PDA

View Full Version : maybe AI DO overheat, sort of?



lowfighter
05-27-2007, 09:19 PM
Yesterday I had a quick fight in a Corsair against an A6M5a AI. The fight went for some time and after a while I had the zero on my tail. He shot at me, I was low and pretty slow, couldn't do much than do the turning cheat, so he kept shooting but suddenly he disengaged. So I thought he must have run out of ammo hehe. So I turned and spotted him flying straight (or climbing a bit, I don't remember exactly). He was flying very slow and I gain fast on him and starting playing cat&mouse with him, rather carelessly http://forums.ubi.com/images/smilies/35.gif. He was doing mild defensive each time I was closing on him. After a couple of this catandmouse suddenly he did a violent maneuver and was on my tail and start shooting. So he still had ammo! But then why did he disengage? So I thought that maybe some sort of very simple overheat model is present with the AI, even if the temperature gauges are not showing any increase while on autopilot. Like after a predefined (long) time the AI's are disengaging and fly "cooling" the engine?
A similar disengagement happened to me some days ago with a Bf109G10 which was trying to climb away from me(P47). After a while, at some 9000 m he suddenly started losing speed and mildly descending. He was easy to catch then. But I thought he disengaged because he run low on fuel (I have no idea how long you can drive a BFG10 at maximum power).
Well, what do you think? Any way to test this in a more controled way?

JtD
05-27-2007, 11:40 PM
I know that in the early days the AI did just that - cooldown breaks. Every now and then they would just disengage, throttle back and come back to combat after a while.

But AI does not overheat, you engine temp isn't even raising when you have the autopilot fly.

lowfighter
05-28-2007, 01:32 AM
Originally posted by JtD:
I know that in the early days the AI did just that - cooldown breaks. Every now and then they would just disengage, throttle back and come back to combat after a while.

But AI does not overheat, you engine temp isn't even raising when you have the autopilot fly.

Yes, clearly there's no raise in temperature indicated on Autopilot, that's why I was thinking it's a very simple "overheat model", the AI flies at maximum settings for a certain time in combat then it disengages and reduce power. I'm still thinking about a test...
JtD,as you said, they have those breaks in the old days, I remember that too. But I think I saw this disengagement, aside from my 2 examples, several times recently, only I didn't pay that much attention. I think it was happening when the fight was long enough.
But I'm also curious if there's another "logical"cause possible for AI disengaging? Hehe, perhaps they just get bored http://forums.ubi.com/groupee_common/emoticons/icon_wink.gif

lowfighter
05-28-2007, 02:52 AM
All right, I did the following test:
I fly a zero agains 16 enemy zeros. I turn off vulnerability (to allow fight go on long enough) and turn off limited ammo (to prevent my autopilot disengage for luck off ammo). As mission starts I go on autopilot and increase play speed to 4x or 8x. The result is negative, my autopilot will fight forever (well I stopped the mission after 40 minutes). Checked with some other plane type too, the same result.
So there must be something else which drives them disengage sometimes (besides the ammo out and/or fuel out, both ruled out in my first example with the A6M5). Hmmm...

Some other things I noticed letting the AI pilot the plane for this long time in combat:
-he would keep the throttle to maximum or close to maximum most of the time but NOT ALL the time.
-sometimes, very rarely, I see redout but never a blackout.

Badsight-
05-28-2007, 03:31 AM
AI on its own

your plane under Autopilot

those are 2 different things , nothing definante has come from Maddox Games to say they are the same thing under the same rules

LEXX_Luthor
05-28-2007, 03:45 AM
AI behavior such as this should be documented -- for one thing, to aid mission sculptors. Having to conduct experiments like these is sad, unless sim devs everywhere are hiding rather poor and slapped together game designs. I doubt that is 100% the case as the AI in various sims have some excellent characteristics. The StrikeFighters series has AI that brilliantly excels at terrain avoidance at extreme low above ground altitudes, and depends on aircraft performance (compare with PF AI crashing in similar settings). But the StrikeFighters AI behavior is undocumented for users as well. Its up to the customer to figure out that AI doesn't use radar -- its a purely visual range dogfight and ground pounding sim, but there is a "hack" that allows BVR AI engagements ... by setting visual range of aircraft to very high distances...but that just makes BVR engagements technically VR engagments in the game.

woofiedog
05-28-2007, 03:51 AM
Very interesting... as I fly offline. Most of the time they run out of ammo blasting away at you or just don't like the flying when it down low... must be the Dust getting into their Eye's.

It could be that chasing the Corsair was warming up the old Sakae 21 radial! Wicked http://forums.ubi.com/images/smilies/winky.gif

lowfighter
05-28-2007, 04:24 AM
Originally posted by Badsight-:
AI on its own

your plane under Autopilot

those are 2 different things , nothing definante has come from Maddox Games to say they are the same thing under the same rules

Oh yes, that's a good point! I was ASSUMING there's no difference between a pure AI and my AI autopilot. You can't do more than that if you want to investigate the AI behaviour. Well you can watch them from externals but that's not much. So yes it's an assumption and it might be wrong.
Now, I went back and tried to reproduce the engagement with the zero. This time the idea was: not shooting at it just try to stay on his 6 as much as possible and see what's happening. So I took a hellcat this time. I stayed on his six as much as I could and he never managed to get on my 6. But he suddelnly disengages, and start to fly again in a pasive way. I play again with him, a bit of boom and zoom (but no shooting). Repeated several times and this time coming down on him I notice that his engine has stopped! I ctrl-F2 and the zero has the prop bent! What's up? Took a screenie but that's not good enough. So I tried once again and recorded track. Again after some time he stoped fighting, but this time he was pretty low, when he did a stupid thing, he rolled a bit and dived into the ground, rubbish rubbish rubbish.
Conclusion: never chase an AI too long because he'll start doing weird things. I'll keep doing a couple more and record them, hope to catch the bent prop AI effect again LOL.

lowfighter
05-28-2007, 07:22 AM
Little update, I'm fascinated. The AI! Artificial Inteligence... Did the same test, chasing a zero, 3-4 times, tracks recorded. It always ended with the same result: AI crashed into ground. Same pattern:
-he gives up the fight and after a while:
-he crashes in about the same way: little roll and quick dive into the ground. Without being really in danger.
It's like they get drunk or fall asleep after a while.

CD_kp84yb
05-28-2007, 07:36 AM
Artificial Inteligence, well skip the Inteligence part

cheers

lowfighter
05-28-2007, 07:50 AM
Originally posted by CD_kp84yb:
Artificial Inteligence, well skip the Inteligence part

cheers

Ahaha, I'd love to discuss your comment but it would deviate this very narrow (and still pretty hazy) thread too much. I suggest we can start a thread with a title like "Any ARTIFICIAL intelligence in the il2 AI behaviour". That would be really really interesting to discuss. http://forums.ubi.com/groupee_common/emoticons/icon_biggrin.gif

Viper2005_
05-28-2007, 08:23 AM
Accelerated time doesn't necessarily make everything happen faster.

For example, last time I checked, you can't open the throttles in a 262 twice as fast at 2x speed, which is rather strange.

I have a feeling that many if not all of the engine related "counters" run in realtime irrespective of time acceleration setting, though I have never spent the time to conclusively test this. Anyway, this could be important in this context, so I suggest that you test it before using a methodology based upon the use of time acceleration.

lowfighter
05-28-2007, 08:36 AM
Originally posted by Viper2005_:
Accelerated time doesn't necessarily make everything happen faster.

For example, last time I checked, you can't open the throttles in a 262 twice as fast at 2x speed, which is rather strange.

I have a feeling that many if not all of the engine related "counters" run in realtime irrespective of time acceleration setting, though I have never spent the time to conclusively test this. Anyway, this could be important in this context, so I suggest that you test it before using a methodology based upon the use of time acceleration.

Hey Viper, I used time acceleration just for saving my time (40minutes normal time judging from the cockpit clock), assuming that everything will happen as normal only it will run faster. But you have a very good point here at least about the human controlled.

So I can surely run it again, start recording and go for a (quick) beer or something. Thanks! http://forums.ubi.com/images/smilies/25.gif

lowfighter
05-28-2007, 11:03 AM
All rigth, did once again the zero Autopilot against 16 AI zeros. Autopilot+invulnerable + unlimited ammonition.
Test1: hit autopilot and start recording and go for a beer. Result: fight lasted less than 15 minutes. Autopilot shot down only 1 enemy plane. Behaviour: pretty defensive from the begining, he started attacking only after most of the enemy were... dead. Yes, although he shot down only one, while he was chased some 8-9 of them died by collisions or by hiting the ground.
Test 2: repeated but again with 8 x time acceleration. Result: the Autopilot after a while managed to land on the grass!? Other details I miss but I have the track, I'll look how and why he crashlanded.
Test3: exactly as test 2 but started autopilot a bit earlier (perhaps 2-3 seconds) than in test 3. Result: fight about 50 minutes judging from the cockpit clock. At the end 11! enemies shot down (compare with test1).

Why did I do 2 tests at 8x time acceleration: because I noticed many times that if you start a mission and hit autopilot the outcome might depend on 1-2 seconds of delay hitting autopilot. So to say it depends on the initial conditions, but it's interesting that if you have a small initial condition difference due to 1-2 seconds delay in hiting autopilot, you might end with a pretty different result at the end of the mission...Cool stuff!
Bah, it begins to show to many variables...

wintergoose
05-28-2007, 12:31 PM
I have seen this behavier of the AI often, I use it if I have a difficult situation I waits for the AI to disengade and then I got them much easier.
It takes a time before they disangade but soner or later they did.
I have not reflected why, thought they was programmed to give up after a while and go for homebace.
That was what they did in the real time.
In real situations the opnonents disangaded much sooner then the AI does in the game.

woofiedog
05-28-2007, 12:36 PM
Quote... Bah, it begins to show to many variables...

http://forums.ubi.com/groupee_common/emoticons/icon_biggrin.gif Nice work!

DuxCorvan
05-28-2007, 12:57 PM
AI doesn't overheat. They just break off when their flight leader gets too far from fight and orders a new waypoint or RTB, and they do it regardless the tactical situation, advantage, danger, etc.

Human pilots -non leaders- don't give a d*mn about their AI leader shouting not to abandon formation, or their gunners protesting "what are you doing!", specially when we are about to have a kill or we are chased, but AI just follows orders as a sheep.

And I don't need to prove it: I say it, and I'm always right, uh? http://forums.ubi.com/images/smilies/354.gif

lowfighter
05-28-2007, 09:13 PM
Originally posted by wintergoose:
It takes a time before they disangade but soner or later they did.
I have not reflected why, thought they was programmed to give up after a while and go for homebace.
That was what they did in the real time.
In real situations the opnonents disangaded much sooner then the AI does in the game.

Probably you're right. As I was doing this 1 vs 1 fights just sitting on their 6, I noticed that sometimes they give up fight sooner than othertimes. And it looks like the tighter you stick to them the sooner they disengage. I'll do one more test, place myself in a fast plane, and have a not-that-fast AI enemy chase me. See if and when he gives up chasing me.


EDIT: not to forget there are temporary disengagements which are perfectly logical. Like when a FW AI is turning with a spitfire, he will diesngage when his speed has decreased too much in the turn. But he's not giving up the fight, just disengage and start gaining altitude and speed. But my curiosity is related to the cases when the AI gives up the fight although they have ammo and fuel enough.

DKoor
05-28-2007, 11:09 PM
I have several tracks Vs Ai.... they never overheat, so that is out of question, that has been tested to death.

Ai will sometimes, if you get enough separation from them, simply disengage and follow the waypoints.

They will do that even in the mid of the combat, if you have damaged him... after some amount of damage Ai simply tries to rtb.

lowfighter
05-29-2007, 02:10 AM
Originally posted by DKoor:
I have several tracks Vs Ai.... they never overheat, so that is out of question, that has been tested to death.

Ai will sometimes, if you get enough separation from them, simply disengage and follow the waypoints.

They will do that even in the mid of the combat, if you have damaged him... after some amount of damage Ai simply tries to rtb.

I wouldn't be so sure, Koor please read the other posts too.
I did another record this time chasing a Bf109F4, from looking at the track one thing is sure the AI radiator flaps are changing position in time. Also around 4 minutes from the start he gives up the fight but I don't understand very well what is he doing at that time with the flaps and radiator, looks like flaps are down and radiator is pretty open...he drops speed...hm some time after that I lose him and he starts fighting again but to me it looks like he's not putting all effort into it, he must be tired or something. But anyway that moment around 4 minutes is very weird to me, if you want to see it just PM me. I'd like to hear some oppinions about what is he doing and why is he doing it.

DKoor
05-29-2007, 02:19 AM
I can give you tracks that are 1 hour long. Ai doesn't overheat. We may discuss something but this - no.

If you want to assure yourself in this fact.... well go play in QMB.
I played vs them for hundreds of hours in QMB.

Not only that..... not only they don't overheat but they do not suffer any damage from any kind of wrong CEM.

If you have saw any of the tracks I post in every other similar thread you'd know what I'm talking about here. I put 109 in sharp dive on 100% pp and then set it on autopilot.
He doesn't overheat, he achieves 950km/h on deck, his elevator works good, his prop achieves mach number (you can't even hear the engine running anymore, so high frequency).

A6M on the other hand achieves also above 900km/h on ~1,5k, he runs forever on wrong fuel mixture at alt etc etc.

Ai doesn't have anything to do with how player aircraft is modeled and that is - a fact.

Due to these facts, you cannot fight Ai in any kind of historical way..... the only way to engage them is to either outturn them which is easy even in 190 (player) Vs Spit fights or if you have enormous energy advantage.

lowfighter
05-29-2007, 02:25 AM
Originally posted by DKoor:
I can give you tracks that are 1 hour long. Ai doesn't overheat. We may discuss something but this - no.

If you want to assure yourself in this fact.... well go play in QMB.
I played vs them for hundreds of hours in QMB.

Not only that..... not only they don't overheat but they do not suffer any damage from any kind of wrong CEM.

If you have saw any of the tracks I post in every other similar thread you'd know what I'm talking about here. I put 109 in sharp dive on 100% pp and then set it on autopilot.
He doesn't overheat, he achieves 950km/h on deck, his elevator works good, his prop achieves mach number (you can't even hear the engine running anymore, so high frequency).

A6M on the other hand achieves also above 900km/h on ~1,5k, he runs forever on wrong fuel mixture at alt etc etc.

Ai doesn't have anything to do with how player aircraft is modeled and that is - a fact.

No prob if you don't want to discuss, this thread is for people who want to discuss. http://forums.ubi.com/groupee_common/emoticons/icon_smile.gif

DKoor
05-29-2007, 02:30 AM
Originally posted by lowfighter:
I did another record this time chasing a Bf109F4, from looking at the track If you are reffering to the NRTK - then RoFL.... they are not accurate.... go ahead test it for yourself - run 109 on closed rad for 5 mins and record NTRK then review it you'll see how rads are opening on track after a while.


one thing is sure the AI radiator flaps are changing position in time. Just this... doesn't really mean anything doesn't it? I mean what can we conclude about this fact? Even if we drop all the NTRK bug issues....


Also around 4 minutes from the start he gives up the fight but I don't understand very well what is he doing at that time with the flaps and radiator, looks like flaps are down and radiator is pretty open...he drops speed... I saw tracks with Ai and player fighters flying with their gears down (of course they did not have gears down in game).


hm some time after that I lose him and he starts fighting again but to me it looks like he's not putting all effort into it, he must be tired or something. But anyway that moment around 4 minutes is very weird to me, if you want to see it just PM me. I'd like to hear some oppinions about what is he doing and why is he doing it. Ai is weird... but after you put hundreds of hours observing their behavior you'll see that they are predictable since they do the same all the time. Same routines.

To clear up many issues you can run the game on arcade so you'll see little clouds above Ai ac saying "I'm on fire" "I'm RTB" "Gunner down" and that'll clear up many issues........

DKoor
05-29-2007, 02:33 AM
Originally posted by lowfighter:
No prob if you don't want to discuss, this thread is for people who want to discuss. http://forums.ubi.com/groupee_common/emoticons/icon_smile.gif I have a question for you.

I gave you a definite answer.
So what is left to discuss?

DKoor
05-29-2007, 03:05 AM
Originally posted by lowfighter:
Yesterday I had a quick fight in a Corsair against an A6M5a AI. The fight went for some time and after a while I had the zero on my tail. He shot at me, I was low and pretty slow, couldn't do much than do the turning cheat, so he kept shooting but suddenly he disengaged. He probably did not disengaged but you have outturned him so he either tried to outrun or outclimb you, but since you have vastly faster ac you could easily catch up with him.


So I thought he must have run out of ammo hehe. It's a possibility that Ai disengages due to low ammo, but in case of A6M5a unlikely since it has 1000 LMG rounds per one nose gun.


So I turned and spotted him flying straight (or climbing a bit, I don't remember exactly). He was flying very slow and I gain fast on him and starting playing cat&mouse with him, rather carelessly http://forums.ubi.com/images/smilies/35.gif. So there....


He was doing mild defensive each time I was closing on him. Depending on skill and altitude.... Zero will put very poor display on 1 Vs 1 deck fights even on Ace Ai skill, take him on higher alts 1-3k and he'll be better since it manages his energy very well. On higher alts they're **** again due to superchargers, US crates being vastly superior etc. etc.


After a couple of this catandmouse suddenly he did a violent maneuver and was on my tail and start shooting. That happens only if you get dragged into scissors or get too close to him.


So he still had ammo! But then why did he disengage? Because you outtunred him. That will happen in any player Vs Ai fighter combat.... you may try it 109 Vs Ai I-16 or so...


So I thought that maybe some sort of very simple overheat model is present with the AI, No.


even if the temperature gauges are not showing any increase while on autopilot. Like after a predefined (long) time the AI's are disengaging and fly "cooling" the engine? No.... that will happen every time you outturn Ai on deck. It will try to outclimb or outrun you. Then sometimes when you close it'll turn and sometimes it wont and will do that crazy rolls until you fill it with lead.


A similar disengagement happened to me some days ago with a Bf109G10 which was trying to climb away from me(P47). After a while, at some 9000 m he suddenly started losing speed and mildly descending. Ai when reached certain alt, simply is like that.... I followed G2 in Yak-1B from deck to around 5,500m and then I gained on him. That is of course if he didn't tried to rtb due to low fuel.


He was easy to catch then. But I thought he disengaged because he run low on fuel (I have no idea how long you can drive a BFG10 at maximum power). Fuel will last very short in 109 full power climb.


Well, what do you think? Any way to test this in a more controled way? I hope this little review of mine helps a bit, since we couldn't get on the track before http://forums.ubi.com/groupee_common/emoticons/icon_biggrin.gif

But to understand better what is happening to Ai, enable arcade mode. Then you'll know 100% when they are really disengaging from the fight.

avimimus
05-29-2007, 05:12 PM
Ahm...The autopilot and the AI are two different things (even though the AI seems to be "on autopilot" sometimes and the autopilot sometimes shoots).

I also noticed that changes in throttle settings don't damage jet engines during the first few seconds of flight. Does this mean that the AI can change throttle on the 004Bs anytime it likes?

lowfighter
05-29-2007, 09:03 PM
Originally posted by DKoor:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by lowfighter:
I did another record this time chasing a Bf109F4, from looking at the track If you are reffering to the NRTK - then RoFL.... they are not accurate.... go ahead test it for yourself - run 109 on closed rad for 5 mins and record NTRK then review it you'll see how rads are opening on track after a while.


one thing is sure the AI radiator flaps are changing position in time. Just this... doesn't really mean anything doesn't it? I mean what can we conclude about this fact? Even if we drop all the NTRK bug issues....


Also around 4 minutes from the start he gives up the fight but I don't understand very well what is he doing at that time with the flaps and radiator, looks like flaps are down and radiator is pretty open...he drops speed... I saw tracks with Ai and player fighters flying with their gears down (of course they did not have gears down in game).


hm some time after that I lose him and he starts fighting again but to me it looks like he's not putting all effort into it, he must be tired or something. But anyway that moment around 4 minutes is very weird to me, if you want to see it just PM me. I'd like to hear some oppinions about what is he doing and why is he doing it. Ai is weird... but after you put hundreds of hours observing their behavior you'll see that they are predictable since they do the same all the time. Same routines.

To clear up many issues you can run the game on arcade so you'll see little clouds above Ai ac saying "I'm on fire" "I'm RTB" "Gunner down" and that'll clear up many issues........ </div></BLOCKQUOTE>


Well DURING the test itself I noticed at around 4 minutes he had the flaps down, also radiator was open somewhat but of course it's not posible in flight to see how much but surely it was not closed. The track shows what I saw when playing: flaps down and rad somewhat open. If I would have seen something which was really different I would have had doubts offering the track for inspection http://forums.ubi.com/groupee_common/emoticons/icon_smile.gif

About the track showing the rad opening although it shouldn't, yes I have seen it too. So perhaps you have a point about AI rad too. Only one thing might be done to check: during the fight pause the game and take screenshots of the AI. It's not perfect but I could still distinguish between radiator positions.

lowfighter
05-29-2007, 09:36 PM
Originally posted by DKoor:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">Originally posted by lowfighter:
Yesterday I had a quick fight in a Corsair against an A6M5a AI. The fight went for some time and after a while I had the zero on my tail. He shot at me, I was low and pretty slow, couldn't do much than do the turning cheat, so he kept shooting but suddenly he disengaged. He probably did not disengaged but you have outturned him so he either tried to outrun or outclimb you, but since you have vastly faster ac you could easily catch up with him.


So I thought he must have run out of ammo hehe. It's a possibility that Ai disengages due to low ammo, but in case of A6M5a unlikely since it has 1000 LMG rounds per one nose gun.


So I turned and spotted him flying straight (or climbing a bit, I don't remember exactly). He was flying very slow and I gain fast on him and starting playing cat&mouse with him, rather carelessly http://forums.ubi.com/images/smilies/35.gif. So there....


He was doing mild defensive each time I was closing on him. Depending on skill and altitude.... Zero will put very poor display on 1 Vs 1 deck fights even on Ace Ai skill, take him on higher alts 1-3k and he'll be better since it manages his energy very well. On higher alts they're **** again due to superchargers, US crates being vastly superior etc. etc.


After a couple of this catandmouse suddenly he did a violent maneuver and was on my tail and start shooting. That happens only if you get dragged into scissors or get too close to him.


So he still had ammo! But then why did he disengage? Because you outtunred him. That will happen in any player Vs Ai fighter combat.... you may try it 109 Vs Ai I-16 or so...


So I thought that maybe some sort of very simple overheat model is present with the AI, No.


even if the temperature gauges are not showing any increase while on autopilot. Like after a predefined (long) time the AI's are disengaging and fly "cooling" the engine? No.... that will happen every time you outturn Ai on deck. It will try to outclimb or outrun you. Then sometimes when you close it'll turn and sometimes it wont and will do that crazy rolls until you fill it with lead.


A similar disengagement happened to me some days ago with a Bf109G10 which was trying to climb away from me(P47). After a while, at some 9000 m he suddenly started losing speed and mildly descending. Ai when reached certain alt, simply is like that.... I followed G2 in Yak-1B from deck to around 5,500m and then I gained on him. That is of course if he didn't tried to rtb due to low fuel.


He was easy to catch then. But I thought he disengaged because he run low on fuel (I have no idea how long you can drive a BFG10 at maximum power). Fuel will last very short in 109 full power climb.


Well, what do you think? Any way to test this in a more controled way? I hope this little review of mine helps a bit, since we couldn't get on the track before http://forums.ubi.com/groupee_common/emoticons/icon_biggrin.gif

But to understand better what is happening to Ai, enable arcade mode. Then you'll know 100% when they are really disengaging from the fight. </div></BLOCKQUOTE>


No I don't think I outturn it. I was turning very softly such that I don't loose speed which was about 400 Km/h.
After he disengaged he was flying very slowly and you notice it that he's not flying it at maximum power. The Corsair is fast but that zero after disengaging was slow and didn't show any intention to remedy that.
Sorry Koor, did the same thing now already 6-7 times and the slowdown after he gives up the fight is always present. You don't need my tracks for that, it's reproducible 100% and you can do the experiment yuorself. I did the experiments with AI set to average such that it was easy to sit on his 6 most of the time. Might be a thing to try it with other skills though...
Again Koor there are different disengagements, some which you mentioned above, one which I mentioned even more above, this are tactical disengagements, and I'm talking about something which has no tactical value: you shoot the guy after this type of disengagement all the time because he's slow and doesn't do anything or much about it, and he's not ....
Look, do the experiment, if you care, a couple of times, and then post here what you see. It doesn't take long, it's a pretty quick thing, less than 10 minutes anyway...
And if you want that I respond anymore to your posts please drop the arrogance and the big words.

Philipscdrw
05-30-2007, 08:11 AM
I dislike fighting the AI because they do two things: they enter an endless climbing spiral when you chase them and it seems impossible to catch them, and they cannot hit you when you're turning because none of them have mastered deflection shooting. And then they switch into RTB mode and ignore your fire. And I have trouble outrunning them (haven't flown much recently, can't remember how true this is). Hence, flying online. But online the enemy don't behave anything like the AI and I cannot compete due to unfamiliarity!

DKoor
05-31-2007, 06:05 PM
Originally posted by lowfighter:
No I don't think I outturn it. I was turning very softly such that I don't loose speed which was about 400 Km/h. OK


After he disengaged he was flying very slowly and you notice it that he's not flying it at maximum power. The Corsair is fast but that zero after disengaging was slow and didn't show any intention to remedy that. OK


Sorry Koor, did the same thing now already 6-7 times and the slowdown after he gives up the fight is always present. You don't need my tracks for that, it's reproducible 100% and you can do the experiment yuorself. OK.... but it'd still be nice if we could see a few of your TRK's.


I did the experiments with AI set to average such that it was easy to sit on his 6 most of the time. Might be a thing to try it with other skills though... Sure because difference is great in two regards especially - shooting and energy management.


Again Koor there are different disengagements, some which you mentioned above, one which I mentioned even more above, this are tactical disengagements, and I'm talking about something which has no tactical value: you shoot the guy after this type of disengagement all the time because he's slow and doesn't do anything or much about it, and he's not .... ....some of which may have everything to do with....


Look, do the experiment, if you care, a couple of times, and then post here what you see. ....various diff Ai skill settings, which you haven't tested (apart from the average).


It doesn't take long, it's a pretty quick thing, less than 10 minutes anyway... OK


And if you want that I respond anymore to your posts please drop the arrogance and the big words. Then please... don't reply to my posts anymore if you find them to offend you in some way!

DKoor
05-31-2007, 06:48 PM
Originally posted by avimimus:
autopilot and the AI are two different things 1.Any chance that we could see a proof of that?
2.If Autopilot isn't the Ai.... then what is it?

lowfighter
05-31-2007, 08:48 PM
Koor, if you PM me I could send you tracks. Or I'll see if I can upload later on M4T.
Coffee time in Romania.

lowfighter
06-01-2007, 01:49 AM
Ok, uploaded 3 tracks at M4T

http://mission4today.com/index.php?name=Downloads&file=details&id=2468

After several plays of first track (A6M5_AI_1)I think it doesn't show any energywise "suspicious" behaviour. But it shows the ground crash I said I noticed in al my trials.
With second and third track it beginns to happen somewhere around 4 minutes from start.
I hope I'll be uploading some more tracks soon with the same plane type for both AI and player and also on a flat landscape map. I already did such a Bf109F4 trial some 4 times without recording.
I've seen once again (in only one of the trials) the flaps drop which is present in the Bf109 track. But in the last unrecorded trials I changed a bit the approach, gave the AI some more freedom whenever I thought he was too close to the suicidal moment. So with that he reactivated several times so I could go on with the trial and notice behaviour for 10-15 minutes. But once the power drop or whatever it is occurs even after reactivating the AI has a mild fighting behaviour.

The pure AI versus autopilot AI question. Possible to set some trials to check that to some extent?
How could we test if the autopilot and pure AI have the same behaviour (energywise, etc)?

DuxCorvan
06-01-2007, 02:36 AM
C'mon don't make your life harder without reason.

Occam's Razor (http://math.ucr.edu/home/baez/physics/General/occam.html) says:

If you press 'A', AI just takes control of your plane. Why would they make it a different 'being'?

msalama
06-01-2007, 06:18 AM
Why would they make it a different 'being'?

+1

No reason, because why would you implement _TWO_ systems with essentially the same functionality?

Hmmm... on the other hand... suppose that, say, Airbus or Boeing is actually using the IL-2 AP / AI code base for their flight directors... would explain the need for redundancy... hmmm...

http://forums.ubi.com/images/smilies/winky.gif http://forums.ubi.com/images/smilies/88.gif http://forums.ubi.com/images/smilies/35.gif OK, getting me coat then.

lowfighter
06-01-2007, 06:42 AM
Originally posted by DuxCorvan:
C'mon don't make your life harder without reason.

Occam's Razor (http://math.ucr.edu/home/baez/physics/General/occam.html) says:

If you press 'A', AI just takes control of your plane. Why would they make it a different 'being'?

Yes, it's very very reasonable to assume they behave the same. But it's good to keep in mind that it might be otherwise, to keep in mind it's an assumption, a very reasonable assumption. But only the developers know for sure, and at least differences must be anyway small and/or manifest only under certain conditions...
How about a little test? Set a fight between a pure AI and the player, starting head on, same altitude, same speed, same skill set in FMB? Hit autopilot when the mission starts. Repeat with different AC types (but same type in each mission).
One problem with such a test is that the time when you hit autopilot (even if it's a matter of less than a second) might matter to some extent to the outcome. It's a funny "system" such that small initial conditions differences can bring very different final results.
I don't know but it's fun to discuss...

msalama
06-01-2007, 07:22 AM
(fumbles for a coathanger)

OK chaps, me again. Will try my hand on Lowfighter's test later today. Will publish tracks too.

The end criteria is us - or, rather, them - fighting 'til either one is down, I take it?

DKoor
06-01-2007, 08:18 AM
Originally posted by lowfighter:
Koor, if you PM me I could send you tracks. Or I'll see if I can upload later on M4T.
Coffee time in Romania. All right........ I'll see them later http://forums.ubi.com/groupee_common/emoticons/icon_smile.gif

msalama
06-01-2007, 11:28 AM
Guys,

It doesn't seem to work. A short report follows:

------------------------------------------------

TESTING CONDITIONS

MAP: Smolensk
WEATHER: Clear
ALTITUDE: 1000m. AGL
AC: 2 x Buffalo Mk.I
AI / AP SKILL: 1) Average 2) Veteran 3) Ace
NOTES: Track recording and AP put on immediately at the beginning.

RESULTS:

1) Average AP / AI

Planes made several h2h attacks w/out firing. Test interrupted after 3 or 4 such passes.

2) Veteran AP / AI

The AP plane shot its AI opponent down on the first pass. Enemy crashed at approx. 20 secs. into the mission.

3) Ace AP / AI

The AI plane shot its AP opponent down on the first pass. Friendly crashed at approx. 20 secs. into the mission.

CONCLUSION:

Either the testing situation is invalid, or, perhaps, the tester.

------------------------------------------------

OK, what am I doing wrong?

lowfighter
06-01-2007, 12:52 PM
Hey M8, you're not doing anything wrong http://forums.ubi.com/groupee_common/emoticons/icon_smile.gif
But the veteran and ace tests are not relevant because they just don't fly and fight enough. The idea is to see rather if AP or AI is getting an advantage after some time, so the shooting headon habits of the veterans and aces simply prevent us seeing this. But the average don't shot at head on so we can observe them longer. You said you stopped the average performance after 3 passes. I did a similar test and let them play for 10 minutes. During these 10 minutes the same pattern would repeat periodically, with clockwise precision, same turn, same head on, same turn same head on....
So this test shows clearly that at average level the autopilot and the pure AI behave identically. Neither of them would turn more tightly and so on. One interesting thing: the autopilot is not pushing the plane too much, the throttle looks most of the time close to maximum but not exactly.

buddye1
06-01-2007, 01:22 PM
hi guys, alot of knowledge here on the IL2 AI.

I am the BOBII AI coder for about two years and I have been flying 1946 for about one month studing the AI. This thread is very interesting to me.

The designs must be very close but I guess different. In BOBII when the player engages auto pilot he in effect becomes an AI and he really has no control of his A/C while in AP.

msalama
06-01-2007, 02:03 PM
Rgr Lowfighter, got it. Will get back to the tests in a day or two - an interesting topic this. S!

msalama
06-01-2007, 02:17 PM
In BOBII when the player engages auto pilot he in effect becomes an AI and he really has no control of his A/C while in AP.

Hi Buddye.

Hmmm... I'm pretty certain this is the case in IL-2 as well - nothing else kind of makes sense (I'm an ex-programmer / program designer myself so thinking, eh, "logical" here). Have to run some tests first to get a clear idea of the thingamajig, though, because what little flying I've done lately has mostly been online!

But on a related matter, provided that you don't mind my asking: now you're the man as regards AI as we know - and anyone not believing me go & check out BoBII's AI, just f*n' excellent - so how would you rate IL-2's AI yourself now that you have IL-2 1946? Just asking out of curiosity, mind, because you if anyone certainly knows this stuff waaaaay better than us garden-variety armchair experts around here http://forums.ubi.com/images/smilies/16x16_smiley-wink.gif http://forums.ubi.com/images/smilies/16x16_smiley-very-happy.gif

PS / EDIT / CLARIFICATION:


he in effect becomes an AI and he really has no control of his A/C while in AP.

Oh, definitely the basic situation in IL-2 as well. Just FYI.

DKoor
06-01-2007, 02:32 PM
I watched the tracks... yes that is typical Ai behavior.
If we take into account what has been said... if autopilot is nothing more than another John Doe IL-2 Sturmo 46 Ai then one thing is clear in the whole story - Ai doesn't suffer the damage from overheat!

It's very easy test (to do that) - run ac to max throttle settings (+WEP if available) and when you overheat run it for some more. Notice the rad temp it reaches over 120?C or something.

Then switch to autopilot.

Then another interesting thing occurs (which I'll test more), when autopilot is engaged "overheat" message disappears. But......... rad temp is not lower!!! Not only that but Ai flies around on that (overheated) rad temp all the time! Suffice is to say - no damage.

buddye1
06-01-2007, 03:10 PM
Originally posted by msalama:
<BLOCKQUOTE class="ip-ubbcode-quote"><div class="ip-ubbcode-quote-title">quote:</div><div class="ip-ubbcode-quote-content">In BOBII when the player engages auto pilot he in effect becomes an AI and he really has no control of his A/C while in AP.

Hi Buddye.

Hmmm... I'm pretty certain this is the case in IL-2 as well - nothing else kind of makes sense (I'm an ex-programmer / program designer myself so thinking, eh, "logical" here). Have to run some tests first to get a clear idea of the thingamajig, though, because what little flying I've done lately has mostly been online!

But on a related matter, provided that you don't mind my asking: now you're the man as regards AI as we know - and anyone not believing me go & check out BoBII's AI, just f*n' excellent - so how would you rate IL-2's AI yourself now that you have IL-2 1946? Just asking out of curiosity, mind, because you if anyone certainly knows this stuff waaaaay better than us garden-variety armchair experts around here http://forums.ubi.com/images/smilies/16x16_smiley-wink.gif http://forums.ubi.com/images/smilies/16x16_smiley-very-happy.gif

PS / EDIT / CLARIFICATION:


he in effect becomes an AI and he really has no control of his A/C while in AP.

Oh, definitely the basic situation in IL-2 as well. Just FYI. </div></BLOCKQUOTE>

Hi msalama,

Thanks for the too kind words about the BobII AI. We are just a small group of volunteers who work on BOBII for the love of it. We do not have the skill/manpower/experience of the IL2 coders.

I purchased 1946 to to study and learn as IL2 is such a super game. Do not get talking about AI as I could go on forever about AI maneuvres, selection, performance, shooting, engage, disengage and (opps see now I doing it).

I feel like I am a guest here and I have only been flying 1946 for one month so it would be unfair for me to make observations on this fine game. I also have a policy of never comparing publicly different flight sims. I just try to respect all flight sims as we have so few.

On the topic of disengage. The poster stated that the IL2 "AI disengeges for no reason and then crashes into the ground". I will look for this in 1946. In BOBII disengage is very important as we simulate very large battles (with 200-400) A/C and the RAF is so badly outnumbered. We do not want a blood bath as we want historicly accurate kill rates. So how to correctly disengage and get everyone home after the fight is a challenge.I think this will also be a big challenge for SOW.

lowfighter
06-01-2007, 11:16 PM
Originally posted by DKoor:
I watched the tracks... yes that is typical Ai behavior.
If we take into account what has been said... if autopilot is nothing more than another John Doe IL-2 Sturmo 46 Ai then one thing is clear in the whole story - Ai doesn't suffer the damage from overheat!

It's very easy test (to do that) - run ac to max throttle settings (+WEP if available) and when you overheat run it for some more. Notice the rad temp it reaches over 120?C or something.

Then switch to autopilot.

Then another interesting thing occurs (which I'll test more), when autopilot is engaged "overheat" message disappears. But......... rad temp is not lower!!! Not only that but Ai flies around on that (overheated) rad temp all the time! Suffice is to say - no damage.

Yes, no damage, no temp variation and so on, that's why I said in the beginning a simple overheat model, but my words are misleading a bit, what I wanted to say a simple model which make the AI reduce power as to mimic they do care about their engines. Just think that what we call AI is a routine in a program. There's no temperature parameter affecting AI, I think everybody agrees about that. But that doesn't prevent the programmer to write an if statement saying something like
if fight>5 minutes
reduce power to 90%
end

But we don't see any throttle reduction when on autopilot!
But we assume AI and autopilot work in exactly the same way.
There's one fundamental difference between the pure AI and the autopilot. The autopilot will fight always only against AI. The AI will fight also against humans. The AI is many times placed in very extreme conditions flying agains humans. The autopilot is not facing such extreme flying by far because the AI opposing him is not trying very hard to winn!
Got to say yesterday I set a fight between the player (set in FMB as average) and AI set to veteran. Made the player invulnerable and hit autopilot. The aim was to place the autopilot in a more extreme situation. I was expecting to see after a while the veteran dominating the average autopilot. Guess what: after a couple of minutes it went into a stalemate, the veteran AI and the average autopilot begann to do clockwork periodical headons. Each time the veteran would shoot at headon of course. But then this is hilarious, not to mention that I placed the veteran AI on my autopilot 6 at the beginning of the mission. http://forums.ubi.com/images/smilies/35.gif

lowfighter
06-02-2007, 02:32 AM
But if I set a mission between a Bf109 AI and the Bf109 autopilot I see the radiator flaps of both are opening and closing during fight and i kind of see a very simple rule:
AC speed low then radiator more open,
AC speed high then radiator more closed.
Doesn't this looks like radiator on auto?

Watched at altitudes less than 2000m.
What do you think?

DKoor
06-02-2007, 06:36 AM
I haven't checked that out I've been busy on The Big Top last nite http://forums.ubi.com/images/smilies/35.gif

Anyhow..... 109 may not be the best example since it has auto rad. We could choose ac like FW-190A9 which doesn't have auto rads.
And see if they are opening.

Point of this being perhaps... the fact that Ai may not be smart enough to set rads on anything other than on AUTO or CLOSED.

This is also interesting thing to check out.

If A9 Ai does open rads at some point that would mean that Ai is capable of opening rads (changing rads position) which may shed more light on the whole picture.

lowfighter
06-02-2007, 11:37 AM
Originally posted by DKoor:
I haven't checked that out I've been busy on The Big Top last nite http://forums.ubi.com/images/smilies/35.gif

Anyhow..... 109 may not be the best example since it has auto rad. We could choose ac like FW-190A9 which doesn't have auto rads.
And see if they are opening.

Point of this being perhaps... the fact that Ai may not be smart enough to set rads on anything other than on AUTO or CLOSED.

This is also interesting thing to check out.

If A9 Ai does open rads at some point that would mean that Ai is capable of opening rads (changing rads position) which may shed more light on the whole picture.

That's a very good idea to test! http://forums.ubi.com/images/smilies/25.gif

lowfighter
06-03-2007, 07:32 AM
So I looked at the radiator position for diferent AC flown by AI.
Corsair, zero and P47 which don't have an auto radiator option appear to fly with radiator closed all the time.
P51 which has auto radiator shows changing of radiator position during fight, opening and closing similar to Bf109, but less pronounced I think.
While flying the zero I noticed the following: if I open the radiator and after that I hit autopilot then the radiator remains in the same position (opened) during autopilot fight.
Also a zero AI which shows the rad closed in game shows rad pretty open AND stationary on the corresponding track.
I'm curious, if I open fully the rad on a zero then hit autopilot, visually I see the rad fully open afterwards. But is it affecting the performance of the autopiloted plane which seems to fly with rad fully open?

DKoor
06-03-2007, 08:44 AM
Originally posted by lowfighter:
While flying the zero I noticed the following: if I open the radiator and after that I hit autopilot then the radiator remains in the same position (opened) during autopilot fight. That's interesting. What would be good to know (but as I see it, difficult to test) is to observe if autopilot loses some speed becasue of closed rad.


Also a zero AI which shows the rad closed in game shows rad pretty open AND stationary on the corresponding track.
I'm curious, if I open fully the rad on a zero then hit autopilot, visually I see the rad fully open afterwards. But is it affecting the performance of the autopiloted plane which seems to fly with rad fully open? Yes.... that's the question... top speed should be affected.
So far I haven't seen evidences which will say otherwise, but it is only our guess at this stage.

If there's a way to make autopilot run full power level, then we could test this.... but I have no idea how to do it. http://forums.ubi.com/images/smilies/16x16_smiley-surprised.gif

lowfighter
06-03-2007, 10:21 AM
The mission below can be used, the autopilot will run at max power, he just want to catch a P47 far away http://forums.ubi.com/groupee_common/emoticons/icon_wink.gif. It shows that if I open the rad fully then hit autopilot it reaches 470 Km/h TAS at 120 m or so (starting with 300 Km/h and 100ma altitude, flying pretty level from 100m to 120 m). There's a reference pure AI flying right in front of the player, if you want to test be careful not to collide at the beginning.

[MAIN]
MAP Crimea/load.ini
TIME 12.0
CloudType 0
CloudHeight 1000.0
player usa0103
army 1
playerNum 0
[Wing]
usa0103
usa0102
g0103
[usa0103]
Planes 1
Skill 1
Class air.A6M5A
Fuel 100
weapons default
[usa0103_Way]
NORMFLY 30430.30 69800.16 100.00 300.00 &0
NORMFLY 100056.34 69819.59 100.00 300.00 &0
[usa0102]
Planes 1
Skill 1
Class air.A6M5A
Fuel 100
weapons default
[usa0102_Way]
NORMFLY 30530.21 69803.47 100.00 300.00 &0
NORMFLY 101107.86 69879.12 100.00 300.00 &0
[g0103]
Planes 1
Skill 1
Class air.P_47D27
Fuel 100
weapons none
[g0103_Way]
NORMFLY 34087.01 69723.39 100.00 400.00 &0
NORMFLY 102458.18 69803.05 100.00 400.00 &0
[NStationary]
[Buildings]
[Bridge]
[House]