy presence until an aircraft physically enters one of their firing cones, within a max distance like one kilometer so they aren't constantly burn cycles.  Once the crew is alerted and go active, give each gun a slight delay before it can physically swing the guns in their specific cone to simulate the time for target acquisition, ranging, and proper gun positioning, that allows surprised and on occasion they can be caught pants down by a proper stealthy fighter approach.  Keep the hard G maneuvering puking and unable to target accurately gunners. 

 

Need a speed limit on how fast they can physically swing and elevate a gun, in all other functions let them be Carlos Hathcock himself, you approach careless you get what's due, and the bomber and attack pilots are reasonably protected.  This should enhanced their firepower when grouped in smart formations placing overlapping cones. 

 

Oh, and make them fighter friend or foe ID blind for one or two seconds when fully combat alerted, the gunners per station will fire on any fighter that moves too fast too close into their cones when under enemy fire, until they can access and make an ID in X milliseconds, simulates the panic and mayhem so allied fighters must also use some common sense and caution.  Closest fighter popping into the cone always the priority, enemy or not.  Once the bullets start flying and every crew is already fully alerted, drop the timings a bit.

 

 

[messsucher]

They can spot stuff outside of their cone of fire, and I think they did that. Else good suggestions.

[[DBS]TH0R]

10 hours ago, kendo said:

The good news is it is being addressed. Hopefully next tweak will get it close to where we want it.   

 

Where was this news posted? I've missed it. Thanks.

[6./ZG26_Custard]

1 hour ago, [DBS]TH0R said:

I've missed it.

So did I ?

[LLv34_Temuri]

On 2/14/2022 at 2:59 PM, ShamrockOneFive said:

That's definitely problematic then.

 

It seems there's a narrow window between sharpshooter and spray and pray modes for the gunners.

No. There aren't sharpshooter AI gunners anymore, at least not in multiplayer environment.

 

Edit: what I mean is that the correction was overdone, I'm quite sure it can be tuned somewhere in between.

Edited February 17, 2022 by LLv34_Temuri

[ShamrockOneFive]

7 hours ago, LLv34_Temuri said:

No. There aren't sharpshooter AI gunners anymore, at least not in multiplayer environment.

 

Edit: what I mean is that the correction was overdone, I'm quite sure it can be tuned somewhere in between.

 

I think we have to consider that this is a multiplayer issue, at least in part, as they are still sharpshooters offline in some instances at higher skill levels.

[Stonehouse]

This translated comment below from turretcontrollerai.txt is interesting.

 

Recalling

AddCoef is default aiming error

ProjSpeedCoef is relating to the relative speed difference between target and gunner. I assume this somehow takes into account the speed difference in 3 dimensions so a vector type calc rather than just a simple difference in m/s.

OverloadCoef somehow takes into account G forces and their impact on aiming

CoefLow,CoefMed,CoefHi,CoefAce are skill based factors and going on stock numbers lower is more highly skilled.

 

You can see that changes to ProjSpeedCoef and OverloadCoef will have a large impact as they are multipliers.

 

The whole resultant of coefficients and g load and speed difference is then multiplied by a random number between -1 and 1 and the tangent taken to come up with an error in meters.

 

Other big factors are relative speed and how many G's the gunner is experiencing at time of shooting.

 

Seems like ending up with a g loading of zero means the gunner is more accurate but I guess you would assume that likely speed diff could be quite high then as it means the gunner's aircraft is accelerating down at -1g as I assume level flight is current overload of 1?

 

<edit> maybe the current loading is zero in level flight so it is more of a difference to 1 g value? This would mean that pushing down at 1 g is a current loading of 2 perhaps (at least not zero anyway) and resolve the question above  

 

Also note that if the relative speed is zero between gunner and target and the current overload is 1 (gunner a/c not manoeuvring) and you set the AddCoef to 1 and OverloadCoef to 0 then I think the error is entirely based on gunner skill values with the random number applied.   <edit> If my thought about current loading being a difference to 1g value instead of the actual g loading is right then level flight is current loading zero and it doesn't matter what the OverloadCoef is for level flight for the case of AddCoef =1 and zero speed diff - error will be pretty much tan (0) - tan(skill). 

 

If AddCoef = 0 then there will be zero error when there is zero speed difference as "( AddCoef + (ProjSpeedCoef * speed diff) )" will be zero and so you'll end up doing tan(0) which is 0. This is regardless of skill or g loading. So you would not want to make AddCoef too small.

 

The skill coefficients seem a bit odd to me. If AddCoef is 1 and OverloadCoef is 0 or current overload is 0 and there is no speed diff, it seems like the formula gels down to 

error = tan (random number * 1 * Skill coeff) and since random number is -1,1 you can think of absolute error as being between tan(0) as zero would be the lowest absolute value of the random number and tan(skill coeff)?

 

If I'm not goofing things up this would mean in this case an ace gets between 0 and 1.55m (1.55 being approx tan(1)) error and a novice gets between 0 and 0.64m error (0.64 being approx tan(10))

 

Maybe I'm just missing something or not understanding but that seems like skill is working back to front? Perhaps someone else can check this.

 

<edit> - that will teach me to rely on internet calculators. Actually tan(1) is 0.01745 and tan (10) is 0.17633 so the error range for an ace in the above scenario is 0 - 0.01745m and for a novice is 0-0.17633m so a factor of 10 roughly which looks in line with an ace skill coeff of 1 and a novice skill coeff of 10 and all is right with the world. Sorry it has been a l-o-n-g time since high school and uni and I guess once upon a time I would have spotted the bad results from the calculator right away but not these days. 

 

The comment about the multiplier coefficients in the formula below having a big impact is ok.

 

Could probably do some spreadsheeting for different AI levels and coeffs with various speed diffs and g loads to tune the blind gunner mod to your satisfaction if you were keen. Remember the order of operations if you do though as the formula is actually evaluated as below:

 

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

 

Quote

 

// 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/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 final zone is obtained

 

 

 

Edited February 18, 2022 by Stonehouse

[oc2209]

6 hours ago, ShamrockOneFive said:

 

I think we have to consider that this is a multiplayer issue, at least in part, as they are still sharpshooters offline in some instances at higher skill levels.

 

Agreed.

 

The more people talk about certain problems they're having, the more it feels like we're playing two different sims (offline versus online).

 

Prior to the most recent AI gunner changes, even LMGs were deadly out to 500m. Nailing you with a headshot was almost a guarantee if you got within ~250 meters.

 

That accuracy has been scaled back considerably, but on Ace level the gunners are still far from being blind or worthless. You're still in plenty of danger once you near the 300m mark. Pilot kills seem to require about 100m to happen consistently. Consistency is the big difference between old gunner AI and the new, from what I can tell. The new AI can take long shots, but it's not as absolutely certain to annihilate you as it was before.

 

Just a few examples from today:

 

Multiple wing hits from 230m while maneuvering:

 

Spoiler

 

Extreme range (470m) Ju-88 hit that damaged my coolant system:

 

Spoiler

 

Here's a spread of hits I took from one Ju-88's burst, from approximately 250m:

 

Spoiler

 

The above hits aren't just pretty DVD markers; they damaged my engine and coolant system.

 

So it appears single player quick mission gunners do not match the 'can't hit the broad side of a barn' reports I'm hearing about multiplayer gunners. I have not tested career AI gunners lately. I know they should theoretically behave similarly, but in my experience career AI differs considerably from quick AI. 

[[RBRI]Khaela]

There's also the point of detection and callouts which is sort of behaving wierdly. 

This is a clip of a bombing sortie we flew on Finnish a week or two ago. 
Gunner seem to be holding on to the gun, but there's no attempt to aim at the target, nor is he calling out any enemy planes until a second fight passes us. And I'm 99.9% sure my gunnery stations were locked, so no human gunner.
https://www.twitch.tv/videos/1301836091

[Stonehouse]

Did up a spreadsheet. Some strange and interesting things in some cases due to the results out of taking tan of the coefficients. Particularly if you fiddle with skill values for speed differences around 30 m/s. You can get situations where making the skill for low twice as bad as stock is more accurate than making it 1.5 times as bad. With a speed difference of 30m/s you get really weird stuff where the max error for low skill of 20 gets -166.43m @ 100m but making the low skill 15 gives a max error of -7639m @ 100m. Just a quirk of results of input values & tan for a specific situation. This gets worse obviously the greater the range. So probably if fiddling skill values expect some odd one-off cases where the numbers combining with tan gives off the scale results. I think it would be extremely hard to come up with a algorithm which was perfect across all situations that wouldn't bring your PC to a halt so I think the devs would have gone for an approximation of the calculations that worked the majority of the time in the interests of performance.

 

The overload coeff did not seem to have as much impact as I expected and the main driver (other than AddCoef which is your default error) appeared to be ProjSpeedCoef. The change from something like 0.05 to 0.2 makes gunners approx 2-3 times less accurate and even if there is a 1m/s speed difference and no g loading the max error at 100m goes from 1.75m to 4.37m for low skill.

 

Note I ended up converting to radians before taking the tangent. This was mainly because the error seemed to get pretty big even for 100m range and level no speed diff if I used degrees. Using degrees a low skill gunner on stock values ended up with a max error of about 54m at 100m. This seemed too extreme even for a low skill gunner. If I have got it wrong the errors will tend to be larger but generally comparing the effects of the skill values and coeffs will be the same. 

 

20m/s is about 50mph and 40 m/s is about 100mph. As a general tip for stock players it seems like generating a speed difference between you and the target bomber will cause the biggest gunner error. Making the target bomber manoeuvre so it is under g loading doesn't make a big difference. 

 

Hope it helps someone tweak turretcontrollerai.txt to give them what they want. If someone particularly wants the Excel file I can upload that too.

Book1.zip

Edited February 19, 2022 by Stonehouse

[migmadmarine]

I think there is also an element of perspective bias that can play in here as well. Especially with the WWI aircraft this is noticeable, since you can look back and watch where the gunner is aiming easily, but it seems that since the changes the gunners, when they miss, are more often missing by absurd margins. Because of this, as the pilot of the aircraft you'll be watching over your shoulder, and thinking "you moron, why are you shooting with the gun pointed 45 degrees away from the target" and getting madder and madder. I can't speak to their effectiveness in multiplayer PVP, but in PWCG co-op and single player, it feels (as the attacking fighter) much more convincing now for the most part than it did in the past, they still hit you, especially if you're attacking stupid, but as they were before my friends and I feared Pe-2s and Halberstadts more than fighters, which seems very backwards.