For thoes that are interested LETS TALK AI PREFORMANCE

[Blood_Hawk23]

"And it begins"


This is what happens when we talk about Mission Builder and DP files at the same time. Its funny, really. It always happens.

DP Files.

When we talk about DPs, no mater what they are for, "Nearly" all measurements are Metric. It is the same for an Aircraft, Vehicle, Infrastructure, Ship, and weapons DP. I'll cover the main points that measurements are used.

[MISC_DATA]

In the [MISC_DATA] section,
Speed is in KPH or KM/h which ever you prefer.
Altitude is in Meters.
Weight is in KGs.

example #1:

Stock Zero,

[MISC_DATA]
unit_family=1
category=1
allegiance=4
max_group_size=8
min_speed=240
cruise_speed=415
max_speed=622
min_alt=50
cruise_alt=6667
max_alt=10000
entered_service=8/1/40 as a note this is MM/DD/Year
crew=1

Example #2:

wep_ja_30kg_gp-Japanese 30 kg bomb,

[MISC_DATA]
unit_family=4
category=37
max_weight=30
weapon_type=1

Example #3

sha_indiana-USS Indiana Stock ship,

[MISC_DATA]
unit_family=5
category=17
allegiance=1
min_speed=0
cruise_speed=30
max_speed=55
entered_service=12/1/41

[GUNSTATIONS]

Now I did say "Nearly" all measurements. Well they threw us a curve ball with the [GUNSTATIONS]. The last measurement is for the round weight. That is measured in Ounces. Why did they do this? I have no idea.
This holds true for ALL [GUNSTATIONS] entries. That is the weight of a single round. I believe the built in ballistics calculator uses it for the trajectory of the round but thats a guess. It is however used for ammo weight. As such it will effect aircraft performance in the case of bombs and drop tanks. (be aware that this armament weight is added to that of the fuel and then the aircraft's dry-weight to give a total weight!). The rest is in Metric. The X,Y,Z is in Meters as is the Range. Speed is in MPS - meters per second.

Note that the gunstation.3 has no weight. That is purely the trigger for the bomb. Its weight comes from the Object_DP.

Example #1:

Stock Zero,

[GUNSTATIONS]
gunstation.0=0,45,1,0.06,750,2,0.01,600,0,40,1d1*11,0,0.616,0.907,0.14973365220456,0,0,0,0,0,0,0.4
gunstation.1=1,47,2,0.12,555,2,0.01,1000,4,40,1d1*30,-1.881,-0.233,0.693,0.273439086965745,0.359239830009869,0,0,0,0,0,5
gunstation.2=1,48,2,0.12,555,2,0.01,1000,4,40,1d1*30,1.896,-0.233,0.693,0.273439086965745,-0.362104505466363,0,0,0,0,0,5
gunstation.3=3,21,8,0.00,0,0,0,0,0,0,0,0,0,0,0,0,0 ,0,0,0,0,0

Example #2:

wep_ja_30kg_gp-Japanese 30 kg bomb,

[GUNSTATIONS]
gunstation.0=3,21,8,0.00,0,0,0,0,0,0,1d1*660,0,0,0 ,0,0,0,0,0,0,0,1058

So, you should now be completely confused.

If not, Great! You're on you way to understanding more of the inner workings of CFS2. Now for the sake of our sanity, I will cover Mission Builder in another post.

If you have questions please, fire away. If you want to know where I have acquired this information then that is simple. Get a copy of DPed. Its in the "How To" doc. For anything else, refer to management.

Till Later,
John

[codeseven]

Thanks John, interesting stuff.

My question is, concerning the AI only, how much of their behavior is 'hard coded' in, i.e. you cannot change it. For example, after an AI aircraft lands it typically slows then veers to the right. That last act of veering to the right seems unchangeable. No matter what parameter is changed in .air, .cfg, .dp, ect. that behavior seems to never change. Can it be?

Also, concerning AI only, you can add a tail hook function to a player aircraft but is it possible to add that function to an AI aircraft? The same player aircraft that will drop a hook seems to ignore that available function when in the sim as an AI aircraft. Again, it seems that the 'hard coding' of the sim will not allow certain changes to the AI.

Unlocking secrets of the AI is the longest running battle in CFS2.

[Blood_Hawk23]

Thanks John, interesting stuff.

My question is, concerning the AI only, how much of their behavior is 'hard coded' in, i.e. you cannot change it. For example, after an AI aircraft lands it typically slows then veers to the right. That last act of veering to the right seems unchangeable. No matter what parameter is changed in .air, .cfg, .dp, ect. that behavior seems to never change. Can it be?

Also, concerning AI only, you can add a tail hook function to a player aircraft but is it possible to add that function to an AI aircraft? The same player aircraft that will drop a hook seems to ignore that available function when in the sim as an AI aircraft. Again, it seems that the 'hard coding' of the sim will not allow certain changes to the AI.

Unlocking secrets of the AI is the longest running battle in CFS2.

From what I've read in Tango's AI handling. Most of what the AI uses comes from the AIR file. We add the tail hook to the CFG. Alot of what is used in the CFG the AI tends to ignore. So I don't think we can give them hooks.

As far as veering to the right, It could be a hardcoded. but it may depend on the Number of planes in the flight. If you have a single AI land it might not turn right. Thats a guess. you figure they would want the AI to clear the runway. but if there are no other planes set to land at that airbase maybe they won't veer off.

you could make a test mission with the player set to "Land anywhere". If that works then you could try to hand edit the base in for the player. I'm only guessing at this. but this works alot of times when dealing with AIs.

an example is the Perseval balloons in the Aerocrate missions. to make them seem like they are floating you have to trick it. Greg has it in his instructions for the balloon if your interested. check the Aerocrate web site.

so thats something to think about.


Oh one other thing, well a question really. The LSO on the ships, does anyone know how that functions?

There are enteries in the DP for it. How do we set them?

[Talon]

From what I've read in Tango's AI handling. Most of what the AI uses comes from the AIR file. We add the tail hook to the CFG. Alot of what is used in the CFG the AI tends to ignore. So I don't think we can give them hooks.

Blood,
Actually the AI use more of the CFG then you think.I have experimented changing parameters in the CFG and it does effect the behavior of the AI.I still don't think you can get the tailhook to work for the AI though.

This thread was about landing at Henderson and I think your answer might be in the airbase data parameters.I'm not an expert but that's where I would start.Don't over think things,sometimes the answer is simple.

[Sarg Willy]

I,ll second Talon,s comments , The Ai and Enemy Ai both read flight tuning spec,s , from the CFG files , In refferance to Elevator effectivness , Rudder effectivness , and Aliron effectivness , normally set to 1.0 , Radical differences occur when changed to =0.5 slows reaction down , and mimicks novice pilot , and conversely =1.5 react,s much faster and has Gallands touch ,


As far as Ai landing and running across the next three states , That is a priority control issue , for player Ai ,
If the enemy Ai are still in route to their base , they have priority 1 and your player Ai get no 2 ,
They will not park , until the E-Ai have all landed and parked , or crashed , ditched , Only then do the player Ai land and park succesfully , when priority switchs them to No 1 control

[codeseven]

As far as Ai landing and running across the next three states , That is a priority control issue , for player Ai ,
If the enemy Ai are still in route to their base , they have priority 1 and your player Ai get no 2 ,
They will not park , until the E-Ai have all landed and parked , or crashed , ditched , Only then do the player Ai land and park succesfully , when priority switchs them to No 1 control

That's interesting, I didn't know that and that would account for the AI sometimes landing and just rolling away.

[Sarg Willy]

One other point , regarding bomber run,s , I have to thank some of the airfile gurus , that sort out flight controls , eliminating the porpose effect , as much as possible , Although I swear mission way points , introduce the effect , no matter how stable the Ai contols are ,
I was test flying the B-25 , with 3 Ai -B25 , and set up 8 , 109e to intercept , As the bombers were straight and level , 2 on my right , 1 on the left , I decided to lock on Auto pilot , ever wounder how the E-Ai target your plane , and pick on you first ,, Because it,s manually flown , you deviate controls more than your Ai flight does ,
Auto pilot confuses their target ability , As lead ship the Ai follow your auto pilot ,
Here,s the best part , you free up processer function , and that means your Ai bomber gunners can track and rotate , process infoe faster , and there hit rates , improved dramatically, It took 5 strings of 8 ,109,s
and the loss of 2 B-25 , before my crew bailed ,
The next bomber I tryed was not stable on Ai controls
and the 109s had no problem tagging the player aircraft first ,

[kelticheart]

.....Unlocking secrets of the AI is the longest running battle in CFS2.

OH, YEAH!

One of the things I would gladly tamper with, should those secrets be eventually unlocked, is the flight behaviour of the AIs during combat.

All of the issues addressed over the years on these pages concerning AIs have been great and the fabolous work done by many very patient members is "above and beyond" any imaginable improvement over the stock sim.

But there's still one that bugs me anytime I engage a dogfight. I made this comment hundreds of times already, I simply find ridiculous that ALL OPPONENTS behave like Japanese Zeros fighters when attacked. Wildcats, Hellcats and Corsairs become all Zeros when I fly as a Japanese in the PTO!

German Bf109 pilots did not use tight loops or hammerhead stalls as evasion or countermeasure manoeuvres, they took advantage of their fuel-injected DB engines and went into sudden dives when attacked. From the gun camera footage that I saw, radial-engined FW190 pilots used the same evasive tactics. So did most of RAF and USAAF fighter pilots, except BoB period Spitfire and Hurricane MkI fighters, whose float carburettor fed Merlins stopped running during positive-G manoeuvres. The gravity forces acting on their carbs during those manoeuvres starved dry the engines which quit on the pilots instantly.

Hence, BoB RAF typical evasive manoeuvre was a hard break left or right, followed by a quarter of a tonneau and then came the dive. This kept the carburettors working as if the plane had a downward gravity force at all times, as in level flight. Later fuel-injected Merlin series cured the problem.

Any WWII fighter pilots who was not at the commands of a nimble, lightweight airplane as the Zero could not fly like Japanese pilots could. Can you imagine a P-47 dogfighting with graceful aerobatics and losing airspeed in no time due to its heavy weight? It would had been certain death for the pilot and WWII would have taken a much different course than it did.

It's amazing how, in CFS2, all negative aspects of any given plane come into play for the human pilot, while all AIs behave almost in the same way, no matter what the aircraft model involved. I tested once a B-29 with the aircraft type changed to 1 (=fighter) in her dp file.

Well, the Superfort tried loops and Zero manoeuvres when I attacked her!

In my opinion, finding the key to change all of the above in AI behaviour, reflecting more the type of aircraft flight envelope involved, would be a quantum leap improvement for CFS2!

Cheers!
KH

[bearcat241]

That's interesting, I didn't know that and that would account for the AI sometimes landing and just rolling away.

Having done a lot of work on AI in this scope of landing behavior, i would have to dispute that premise. I've seen them roll "over the hills" even on AI test missions set up specifically for landing improvements with no enemy involved. This particular behavior is always in the air file. Once i applied the correct tweaks, it stopped. And in combat scenarios with enemies still active on RTB, my improved AI air files still did the job on landing.

In my opinion, finding the key to change all of the above in AI behaviour, reflecting more the type of aircraft flight envelope involved, would be a quantum leap improvement for CFS2!

In some ways its cool the way we keep going over this same old ground for the sake of discussion, but this quantum leap is tied directly to the amount of the "unfinished" AI code we can access and we all know where the keys are. Until we get that in the hands of qualified programmers, its all just more talk and wishful thinking. The incidences noted above of attempted carrier landings by AI (i've witnessed this myself) and the random, "dog-fighting" heavy bombers are good examples of the unfinished behavioral code. Remembering the times of CFS2's rushed release, the eager market and the competitive pressure from other makers at the time, i can see the how and why the code was never completed in the interest of meeting the deadline of that hallowed release date. And now, the keys of the kingdom lie well beyond our reach. Meanwhile, we've done just about every trick we could think of by tweaking flight dynamics, but many behavioral things still remain locked away in the AI control modules.

But hey, let's keep the discussions coming, and let's hope the gatekeepers are eavesdropping, even after 13 years of market life

[codeseven]

Having done a lot of work on AI in this scope of landing behavior, i would have to dispute that premise. I've seen them roll "over the hills" even on AI test missions set up specifically for landing improvements with no enemy involved. This particular behavior is always in the air file. Once i applied the correct tweaks, it stopped. And in combat scenarios with enemies still active on RTB, my improved AI air files still did the job on landing.

Hi Bearcat.

It's been quit awhile since messing around with them but I know could consistently get the AI to do carrier landings and slow almost to a stop but then that last few feet they would make that turn to the right and roll off the deck. For the heck of it I even had a carrier set up to sudden make a sharp turn to starboard to try to keep an AI on deck, but to no avail. I remember thinking back then, if I could only get them to remain in a straight line when landing I could keep them on deck. The best scenario would of coarse be that they had a tail hook function but I know that's all in the original codeing.

Could you explain what particular air file tweaks you applied for landing improvements? Thanks

I agree that this subject has come up many times through the years, which just shows how frustrating the AI can be. But, yes, I too don't think the AI codeing was ever finished and key functions were left out. It's a bummer to think 'what could have been' had the coders been allowed to finish. I don't know if it's even possible now for someone to decompile, fix and recompile the code for the AI but geez, wouldn't that be great! (imagine the list of requested features).

[Sarg Willy]

Hi Yah BC , how,s it going , You have every write to contest it , Forgive me but , I never intended that to be the only reason Ai refuse to park , That was just a fluke to find , happen to be cycling through chase view and tab , and saw the last E-Ai crash on landing , in Europe , as the player Ai were running to London from Hawkinge the roaches stoped ,
I think you and I both know there must be 300 Me109e airfiles alone out there and they,ve been moded by what 500 members at the out house , I,ll wadger there,s 49 other causes , that creat the same symptom , Have a beer matey ,


Now as for the M$soft , NDS watchmacallits , Rethread the game of CFS-2 in Quad spectrum , a processor for the player , a processor for E-Ai , a processor for P-Ai , a processor for scenery , and we,ll all buy a new computor , and the update as well !!

[bearcat241]

Hey there Sarg...good to here from ya.

Codeseven, the basic tweaks you'll need for good landing behavior on touchdown is (1) increase the brake strength by a good scale, (2) put the tail-wheel or nose-wheel turn radii to 45 degrees or more (i prefer 75 degrees on everything), (3) make sure your flap lift and pitch in the air file are set to '0' (yes, AI are programmed to read flap parameters for landing), and (4) don't set the power scalar in the cfg any higher than 1.1, maybe 1.18. Any higher and you'll have to adjust the brake strength even more.

[codeseven]

Hey there Sarg...good to here from ya.

Codeseven, the basic tweaks you'll need for good landing behavior on touchdown is (1) increase the brake strength by a good scale, (2) put the tail-wheel or nose-wheel turn radii to 45 degrees or more (i prefer 75 degrees on everything), (3) make sure your flap lift and pitch in the air file are set to '0' (yes, AI are programmed to read flap parameters for landing), and (4) don't set the power scalar in the cfg any higher than 1.1, maybe 1.18. Any higher and you'll have to adjust the brake strength even more.

Thanks! I'll give it a try.

[Allen]

With my test geting the AI to stop after it lands has to do with the Aircraft.cfg [contact_points] more than any thing you can do in the airfile. Changing the landing gear compression and/or damping will make the AI take longer to stop or be unable to stop. Changing the tail-wheel turn radius to from the stock 180 degrees will make the AI take longer to turn off of the runway and park.

The best thing I have found to do is use the compression and damping from a stock aircraft that is close to the weight of the aircraft you are building/importing. As for the tail-wheel aircraft if they are going to be used in places that have lots of open space you can change the radius to from the stock 180 degrees but if you going to be in the PTO with its small runways it best to stick with the stock 180.