Sky_Wolf Posted November 14, 2020 Posted November 14, 2020 The AI pilots often continue to fly/fight rather than return home in situations when in real life you would expect them to disengage. Examples include when their plane is severely damaged and when they follow you back over 10 km to your base. Some of us think that the AI algorithms should be modified to make AI disengagement more frequent and more realistic. I'm interested to know how many customers think this should be improved, and how many think that it isn't much of a problem. Here is an example of a thread in the forums that deals with this question:
Gambit21 Posted November 16, 2020 Posted November 16, 2020 This is just editor logic which is easy to do - it's up to the mission designer. I can make AI disengage at any time for any reason I want basically. For instance if 8 109's attack your flight, I can make the 109's "bug out" when 5 of them are destroyed, or if the leader is killed, or after a certain amount of time etc.. As far as the career is concerned, this is possibly simply with modified mission logic....which might not be that easy across every career in the game...but possible. 1
jollyjack Posted November 17, 2020 Posted November 17, 2020 You got me curious: do you do that with damage percentage level addressing a counter pointing to a return to base ? Never realized this was possible. I'll have a look if i can find a mission using such a scheme. I noticed a FW returning home after hitting it with a P51 .50 amno in Tip's Lullaby campaign quite in vain for a while, and then the FW ran off suddenly.
Dezmond Posted November 17, 2020 Posted November 17, 2020 *Especially* now that G fatigue is such an issue. I'm not exactly a Professional Warrior but I feel I am losing a fight and want to make a run for it... Quite Often. Think the AI should do the same ?
Gambit21 Posted November 17, 2020 Posted November 17, 2020 5 hours ago, jollyjack said: You got me curious: do you do that with damage percentage level addressing a counter pointing to a return to base ? I can do it with anything I want, but I generally choose to use a certain percentage of the enemy flight killed or critically damaged. So I'll used an "on killed" report from each aircraft to a "1" Counter, then each one of those into a counter set to whatever value I want. So let's say there are 8 109's, and I want them to become "demoralized" and disengage when 5 of them are killed....so I'll set the counter to 5, then pipe that into a Force Complete "High" MCU, and a "bug out" waypoint set to "High" If I want to get fancy I can use a random output to pick one of several conditions for disengaging, but then we're getting into territory where the player will never notice the difference, yet you've made the mission much more difficult to test. Best to keep it simple. 1 1
Thad Posted November 17, 2020 Posted November 17, 2020 Yes, there isn't much a good mission builder can't do with the editor. Or at least simulate happening in a mission. 1
jollyjack Posted November 17, 2020 Posted November 17, 2020 2 minutes ago, Thad said: Yes, there isn't much a good mission builder can't do with the editor. Or at least simulate happening in a mission. Seems only Gambit is the only one revealing real secrets LoL.
Thad Posted November 17, 2020 Posted November 17, 2020 No secrets actually. Every thing is available for inspection from within the editor. But, even that takes some insight or experience with mission building.
Avimimus Posted November 17, 2020 Posted November 17, 2020 Yes. It should at least be an option. It should be possible to get AI to break-away after a single pass (especially if outnumbered - jumping the enemy for a single pass and fleeing was a common tactic). It should be possible for AI to accidentally break-away after a pass because they fail to visually re-acquire a target. It should be possible for AI to RTB if separated from allies. Adding in these scenarios using a 'realistic morale' or 'realistic disengagement' difficulty setting would be the greatest single improvement to realism in this sim. P.S. Yes, some pilots were persistent in fighting when outnumbered, at a disadvantage, or damaged - even ramming the enemy. However, it was more common for most aircraft to end up separated or low on ammo - trying to regroup or return home after even a very brief fight.
Gambit21 Posted November 18, 2020 Posted November 18, 2020 4 hours ago, Avimimus said: Yes. It should at least be an option. It should be possible to get AI to break-away after a single pass (especially if outnumbered - jumping the enemy for a single pass and fleeing was a common tactic). I'm fairly sure I could get this to happen if I wanted with a proximity trigger and timer, but it would be cumbersome to add it to the career. 4 hours ago, Avimimus said: It should be possible for AI to accidentally break-away after a pass because they fail to visually re-acquire a target. Possible with scripting, but with regard to AI coding, how would the sim know when this is feasible or completely implausible. 4 hours ago, Avimimus said: It should be possible for AI to RTB if separated from allies. Not compatible with the current logic (if you're talking about aircraft within a single flight) 4 hours ago, Avimimus said: Adding in these scenarios using a 'realistic morale' or 'realistic disengagement' difficulty setting would be the greatest single improvement to realism in this sim. Given all that's involved there, (too much to start winding down the path) and all the variables, I don't see how that's possible other than conditions I already mentioned which are possible in with the editor logic already.
JG7_X-Man Posted November 18, 2020 Posted November 18, 2020 (edited) On 11/15/2020 at 10:46 PM, Gambit21 said: This is just editor logic which is easy to do - it's up to the mission designer. I can make AI disengage at any time for any reason I want basically. For instance if 8 109's attack your flight, I can make the 109's "bug out" when 5 of them are destroyed, or if the leader is killed, or after a certain amount of time etc.. As far as the career is concerned, this is possibly simply with modified mission logic....which might not be that easy across every career in the game...but possible. Bingo! I am about to post a question to the mission making forum about how to use the On Event: Critical Damage setting the Damage: Threshold to ##% for AI "bug out", after I search JimTM's manual. This is type of behavior coincides with real life human action. When your aircraft is severely damaged, you don't stay and fight, you call out to the leader and you are assigned a wingman to escort your butt home (which is another advantage of the Luftwaffe's Finger-four idea). Why, because good leaders know that aircraft are easier to replace than pilots, something the Luftwaffe Generals didn't figure out, until it was too late (good thing they didn't...). Which is another beef I have with AI gunner logic. When they are being fired at, they don't stop (i.e. take cover), then start back up again after a few seconds. You have to kill the gunner for him to stop. This may not be correct for the nose, tail or ball turret gunners (any position that you have to be strapped in) but not the waste gunners. Zero fear factor, maybe Adrenaline? Edited November 18, 2020 by JG7_X-Man
Avimimus Posted November 18, 2020 Posted November 18, 2020 13 hours ago, Gambit21 said: 19 hours ago, Avimimus said: It should be possible for AI to accidentally break-away after a pass because they fail to visually re-acquire a target. Possible with scripting, but with regard to AI coding, how would the sim know when this is feasible or completely implausible. Interesting. I'd been think about possibly having the AI do a LOS check... and if the target is obscured with each passing second there is a chance of losing the target. The AI could then engage in a behaviour like a right-angled turn and attempt to acquire a new target. 13 hours ago, Gambit21 said: 19 hours ago, Avimimus said: It should be possible for AI to RTB if separated from allies. Not compatible with the current logic (if you're talking about aircraft within a single flight) In the original Il-2 ("Il-2 1946") there is an AI check which will let the AI call for help if an allied aircraft is within 8000 metres... I always thought this could be reversed... if there is no other allied AI with 4000 metres there would be a chance of the AI aircraft deciding to RTB. Anyway - it is quite interesting to hear that some behaviours might actually be possible within the current mission editor! I should go back to learning it (although I still have an ambition to simulate a Jock column in Il-2 Cliffs of Dover first).
Sky_Wolf Posted December 20, 2020 Author Posted December 20, 2020 Ok, so here is another example of how the enemy AI pilots rarely disengage when you would expect them to. At around 6:15 into the video a 109 begins leaking fuel and radiator fluid. The 109 is hit again in four more attacks by the MiG - the 109 is not looking good after the fourth (at ~8:50 into video). But sure enough the 109 does not disengage. At around 11:00 into the video the same 109 sneaks up on the MiG's six and begins firing. IMO this has to be addressed. The 109 should have disengaged, maybe by diving and flying low back to base. 2
Dutch2 Posted December 23, 2020 Posted December 23, 2020 On 12/20/2020 at 7:48 PM, Sky_Wolf said: Ok, so here is another example of how the enemy AI pilots rarely disengage when you would expect them to. At around 6:15 into the video a 109 begins leaking fuel and radiator fluid. The 109 is hit again in four more attacks by the MiG - the 109 is not looking good after the fourth (at ~8:50 into video). But sure enough the 109 does not disengage. At around 11:00 into the video the same 109 sneaks up on the MiG's six and begins firing. IMO this has to be addressed. The 109 should have disengaged, maybe by diving and flying low back to base. Another point, damaged AI plane should also trying to spare the plane, not like now still doing stunts at an high performance. This effect can be clearly seen in FC when a strut has been broken and the AI plane is still running/diving/turning fast. While we are discussing the AI behavior since 2009 in RoF and for some strange reason the developers still seems not to get this job right. I would suggest for 1C to make the AI adjustable in different aspects, like we can do in OBD WoFF and WoTR cfs. Here if I well can remember you can adjust the AI on 3 aspects, Skill/engage/??? so you can adjust the whole AI on your personal taste, that way will end 80% of the future AI discussion. 1
Avimimus Posted December 23, 2020 Posted December 23, 2020 9 hours ago, Dutch2 said: While we are discussing the AI behavior since 2009 in RoF and for some strange reason the developers still seems not to get this job right. I would suggest for 1C to make the AI adjustable in different aspects, like we can do in OBD WoFF and WoTR cfs. Here if I well can remember you can adjust the AI on 3 aspects, Skill/engage/??? so you can adjust the whole AI on your personal taste, that way will end 80% of the future AI discussion. I"ve got doubts about that... I've seen it in SDOE and Cliffs of Dover... it is very easy to get the settings wrong, and with linear sliders used to determine the behaviour of a non-linear dynamical system - results are extremely unpredictable. What I think would work better is breaking the AI into a series of separate behaviours (likely already the case), but making whether those behaviours are triggered more adjustable and more contextual. So being able to take a 'dive and RTB' manoeuvre and adjust: - The probability it is triggered - The rules which can trigger it For instance, being outnumbered should increase the likelihood. Having altitude should increase the likelihood of a dive. Being damaged should decrease the likelihood of a dive. Having a faster enemy closing should increase the likelihood of the 'RTB manoeuvre' aborting and the AI trying to fight it out (as it can't disengage successfully). So you end up with a series of basic manoeuvres/behaviours which can be tested extensively to ensure they work (bug free and reliably) ...but one also has more rules for triggering when and where they take place (and these rules can be adjusted). Such a modular structure could also be used to create more context, time-period, or even aircraft specific manoeuvres (e.g. A Sopwith Triplane, Fokker Dr.I or Fokker D.VIII might try to climb to disengage, while a Fokker D.VII or a Spad would dive to disengage)!
Recommended Posts