Jump to content


Photo
- - - - -

Spad AI: calling out HT


  • Please log in to reply
59 replies to this topic

#1 gavagai

gavagai
  • Posts: 15542

Posted 13 October 2013 - 15:32

New Spad 13 to test versus Criquet's Spad 13. I think I found a good middle path. Let me know what you think. Use the "ace" setting for tests.

————————
Ok HT, I made a test of Criquet's Spad AI versus my own. We all know the RoF AI is bad with "boom and zoom" fighters because the AI algorithm only looks to find a target and turn at it. The best we can hope for is making it less bad. So, when I try a 5vs5 fight between the Spad 13 (using autopilot) and the Albatros D.Va, for example, the Spads take losses no matter what I do with AI script. However, in a 5 vs 3 fight, my Spad AI mops the floor with the Albatrosses. Criquet's AI does not.

—-

My series of tests went like this. 5 Spads at 3km, 1.5km away from 3 Albatros D.Vs at 2.5km, face to face.

With my Spad AI, all of the Albatrosses would be shot down, and sometimes we would lose 1 or 2 Spads, sometimes none.

With Criquet's AI, I repeatedly had multiple Spads shot down and 1 or 0 Albatrosses shot down. In fact, 0 was the more common loss ratio for the German side. In one test 4/5 Spads were shot down for the loss of zero Albatrosses.

So, why is Criquet's AI Spad so much better? I agree that hit and run tactics were the norm, but what's the point if the AI can't score a kill? Should I make the Albatros D.II use hit and run tactics, i.e. make it historical, even though the aircraft in RoF is totally unsuited to that task? I think you understand where I am going with this.

If you recall, what I did with the Spad 13 AI was to make it use vertical tactics. That way the flight maintains some cohesion and offers mutual support, instead of fighting like a herd of cats.

So, what I'm arguing is this: how we judge the AI script should be based on its empirical success. Whether it matches some idea we have of how the aircraft was flown is immaterial if it cannot shoot anything down.

————-

Here, I added the scripts so anyone can compare for himself with JSGME. I think you'll find that my "ace" is a pretty dangerous opponent! Again, I have lot of respect for what Criquet did, but I think we can still improve.

Attached Files


  • 0

#2 SYN_Adamfp3

SYN_Adamfp3
  • Posts: 82

Posted 13 October 2013 - 15:40

Prepare for handbags
  • 0

#3 gavagai

gavagai
  • Posts: 15542

Posted 13 October 2013 - 15:46

Prepare for handbags

:x

I'm forcing the issue because I think these changes would be great for RoF Refined 0.2, which will include Fubar's engines and the removed lens flare.
  • 0

#4 HotTom

HotTom
  • Posts: 8177

Posted 13 October 2013 - 16:16

If you recall, what I did with the Spad 13 AI was to make it use vertical tactics. That way the flight maintains some cohesion and offers mutual support, instead of fighting like a herd of cats.

So, what I'm arguing is this: how we judge the AI script should be based on its empirical success. Whether it matches some idea we have of how the aircraft was flown is immaterial if it cannot shoot anything down.

Gav,

First, unless you've changed it, you did not make the SPAD use vertical tactics. You made it use horizontal tactics and it just went round and round and down and die. Is this yet another new and improved version?

Second, despite all your charts and graphs and waving around your amateur historians, you now have completely reversed yourself and say the only test of an FM AI is not how historically accurate it but is how efficiently it shoots other planes down? :roll: :roll: :roll:

I'll give it a try, buddy, but it is Sunday and I have a football game (Go Lions!) and definitely a baseball game (Go Tigers!) to watch today so I may not get it tested today. Why don't you do the same? Plop down on the couch with your son and watch some sports. Get some Quality Time With The Kid in. Rest your brain. I sense it's overheated. I think I smell smoke. :mrgreen:

:S!:

HT

PS: Ty Cobb lives! Go Tigers!

Attached File  Cobbstealing3rd.jpg   35.08KB   854 downloads
  • 0

#5 gavagai

gavagai
  • Posts: 15542

Posted 13 October 2013 - 17:14

What is an "FM AI?"

I'm talking about tactics HT. By your argument Criquet should have scripted the N28 AI to try to turn fight. What is the point of Criquet's Spad 13 if its offensive ability is nil? It is not even good at running away…

Against Criquet's "Ace" I can win 5vs1 with the Mercedes Fokker D.VII. The AI Spads just immediately scatter and allow me to pick off their scouts one by one.
  • 0

#6 HotTom

HotTom
  • Posts: 8177

Posted 13 October 2013 - 17:42

Gav,

I think where your mistake (IMO of course) occurs is setting up Quick Mission furballs. The SPAD didn't furball. It attacked and escaped, climbed and returned. You keep trying to make it into a turnfighter.

Here is what Major Charles Biddle, who commanded the USAS 4th Pursuit Group, wrote in a letter home in November 1917 when he was still with the French and flying SPADs (I wish I could copy it like your hero T-C did endlessly but I don't have a scanner):

"You may think it sounds foolish, or as if one was blowing a bit of talk about, attacking five when you were only two, but an attack does not necessarily mean that you charge into the middle of them and mix it up. On the contrary, you can, by diving at high speed from above, get in some shots, and then by using your speed climb up above them again before they get in a shot.

"If you remember to leave your motor on as you are diving, and in this way come down as fast as possible, without at the same time going so fast as to interfere with your shooting, the great speed in this way will enable you to make a short steep climb and thus regain a position perhaps 200 meters above the heads of the Huns where they cannot effectively shoot at you. I am now of course speaking of an attack on a group of single seater machines.

"If the engagement ends here, the chances of bringing one down are not great, but you can sometimes by such methods, and by, for instance, hitting some part of the machines, so worry the Huns that one will in the general confusion get separated from his comrades so that you can get a fair crack at him"

(See Page 14: http://books.google.... biddle&f=false" onclick="window.open(this.href);return false;">http://books.google....books?id=NrO1Y3 … le&f=false)

(http://www.theaerodr.../usa/biddle.php" onclick="window.open(this.href);return false;">http://www.theaerodr.../usa/biddle.php)

This sure sounds like BnZ and/or energy tactics to me, not vertical or horizontal maneuvering. It's certainly the way I fight the SPAD and SE5a and F2b: Attack to split up their formation and then go after the loners.

It doesn't sound at all like what you are trying to simulate with your Mod (which I gather hasn't been changed since we tested it it since you didn't answer my question: Is this the original?). If it's the original Mod, my opinion hasn't changed: It sucks.

But I will give it a go.

:S!:

HT

PS: Ty Cobb lives! Tigers Rule!

Attached File  Cobbstealing3rd.jpg   35.08KB   823 downloads
  • 0

#7 HotTom

HotTom
  • Posts: 8177

Posted 13 October 2013 - 18:07

Prepare for handbags

LOL. As my Navy friends say: "Stand by for heavy roll!" :mrgreen:
  • 0

#8 Thaatu

Thaatu
  • Posts: 740

Posted 13 October 2013 - 18:13

I played five or six quick missions with the Spad AI test and when I was on its tail the AI started turning slightly to the left and began a steep climb. I was able to follow it only a few times as I couldn't climb as much. Once the Spad reach stalling point, it went into a leftward dive and began climbing again, unless it could fire at someone. I only tested it with Spad XIII vs. Fokker D.VII, with different numbers, but the Spads turned out to be impressively difficult for me and my AI buddies.

HotTom, I would also support BnZ tactics for appropriate planes, but a lot of skill is required even from a human player to properly pull it off. Criquet's AI's attacks are too easy to evade and the AI doesn't react when it's getting shot at during extending. This makes the AI too vulnerable, even if it has the advantage in numbers.

No one is doubting that BnZ tactics were used, but even Biddle is talking about it in an event that your flight is outnumbered. He even says "…but an attack does not necessarily mean that you charge into the middle of them and mix it up. On the contrary, you can, by diving at high speed from above…", indicating that it isn't the only tactic their squadron used. I just wish AI could be made to switch tactics depending on the situation, especially the enemy's numbers and planes.
  • 0

#9 HotTom

HotTom
  • Posts: 8177

Posted 13 October 2013 - 18:26

Yes, I agree, I wish they had more than one tactic.

It sounds to me as though you are testing them by flying as Hun (uhm German). Your reaction is interesting. It sounds as though Gav is doing the same thing. And he is giving the SPAds a numerical advantage. How about trying 2 SPADs versus 5 Germans as Biddle describes?

I hadn't tried them by flying them because the Mod really is about the AI, not my skills (or lack of them). I just set up a Quick Mission SPADs versus D.VIIs and let the AI fly while I sat in my cockpit on autopilot (AI was flying) and watched what happened. I tried it several times, from both sides, all with the same result.

What I saw was the SPADs just kept turning and turning and losing altitude while the D.VIIs constantly had altitude advantage. My SPAD wingmen and I were forced down every time (usually by head-on attacks that destroyed our engines). When I rode in the D.VII (again without me flying it), the same thing happened: SPADs turned and died.

If Gav hasn't made any changes to this Mod, it is unacceptable in my book. Any BnZ plane attempting to turn is going to lose. I've flown BnZ planes my entire sim flying career (going back about 20 years) and the rule never changes: Don't turn. Ever. Don't pull up with an enemy right behind you. Ever.

You say the AI doesn't react when it gets shot at extending. That's correct. It shouldn't react. Other than jinking a bit the object of the exercise is to get out of range. You will note in Biddle's description that he considered 200 meters height advantage sufficient to be safe from the Hun guns. Obviously, with the German prop hangers we have in RoF you need more vertical separation.

:S!:

HT
  • 0

#10 gavagai

gavagai
  • Posts: 15542

Posted 13 October 2013 - 20:07

I promise you that your own flying does not match the opinions you express here.

2vs5 with the AI is a losing proposition no matter what AI scripts you run.
  • 0

#11 HotTom

HotTom
  • Posts: 8177

Posted 13 October 2013 - 20:21

My own flying may not but my tactics are the same and they work.

Regarding testing, I do it AI only because I want to see what the AI will do absent and human influence.

You still haven't answered my question. Is this the same Mod I already tested?
  • 0

#12 gavagai

gavagai
  • Posts: 15542

Posted 13 October 2013 - 21:20

No, it is not exactly the same. There's another adjustment I need to make, but it is still very similar.

I did my tests AI only, the same way as you.
  • 0

#13 HotTom

HotTom
  • Posts: 8177

Posted 13 October 2013 - 21:37

No, it is not exactly the same. There's another adjustment I need to make, but it is still very similar.

I did my tests AI only, the same way as you.

Okay, well, if it's revised, I'll take it for a ride. :S!:
  • 0

#14 gavagai

gavagai
  • Posts: 15542

Posted 14 October 2013 - 00:12

You won't notice a large difference. I tried an adjustment for decreasing its pitch aggressiveness, but inadvertently made it so that it will sometimes auger into the ground!

There is a min and max airspeed, in meters/second, where elevator inputs are dampened to the point of being completely suppressed.

So, at the min airspeed and slower, there is no elevator movement. At the max designated airspeed and faster, there is full elevator input. Setting the min too high without decreasing the maximum bank angle can lead to disaster! Criquet's pilots do this, too, but they're never banking enough for it to be a problem.

———-

I think I've got it right now. I just tried the 5vs3 test as a human on the German side. While I was shooting a Spad his wingman came up behind me and killed me in one burst, and that's with the reduced lethality mod. I haven't been shot down like that by the AI in a long, long time.
  • 0

#15 HotTom

HotTom
  • Posts: 8177

Posted 14 October 2013 - 01:43

Well, that's how I usually get killed in PWCG, more enemies than I can keep an eye on. :oops:

I'll try it out tomorrow (Monday). :S!:
  • 0

#16 HotTom

HotTom
  • Posts: 8177

Posted 14 October 2013 - 18:02

Here is my AAR:

Both tests 3 SPAD 13s vs 3 Fokker D.VIIs, all set at Veteran, meeting in a skirmish at 4,000 meters.

1. Criquet's Mod. This was interesting because I had never actually tested his Mod to see what it would do. The good news is: Criquet's Mod extends very nicely. If no one is chasing it, it goes out about 1.5 km and then turns back. If someone is chasing, it will keep extending until horizontal separation is more than 1.5 km. The bad news is: The SPAD doesn't climb at all either extending or returning. The fight settled at 9,500 feet (I use jj's British gauge Mods) and stayed there the whole time. The SPAD actually tried to attack from below repeatedly, which isn't the way it's supposed to work.

After about 15 minutes I called it a draw and gave up. All six planes were very much still alive. If the primary goal is to survive the fight (and I think it is), both planes did very well.

2. Gavagai's Mod: Compared to the original version, they are climbing a bit more in turns but that only bleeds more energy. They still are turning far too much and they aren't very good at it. I (or my AI alter ego, actually) was wounded in less than 30 seconds and seriously wounded in less than a minute. The fight spiraled downward and ended up at treetop level.

In the end, all three SPADs went down with engine damage. One D.VII went down with engine damage, two D.VIIs survived with what appeared to be little or no damage.

My opinion of your Mod remains unchanged: You're still trying to make the SPAD into a turnfighter, which it never was for very good reason.

Go back and read Biddle's comments on SPAD tactics that I posted earlier in this thread. If you can make a Mod that does that, I will gladly use it.

My opinion (I don't vote in these pointless polls that aren't a valid statistical sample of anything): Criquet's Mod, while flawed, is far more representative of true SPAD tactics and I'm sticking with it.

Nice try but no cigar! :S!:
  • 0

#17 PatAWilson

PatAWilson
  • Posts: 3381

Posted 14 October 2013 - 21:08

Still need to differentiate between what we might be able to make the AI do and what will need a code change to do. The tactics as described in HT's post would need a code change. No fiddling with parameters would ever accomplish that. The code change would look like this:

Case 1:
1. AI is aware of an altitude advantage. IMHO it is aware of this only in the sense that it thinks it is a a bad thing, and must immediately dive to get co-alt.
2. AI is aware of its numerical disadvantage. This is obviously not the case.
3. AI makes a single diving pass and extends away - and does not return. The AI does not know how or why it should do this either.

Three swings and three misses - take a seat on the bench. No fiddling with data is going to make AI able to do something that it is not coded to do.

Case 2:
1. AI is aware of an altitude advantage.
2. AI has equal or greater numbers.
3. AI makes diving pass but then stays and fights, using energy maneuvers to maintain an altitude advantage.

Strike 1 and 2 are the same. Strike 3: you are not going to get case 2 either because the AI flat out does not know how to perform these maneuvers.

This does not mean that fiddling with existing parameters is useless, and I certainly applaud Gav for taking on this task. Just saying - let's get real about expectations.
  • 0

#18 HotTom

HotTom
  • Posts: 8177

Posted 14 October 2013 - 21:48

I agree. There is only so much Criquet could do without getting into the hard code. If his modded AI would climb as it returned to the fight it would be more effective and realistic by again attacking from vertical separation.

I'm not sure "three and out" would work. I just flew a SPAD mission today and chewed a D.VIIF to shreds but still couldn't knock him down. I ran for home when I was out of ammo (which is something else the AI don't do, they keep fighting without bullets). I would say remaining ammo ought to be the limiting factor in the duration of a fight.
  • 0

#19 gavagai

gavagai
  • Posts: 15542

Posted 14 October 2013 - 22:51

One case you don't consider HT. If Criquet's Spads start above the Fokkers, they dive far, far below the altitude of the Fokkers because they are restricted from pulling up in time. In the tests I ran that often resulted in the Spads being immediately shot down.

What if I did this? How about a Spad that maneuvers hard for a shot, but runs like a sissy the moment anything is behind it? It will still have the negative aspect of Criquet's Spads where the flight loses all cohesion the moment there is enemy contact, but there's a much better chance that they actually shoot something on the first pass.
  • 0

#20 HotTom

HotTom
  • Posts: 8177

Posted 14 October 2013 - 22:58

LOL, it's the "maneuvering hard" that gets SPADs in trouble. The whole idea is to start with a full speed run. Then what I normally do is start maneuvering only when I've crippled the bad guys and degraded their performance. By the end game, I'm almost turn fighting with them (as long as there are other Huns in sight). SPADs do turn. A bit.

Regarding unit cohesion, no the AI are not hard coded to do that and that's a shame. In that mission I just described to Pat, it was pretty much a draw. No SPADs lost, no D.VIIFs lost in a 3 on 3 fight. The Huns scattered and the fight was all over a grid square. That was with Criquet's AI and I'd say that's a pretty good result.
  • 0

#21 Thaatu

Thaatu
  • Posts: 740

Posted 16 October 2013 - 19:01

Bear with me…

I ran 24 tests, eight with each available AI: Vanilla, Criquet’s and Gavagai’s. Mission parameters are: 5 SPAD XIII vs. 5 Fokker D.VII, all ace. SPADs start at 3000 meters, Fokkers at 2500 m, facing each other at a distance of 3000 m. Each skirmish was recorded and dissected afterwards. The recording ended when the last of the planes from the losing side either crashed or had an engine malfunction, or due to issues relating to real life. Light damage means less than three damaged surfaces, heavy means more than that but an intact engine, critical means that the plane can’t fly much longer and engine failure means engine shutdown (usually due to fuel leak). Also with Vanilla and Gav's AI I indicate how much time it takes for the fight to drop from the initial 2500 meters to 300 meters. As Criquet's AI's fights are considerably longer, I didn't analyze them with the same accuracy as the others. The stated outcomes are my interpretations, so they can be disagreed with.

THE TESTS:

Vanilla test 1: German victory (narrow). The survivor is one undamaged D.VII. The altitude dropped below 300 meters in 3 minutes 30 seconds. One undamaged SPAD rammed into a forest. The fight lasted 9 min 58 s.

Vanilla test 2: German victory (clear). The survivors are three D.VII (one undamaged, one critically damaged and one with an engine failure). Below 300 meters in 3 minutes 40 seconds. Duration 4 min 30 s.

Vanilla test 3: German victory (narrow). The survivors are three D.VII (one lightly damaged, one with an engine failure). Below 300 m in 3 min 20 s. Duration 8 min 20 s.

Vanilla test 4: French et al. victory (clear). The survivors are three SPADs (one undamaged, two lightly damaged). Below 300 m in 3 min 15 s. One undamaged SPAD rammed to the ground. Duration 5 min 30 s.

Vanilla test 5: French et al. victory (clear). The survivors are three SPADs (two undamaged, one heavily damaged). Below 300 m in 3 min 50 s. Duration 4 min.

Vanilla test 6: German victory (clear). Survivors are two D.VII (one undamaged, one lightly damaged). Below 300 m in 4 min. One undamaged DVII landed. Duration 7 min 30 s.

Vanilla test 7: German victory (crushing). Survivors are four D.VII (all undamaged). The fight ended in 500-1000 meters altitude. Two SPADs rammed into each other. Duration 2 min 45 s.

Vanilla test 8: German victory (clear). Survivors are two D.VII (all undamaged). Below 300 m in 3 min 20 s. Two other D.VIIs landed successfully after engine failures. Duration 9 min 40 s.

Observations: Starts with a nice dive from above while firing. Loses a lot of altitude when turning. Encounters turn into chaotic and dense dogfights, during which a lot of collisions occur. The SPADs switch their targets rapidly when they have an opportunity. They get a lot of kills because of these snapshots.



Gavaqai test 1: French et al. victory (narrow). Survivors are two D.VII (one lightly damaged, one heavily) and two SPADs (all undamaged). Below 300 m in 5 min 30 s. The planes apparently lost sight of each other although they were within visual range. Duration 8 min 13 s.

Gavaqai test 2: German victory (crushing). Survivors are five D.VII (three undamaged, one lightly damaged, one engine failure) and two SPADs (one undamaged, one severely damaged). The fight ended at about 1 km altitude. The remaining undamaged SPAD lost sight of the D.VII although within visual range. Duration 7 min 30 s.

Gavaqai test 3: German victory (crushing). Survivors are five D.VII (three undamaged, one lightly damaged, one engine failure). The fight ended at 1.5 km altitude. At the 40 second mark one of the SPADs was destroyed in a collision, in which a D.VII lost an elevator. Duration 5 min 30 s.

Gavaqai test 4: German victory (clear). Survivors are three D.VII (all undamaged) and one SPAD (engine failure). Below 300 m in 6 min 30 s. One of the SPAD AIs apparently lost control and just circled downward into the ground. The remaining SPAD actually managed to climb away from the three D.VIIs before the engine failure. Duration 9 min 20 s.

Gavaqai test 5: Draw. Survivor is one D.VII (critical damage). Below 300 m in 6 min 30 s. One of the SPADs shed its wings with a barrel roll dive and rammed into a D.VII at the 4:40 mark. Prior to this the fight was impressively even, without any losses to either side. When the situation was three SPADs against one D.VII, two of the SPAD AIs lost control and circled into the ground. The last undamaged SPAD made a barrel roll too close to the ground while shooting the remaining crippled D.VII and was smashed. Duration 12 min 20 s.

Gavaqai test 6: German victory (clear). Survivors are five D.VII (two undamaged, one critically damaged, two engine failures). The fight ended at 1 km altitude. At the 40 second mark one of the SPADs was apparently badly wounded and lost control of the plane. Duration 5 min 30 s.

Gavaqai test 7: Draw. Survivors are one D.VII (undamaged) and one SPAD (heavily damaged). Below 300 in 6 min 30 s. Two D.VII rammed into each other and two SPAD AIs lost control and circled into the ground. The fight ended when bullets ran out. Duration 10 min 20 s.

Gavaqai test 8: German victory (clear). Survivors are three D.VII (two undamaged, one lightly damaged). The fight ended at 1.5 km altitude. Two SPADs collided with two D.VIIs in the beginning, which is why the fight was short. Duration 4 min 30 s.

Observations: At the beginning instead of diving at the enemy the SPADs barrel roll slightly under them and attack from there. Apparently there’s a higher chance of AI freezing and circling to the ground. On average the SPADs lose less altitude than with Vanilla AI, but they change targets less opportunistically.


Criquet test 1: German victory (crushing). Survivors are four D.VII (three undamaged, one lightly damaged). Duration 19 min 50 s.

Criquet test 2: Draw. Survivors are five D.VII (four undamaged, one engine failure) and five SPADs (three undamaged, two engine failures). The video ended with two D.VIIs chasing the remaining SPADs. Duration 21 min.

Criquet test 3: Draw. Survivors are three D.VII (two undamaged, one heavily damaged) and two SPADs (one undamaged, one lightly damaged). Ended with the D.VIIs chasing the SPADs. Duration 39 min 30 s.

Criquet test 4: German victory (clear). Survivors are five D.VII (four undamaged, one heavily damaged) and three SPADs (one undamaged, one critically damaged, one with an engine failure). Ended with the D.VIIs chasing the remaining SPAD. Duration 31 min 30 s.

Criquet test 5: German victory (crushing). Survivors are five D.VII (four undamaged, one engine failure) and one SPAD (undamaged). Ended with one D.VII chasing the SPAD. Duration 39 min.

Criquet test 6: German victory (clear). Survivors are four D.VII (two undamaged, one critically damaged, one engine failure) and one SPAD (engine failure). Duration 17 min 30 s.

Criquet test 7: German victory (clear). Survivors are four D.VII (three undamaged, one engine failure) and one SPAD (undamaged). Ended with the D.VIIs chasing the remaining SPAD. Duration 25 min 30 s.

Criquet test 8: German victory (narrow). Survivors are three D.VII (all undamaged) and two SPADs (all undamaged). Only one D.VII was active at the end. The rest of the planes were just circling around although they were in visual range.

Observations: The SPADs attack initially by diving down to the same level or below the D.VIIs, then turn inverted while firing and dive under the D.VIIs before extending. When a D.VII is close on their six, the SPADs make a “hanging split-S” ie. roll inverted and stay there for a few seconds before applying pitch in incremental steps. This makes their movement seem unnatural. SPADs change targets less opportunistically than with Vanilla AI.


For haphazard scoring let's go with a basic 1 to 3 (warning, there may be bias with this, but I'm too lazy). 0 for a defeat, 1 for each for a draw, 1 for a narrow victory, 2 for a clear victory and 3 for a crushing victory.



Vanilla AI
German = 11
French et al. = 4

Gavagai's AI
German = 14
French et al. = 3

Criquet's AI
German = 15
French et al. = 2


CONCLUSIONS

Although Criquet's AI's SPAD scores are the lowest, this does not necessarily mean that it handles combat against AI worse than the others. Most of the losses suffered in Criquet's AI tests were from minor fuel leaks resulting in engine failures as the fights dragged on. Both modded AIs seem to suffer from inflexible target acquirement. Whereas Vanilla AI flies worse than both modded AIs, it apparently shoots quicker, gaining snapshot kills at least with ace skill level. Vanilla AI is also better at diving from above. Criquet's could use some more work with the little maneuvering it does. Sometimes the jerking motions, especially when doing a Split-S, ripped even lightly damaged wings apart. Gavagai's AI holds the altitude advantage in their fights but suffers greatly for the inflexibility in targeting. Also for some reason it seems to freeze more frequently than Vanilla AI (I didn't check if Criquet's was freezing at all). Most likely it is just a problem with Quick Mission, where AI doesn't have any waypoints. I think it would be best if both the modded AIs were further developed and tested. If diving and targeting behaviors were adjusted to being more aggressive, it might be a fundamental improvement. If it can be done that is…



So, how many tests do we need to get a statistically valid outcome? A hundred? A thousand? That's it, I'm out.
  • 0

#22 10ps

10ps
  • Posts: 1069

Posted 16 October 2013 - 19:47

Hey HT, keep in mind the SPAD is absolutely flyable in a very aggressive way instead of extending away and coming back for a pass, even more so the SE5a. In fact, you can almost fly them "IL2 style": dive, zoom up vertically, dive down almost vertically again etc. This requires a very good aim and a lot of practice, of course, but it's definitely possible. I'm actually more afraid of such SPAD pilots than those doing hit and run attacks. Too defensive, IMO. It's not too easy to describe this, you've got to see it I'd say.

Regarding the mods… even if I don't usually fly offline I think I will try them when I get my computer back running.
  • 0

#23 gavagai

gavagai
  • Posts: 15542

Posted 16 October 2013 - 19:48

Wow, thanks thaatu. The circling thing is an error that I was able to fix.

Interesting that Vanilla scored the best.
  • 0

#24 HotTom

HotTom
  • Posts: 8177

Posted 16 October 2013 - 20:07

Hey HT, keep in mind the SPAD is absolutely flyable in a very aggressive way instead of extending away and coming back for a pass, even more so the SE5a. In fact, you can almost fly them "IL2 style": dive, zoom up vertically, dive down almost vertically again etc. This requires a very good aim and a lot of practice, of course, but it's definitely possible. I'm actually more afraid of such SPAD pilots than those doing hit and run attacks. Too defensive, IMO. It's not too easy to describe this, you've got to see it I'd say.

Regarding the mods… even if I don't usually fly offline I think I will try them when I get my computer back running.

10ps, that's pretty much what Biddle described (did you read that in Post #6 or are just chirping without reading the thread?). It's fine as long as no one is chasing you. If the bad guys are on your six, you must extend. If you pull up into a zoom climb with someone behind you, you are dead.

The SE5a used to be good at this, too, until 777 "fixed" it by nerfing the zoom climb.

Gav, you're still trying to fix a problem that doesn't exist. With the exception of climbing after the extend, Criquet pretty much fixed it.

:S!:
  • 0

#25 10ps

10ps
  • Posts: 1069

Posted 17 October 2013 - 12:32

Nope, not what I mean. Where does he speak of zooming up vertically and dropping back down immediately to keep the pressure on the enemy? I read it as 'do one attack, climb, engagement over'. Then maybe I misunderstood it for whatever reason. Also I'm not a native speaker.
  • 0

#26 Hellshade

Hellshade
  • Posts: 786

Posted 17 October 2013 - 12:57

For whatever it's worth - which probably isn't anything - I'm more interested in dangerous opponents than I am in historically accurate ones. I'd rather have the AI use tactics that aren't necessarily accurate but that make them more dangerous to fight and harder to kill. If we could have both, that of course would be great but given that the FMs are way off from their historical counterparts, trying to make AI pilots use historical tactics with non-historical flight models seems like a losing proposition to me. Even if we were able to agree on exactly what the tactics employed during the war were - they would be rendered all but useless against the sims FMs.

To me, we seem to have been going about it all backwards. We (as a community) have been trying to make the AI use historically correct tactics for the planes in order to make the engagements more realistic. But the pilots back in the day developed the tactics that they did based upon the strengths and weaknesses of the planes they were flying in and flying against. They weren't flying against pfalz-a-copters. If they were, they would have developed different tactics.

At the risk of being called a heretic of WWI history, I suggest the AI gets modded to develop successful strategies against the FMs that are in the sim. Will the battles be historically accurate in the strategies applied? No. But neither is using historically accurate strategies against, in some cases, radically different flight models. What you would get by modding the AI to use tactics that are designed to work against the sims current FM are dogfights that are much harder to win.

If I'm never going to truly get historically accurate air to air battles until the FMs are fixed, then I'd at least prefer to have challenging ones. The point of making the AI use historically accurate tactics in the interests of recreating history seems mute when flying with and against planes that don't have historically accurate performance. It's a red herring. Without accurate FMs, you just can't get there from here.
  • 0

#27 10ps

10ps
  • Posts: 1069

Posted 17 October 2013 - 16:27

Absolutely. I don't need turnfighting N28s and BnZing Alb D.IIs… :lol:
  • 0

#28 HotTom

HotTom
  • Posts: 8177

Posted 17 October 2013 - 16:30

Typical MP arcade gamers. :roll:

"I don't care if it's accurate, I just want a good fight."

"Oh! No! Don't make me read a history book!" :o

You guys should stick to playing Tetris. You're missing the whole point of a simulation. :oops:
  • 0

#29 Hellshade

Hellshade
  • Posts: 786

Posted 17 October 2013 - 17:02

You misunderstand my point completely sir. You can't have a historically accurate simulation of battle tactics if the FMs aren't also historically accurate. So since that's basically off the table, at least make them challenging. Would I prefer it all to be historically accurate? Absolutely.

Please tell me how you will recreate historically accurate combat tactics using Pfalz-a-copter Flight Models. It simply cannot happen.
  • 0

#30 HotTom

HotTom
  • Posts: 8177

Posted 17 October 2013 - 17:22

I agree that we have got to do the best you can with what 777 has given us. It is what it is. Many of the FMs are deeply flawed and, no, you can't truly recreate WWI air battles with them.

But you lose me when you say: "I'd rather have the AI use tactics that aren't necessarily accurate but that make them more dangerous to fight and harder to kill."

That's where my Bullshit Alarm goes off. :mrgreen:

Sorry, but I would rather modify my own tactics to adapt to the peculiarities in RoF than taking something that is reasonably accurate (such as Criquet's SPAD mod) and turn it into something that never was (like what Gav is trying to do: Make the SPAD into a turnfighter).

Making the AI less historically accurate (which is what this thread is about) is not an acceptable solution to me.

:S!:

HT
  • 0

#31 Thaatu

Thaatu
  • Posts: 740

Posted 17 October 2013 - 18:04

HT, you do know you can choose whatever AI mod you like. If you don't like this one, maybe you'd be better off not insulting those who may like it. Really, once all those smilies you incorrectly use are filtered out, your arrogant and ironic facade disappears and shows your hostility and aggressiveness.

Ah, the loud minorities… :roll:
  • 0

#32 HotTom

HotTom
  • Posts: 8177

Posted 17 October 2013 - 18:09

HT, you do know you can choose whatever AI mod you like. If you don't like this one, maybe you'd be better off not insulting those who may like it. Really, once all those smilies you incorrectly use are filtered out, your arrogant and ironic facade disappears and shows your hostility and aggressiveness.

Ah, the loud minorities… :roll:


Thank you for your input, Thaatu.

You tell me not to be insulting while you are insulting me?

Really? :roll:

Gav said in the title he wanted my input.

I don't like his mod because it is unrealistic and unhistorical and I backed it up with sources.

What is it about that you don't understand? :oops:
  • 0

#33 HotTom

HotTom
  • Posts: 8177

Posted 17 October 2013 - 18:17

Nope, not what I mean. Where does he speak of zooming up vertically and dropping back down immediately to keep the pressure on the enemy? I read it as 'do one attack, climb, engagement over'. Then maybe I misunderstood it for whatever reason. Also I'm not a native speaker.

No, you are understanding it correctly and I get it that you are talking about fighting in the vertical plane. Remember, we're talking about fighting AI here and my experience has been if you go vertical with a Pfalz or a Dr.1 on your tail, they will just hang on their props and chew you to pieces.

Yes, Biddle is talking about multiple BnZ attacks, which is what I try to do. With the SE5a I have to extend further because it doesn't zoom climb as well as the SPAD but the tactic is the same.

Multiple BnZ attacks is what they did. If you can find me a source that says they fought in the vertical plane, please post it. I've never seen it but I haven't seen everything.

:S!:

HT
  • 0

#34 Hellshade

Hellshade
  • Posts: 786

Posted 17 October 2013 - 18:42

Making the AI less historically accurate (which is what this thread is about) is not an acceptable solution to me.

Then don't install the mods. :S!:
  • 0

#35 HotTom

HotTom
  • Posts: 8177

Posted 17 October 2013 - 18:47

Making the AI less historically accurate (which is what this thread is about) is not an acceptable solution to me.

Then don't install the mods. :S!:

I haven't. I pitched Gav's Mod after testing it. Went back to Criquet's Mod.

I find it somewhat ironic that our self-proclaimed "rivet counter" is now drifting off into Fantasy World with this Mod, though.

:S!:

HT
  • 0

#36 Thaatu

Thaatu
  • Posts: 740

Posted 17 October 2013 - 19:41

I called you hostile and aggressive. Don't you agree with that?

Typical MP arcade gamers.

"I don't care if it's accurate, I just want a good fight."

"Oh! No! Don't make me read a history book!"

You guys should stick to playing Tetris. You're missing the whole point of a simulation.
You could fill a poetry book with these posts.

Btw, the quote from McCudden didn't convince you that maybe fighter tactics weren't as well established in the four years after the invention of air warfare as might be assumed by some modern observers, who have gained more flying hours in a simulator than most of the pilots in WW1 got in their lives. Add to this the fact that high speed airplanes began to emerge only after the latter part of 1916, making the time needed to solidify the "BnZ tactics" only two years. I don't argue that most Americans may have fought that way by the end of 1918, but what about the more numerous British and French. How long did it take for them to fight with uniform procedure? I know you like to demand sources, but this is a really minute historical detail, which the professionals don't usually bother with. Finding sources for this takes time, which I'd rather use for my own research in my own field. Call this a hypothesis if you will, but from my experience what are often perceived as strict divisions in history quickly break when research is focused on them.
  • 0

#37 Hellshade

Hellshade
  • Posts: 786

Posted 17 October 2013 - 19:45

Gav,

You're one of the most active modders at the moment and the OP of this thread. What do you think of the idea? Code the AI to fight to the strengths of the planes FM in the sim as opposed to what is historically accurate. The heck with what they did, how would they try to survive if they were flying the planes as they are in the sim. I think the dogfights would be more intense and there might be fewer kills with less predictable AI.
  • 0

#38 HotTom

HotTom
  • Posts: 8177

Posted 17 October 2013 - 19:57

I called you hostile and aggressive. Don't you agree with that?

Typical MP arcade gamers.

"I don't care if it's accurate, I just want a good fight."

"Oh! No! Don't make me read a history book!"

You guys should stick to playing Tetris. You're missing the whole point of a simulation.
You could fill a poetry book with these posts.

Btw, the quote from McCudden didn't convince you that maybe fighter tactics weren't as well established in the four years after the invention of air warfare as might be assumed by some modern observers, who have gained more flying hours in a simulator than most of the pilots in WW1 got in their lives. Add to this the fact that high speed airplanes began to emerge only after the latter part of 1916, making the time needed to solidify the "BnZ tactics" only two years. I don't argue that most Americans may have fought that way by the end of 1918, but what about the more numerous British and French. How long did it take for them to fight with uniform procedure? I know you like to demand sources, but this is a really minute historical detail, which the professionals don't usually bother with. Finding sources for this takes time, which I'd rather use for my own research in my own field. Call this a hypothesis if you will, but from my experience what are often perceived as strict divisions in history quickly break when research is focused on them.

Thaatu: I (respectfully) disagree. Most (not all) of the American group commanders and squadron commanders (Hartney, Biddle, Thaw, etc.) flew for the British or French before the USAS was created and they were the ones who determined the tactics. They had a very good idea what worked and what didn't work. The USAS went from a turnfighter (N28) to a BnZ fighter (SPAD) and they caught on pretty quickly to what worked and did not work. The Biddle quote I used was from a letter he wrote in November 1917 while still flying SPADs for the French. With the adoption of the SPAD and the replacement of the more agile Nieuports, the trend in French fighters was toward a BnZ plane. The British, too, had already moved in that direction with the SE5a and the F2b and even the Sopwith Dolphin. BnZ tactics were pretty much universal by the second half of 1917 for the Hispano-powered Allied fighters.
  • 0

#39 HotTom

HotTom
  • Posts: 8177

Posted 17 October 2013 - 19:58

Gav,

You're one of the most active modders at the moment and the OP of this thread. What do you think of the idea? Code the AI to fight to the strengths of the planes FM in the sim as opposed to what is historically accurate. The heck with what they did, how would they try to survive if they were flying the planes as they are in the sim. I think the dogfights would be more intense and there might be fewer kills with less predictable AI.

You know that is the definition of an arcade game, right? :mrgreen:
  • 0

#40 Hellshade

Hellshade
  • Posts: 786

Posted 17 October 2013 - 21:03

You know that is the definition of an arcade game, right? :mrgreen:

Isn't that funny? I thought it stopped being a sim and became and arcade game the moment the Devs decided to leave the Pfalz-a-copter, the over modeled Camel and several other horrifically inaccurate Flight Models as they were. But if you feel it's still a sim flying like that, I'm happy that you are happy sir.
  • 0


0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users