AI Performance

[e.a.wicklund]

If I make edits to the flight_tuning section of aircraft.cfg, does the AI get to use these settings too?

Or does the .air need to be changed so the AI can use that difference too?

[bearcat241]

For any model, AI uses same configurations as the player in general, with a few takeaways...

[No Dice]

This may Help ??

AI Flight Tuning Tutorial by Tom Sanford
find required Air File Manager in the design section

Download


http://thefreeflightsite.com/Vietnam_Projects_2.htm

[e.a.wicklund]

This may Help ??

AI Flight Tuning Tutorial by Tom Sanford
find required Air File Manager in the design section

Download


http://thefreeflightsite.com/Vietnam_Projects_2.htm

Brilliant! This looks really useful. Thanks for the info.

[Captain Kurt]

Here's a bit more info to think about regarding AI performance tweaking.

As Tom Sanford notes in the tutorial that No Dice posted above, to get the AI to perform well - especially in ground attacks - you sometimes have to lower the Moment of Inertia values in the air file. That messes with the player aircraft control ability so you have to put the original Moment of Inertia values into the cfg file. The player aircraft will use that while the AI uses the air file.

This is also true for power and thrust settings. Wingmen continue to pull away from you in mission formations or run over you on the runway before you can get started or clear the runway? Increase the power and/or thrust scalar number in the cfg file until the player plane can hang with the AI. The AI performance doesn't change.

Sanford says this indicates that the settings in the air file take precedence for the AI and the settings in the cfg file take precedence for the player aircraft.

That's not entirely true however.

Adjusting the contact points (points where the wheels touch the runway), the static pitch, and the static cg height in the cfg file to get the player aircraft to sit on the ground correctly will also adjust how the AI aircraft sit on the ground.

Does your plane fly nose high or nose low? Adjusting the empty weight cg (center of gravity) position fore or aft in the cfg file will fix this, and the AI aircraft get fixed as well.

IMHO the AI use the air file for power and flight performance functions and the cfg file for the "visual" attributes of the plane. Experiment with this yourself and see what conclusions you draw.

[e.a.wicklund]

Here's a bit more info to think about regarding AI performance tweaking.

As Tom Sanford notes in the tutorial that No Dice posted above, to get the AI to perform well - especially in ground attacks - you sometimes have to lower the Moment of Inertia values in the air file. That messes with the player aircraft control ability so you have to put the original Moment of Inertia values into the cfg file. The player aircraft will use that while the AI uses the air file.

This is also true for power and thrust settings. Wingmen continue to pull away from you in mission formations or run over you on the runway before you can get started or clear the runway? Increase the power and/or thrust scalar number in the cfg file until the player plane can hang with the AI. The AI performance doesn't change.

Sanford says this indicates that the settings in the air file take precedence for the AI and the settings in the cfg file take precedence for the player aircraft.

That's not entirely true however.

Adjusting the contact points (points where the wheels touch the runway), the static pitch, and the static cg height in the cfg file to get the player aircraft to sit on the ground correctly will also adjust how the AI aircraft sit on the ground.

Does your plane fly nose high or nose low? Adjusting the empty weight cg (center of gravity) position fore or aft in the cfg file will fix this, and the AI aircraft get fixed as well.

IMHO the AI use the air file for power and flight performance functions and the cfg file for the "visual" attributes of the plane. Experiment with this yourself and see what conclusions you draw.

Really good info there. I asked this question because I did a little flight-tuning on this corsair and got it flying as I expect. Still not a good turner, just a better BnZ plane. I took up against the default Georges and took them down with relative ease. As always I strive for good game balance and for the right reasons. So I took the Georges up against the corsair. With the default George, I shot down these new and improved corsairs even more easily. Made me wonder: is the AI really getting as much out of the plane as I do? The game is player-centric, of course. You're supposed to win, but fight for it. Well, I want to have to fight harder. I'll be testing this new information out and see what I get.

[bearcat241]

...Made me wonder: is the AI really getting as much out of the plane as I do? ...

The answer is no. Although the player and AI share the same essential flight modeling, actual performance results come down to application, where the advantage goes to the player. With any flight dynamics, the player is ever evolving in learning how to use knowledge and tactics to best exploit the strengths of his model, overcome the model's weaknesses and win the fight. The AI don't have a learning curve and are relying on the linear programming of the ai_playr.dll to win fights against both player and other AI.

The FDE takeaways for the AI that i alluded to earlier allow them to do maneuvers in certain combat situations that a player can't follow, even when flying the same model. IMO the stock Zero and Oscar fly like pregnant pigs for the player in ACM - it's why i never liked using them in past multiplay. But as AI adversaries they can put up a good fight against all of the stock Allied models, so we see the advantages of FDE takeaways in effect.

But even with these advantages they can still be obliterated by a smart player because there are certain player maneuvers they just can't comprehend and compensate for. For example, get an AI "anything" (set as Ace & Highly aggressive) on your six in a Wildcat, go balls to the wall in a shallow high speed dive and then pull up into a full, tight overhead loop and see what happens. I guarantee they won't react like a human would in multiplay given the same move. They simply don't push the envelope when you show them something outside of their standard "turn & burn" programming and default to maneuvers that make them easy targets.

I spend 95% of my flight time in QC, so i would say from experience that the AI does not get as much out of a model as a player would...

CheerZ!

[Roxane-21]

AI is programmed to act like an A6M2 pilot (remember: CFS2 is a Microsoft game, Microsoft is an American company, Japan was America's enemy during WW2, so in CFS2 you are supposed to fly the US side against the Japs).

If you fight against AI Corsairs, they will act like they are actually flying Zeros, and will try to dogfight with you. Every Corsair pilot who tried to turn at low speed against a Japanese fighter in real life can tell you (if he even survived to do so): DON'T DO THAT!

[bearcat241]

...Every Corsair pilot who tried to turn at low speed against a Japanese fighter in real life can tell you (if he even survived to do so): DON'T DO THAT!

Its why i compensated for this by giving all of my favorite American models much improved elevator and aileron effectiveness. To your point that the AI programming is universally rigged for TnB over BnZ tactics, i decided that would give my squads a better fighting chance in the merge. They turn tighter and roll faster at lower speeds than stock FDE, which are a big disappointment out of the box.

The RL Hellcat was not as fast as the Corsair in top speed, but it was a great turner with a lot of power and held the distinction of being the premiere Zero killer in WWII, and also had the highest kill ratio for the entire PTO war against all types. Most first hand accounts i've heard and read of former Hellcat drivers indicate a lot of success in the turning fight, even against the experienced Japanese pilots of Rabaul and Truk in 43 and 44.

Take away its bomber and transport shootdowns, you still have an impressive record in fighter vs fighter action, even in engagements with the late war hotrods like the Ki-84, ki-61, ki-100 and others that the Japanese were operating with in the defense of the Japan homeland in 44 and 45. And unlike the kamikaze corp of that period, the homeland defense pilots were the best of the IJN and IJA - highly trained, battle-tested survivors from other regions, reassigned solely to defending the homeland against American airstrikes and raiding fighters. These guys were good dogfighters. So overall, the Hellcat's record was not managed by strictly BnZ fighting as many would like to claim, but by mixing it up with hard core dogfighting and good team tactics.

That said, some surviving Japanese aces who flew the latest and best fighters have been quoted as saying that the plane they feared and respected the most in the turning fight was the Corsair. They have said that it was wicked fast, had a fast roll rate at low speed (strong torque action i presume), as well as a good turn rate aided by its massive wings and huge rudder (7.3 - 8.0g Corner speed @ 278-300 mph is extremely strong for a 9000+ lb plane that's not supposed to be dogfighting at those speeds ???). I once read accounts of one Japanese ace whose name escapes me right this moment who flew Ki-84's and Ki-100's against both Hellcats and Corsairs in the DOJ chapter who even claimed that the Corsair had a tough hide from a six o'clock position, with bullets and cannon shells bouncing off its sleek forward tail and fuselage sections in high speed pursuits.

[bearcat241]

AI is programmed to act like an A6M2 pilot...

Only to a very limited degree would that be true. They could have researched the subject a bit more thoroughly. From my admittedly casual findings, one of the absolute favorite tactics of early war Zero pilots against American fighters of that period - before the Hellcat's debut - was to take the fight into the vertical to exploit the stronger climb rate and lighter weight of the Zero. It was said that if a Wildcat followed a Zero into a full loop the Zero would be on the Wildcat's six inside of the second turn. The claim is that many Japanese aces were made against the Wildcat in the vertical fight, using its heavier bulk and slower rate of climb against it for an early stall. One would think this kind of maneuvering and aggressiveness could have been easily mimicked in the programming.

I've seen a lot of AI Zero's pull a half-hearted chandelle as a defensive move to vertical, but fail to convert it into anything offensive...just kept running if i let them...that's not the way an A6M2 pilot wins a fight against a slower-climbing heavier foe.

I could go on...

[Roxane-21]

7.3 - 8.0g Corner speed @ 278-300 mph is extremely strong for a 9000+ lb plane that's not supposed to be dogfighting at those speeds ???

This is not what I would call "low speed". While dogfighting against AI fighters airspeed quickly drops below 200 mph - a range of speed the Corsair can easily be outturned by any Japanese fighter. You then need to exit the turning fight and gain airspeed before re-engaging, thing AI Corsairs will never do.

I've seen a lot of AI Zero's pull a half-hearted chandelle as a defensive move to vertical, but fail to convert it into anything offensive...just kept running if i let them...that's not the way an A6M2 pilot wins a fight against a slower-climbing heavier foe.

That's true. I always had the feeling that AI fighters loose their mind when they come to inverted flight. Even if I have a more agile enemy fighter near my tail, I often find escape making an inverted loop. I once survived an entire BoB Bf-110 campaign that way.

[e.a.wicklund]

AI is programmed to act like an A6M2 pilot (remember: CFS2 is a Microsoft game, Microsoft is an American company, Japan was America's enemy during WW2, so in CFS2 you are supposed to fly the US side against the Japs).

If you fight against AI Corsairs, they will act like they are actually flying Zeros, and will try to dogfight with you. Every Corsair pilot who tried to turn at low speed against a Japanese fighter in real life can tell you (if he even survived to do so): DON'T DO THAT!

That explains a lot. When I tweaked the corsair FDE, I didn't specifically work on turning ability. I thought that was good as is. When I took a George against it, the AI corsair pilots were trying to fight the George's fight. Bad move. I didn't configure it for TnB.

For me, I found the default A6M to be an excellent dogfighter. Attacking anything with it is easy to me. Often I find it relaxing. Still, based on what you folks are saying, I need to tweak the .air file anyway, just to make it a more dangerous opponent as an AI.

So then, a question for the group. Is it possible to influence the FDE and force the AI to fight in the vertical?

[Ivan]

Gentlemen,

The reality may be a whole lot simpler than what you are describing.
Consider from a programming standpoint if you were designing the "tactics" into the AI Pilot WITHOUT knowledge of the relative performances of the aircraft involved.
It is very simple to calculate the flight path needed for a "Lead Pursuit" and to a lesser extent a "Lag Pursuit", but WHEN does your AI pilot change to vertical maneuvers? Under what conditions would it attempt a High or Low Yo-yo?

As Human pilots, we come into a game with a preconception of the relative performance of our aircraft and the opposing aircraft.
Note that with the mention of the A6M2, we all have a pretty good idea of the expected characteristics though we may argue a bit about the details. The same is true about Hellcat, Corsair, Hayate and Ki 100. Was the Ki 100 faster than a typical Hellcat? We may disagree but only about detail.

Now imagine if we are the AI programming and need to make decisions ahead of time without knowing anything about the aircraft and without being able to adapt to what we see. A turn fight resulting from flying pursuit curves is about the only reliable canned tactic that will work.
To do better than that requires some much smarter AI programming than I believe was done for these games.

(Just my view of the world.)
- Ivan.

[e.a.wicklund]

Only to a very limited degree would that be true. They could have researched the subject a bit more thoroughly.

So are you saying the AIis "not" designed to fight with an A6M? Or that they could've researhed it better? I'm guessing the latter, and would agree with that.

From my admittedly casual findings, one of the absolute favorite tactics of early war Zero pilots against American fighters of that period - before the Hellcat's debut - was to take the fight into the vertical to exploit the stronger climb rate and lighter weight of the Zero. It was said that if a Wildcat followed a Zero into a full loop the Zero would be on the Wildcat's six inside of the second turn. The claim is that many Japanese aces were made against the Wildcat in the vertical fight, using its heavier bulk and slower rate of climb against it for an early stall. One would think this kind of maneuvering and aggressiveness could have been easily mimicked in the programming.

I've heard the same thing. A lot of F4Fs were destroyed using this tactic. Though, these pilots had a rude awakening when they tried the same against the Hellcat. The newer machine could stay with them and not stall out.

[bearcat241]

...Now imagine if we are the AI programming and need to make decisions ahead of time without knowing anything about the aircraft and without being able to adapt to what we see. A turn fight resulting from flying pursuit curves is about the only reliable canned tactic that will work.
To do better than that requires some much smarter AI programming than I believe was done for these games.

Yes...preemptive programming is tougher to write for AI and only good for initiating the fight. But after the fight is on, the designers relied mainly on reactive programming to keep it going and didn't put enough variety into the reactions. It doesn't take long before every reactive maneuver looks canned and predictable. Eventually it becomes less like simulated A2A combat and more like target practice because you know where you have to be before the target "arrives".

EAW: of course i meant the latter in your question. The CFS2 AI design is applicable to any model. I've just never been very impressed with the engine's decision-making. This sim was my first venture into WW2 aerial simulation back in 2001. I came directly from Jane's world of jet combat where the AI were very unpredictable, player-hungry and deadly - with missiles to top it off.

It didn't take much effort for me to adapt to a slower pace of action, situational awareness and real-time anticipation in CFS2. The only remedy i ever found to make the AI engine think faster and more aggressively - out of the box as they say - is to increase the sim rate speed to 2x. Gets crazy after that!

Attached is a gift, some of my homemade chili...good for AI...give me your honest opinion.

[e.a.wicklund]

Attached is a gift, some of my homemade chili...good for AI...give me your honest opinion.

Yeah, that is a bit too much for me. I was constantly blacking out during every turn, no matter how shallow, because the plane is so fast. For me, a plane like this unbalances the game. Thanks for posting it though. I appreciate that.

[bearcat241]

That's why its called chili

[sixstrings5859]

You are correct Bearcat. The AI in CFS2 is predictable but fun if you want high scores. Later combat flight sims i use have far better AI's and are more challenging but that just goes with advancements made. The IL-2 BoX series has one of the best AI i've ever seen,but they are new releases.CFS2 was good for back in the day but IL-2 '46 still gets the nod in the AI department for a older sim,with all the mods made for it. Glad i got back into CFS2 as it is still enjoyable to fly and easy to master the aircraft. Sometimes it's nice to get away from serious combat flight sims and just have fun.Regards,Scott

[bearcat241]

Copy that...i totally agree on your IL2 AI point. Think of CFS2 as gunnery school and IL2 as the actual

[e.a.wicklund]

In aircraft.cfg, there's a parameter called Elevator_Effectiveness. Is there a corresponding parameter in the .air file for this?

Since .air files are read for the AI first, I'd like to change a parameter for them, and maintain the current setting for the player. So, how to improve Elevator_Effectiveness for the AI, but not for myself in aircraft.cfg? As is stands, I don't see a directly correlating parameter.