4.11 - AI debugging

[Uzin]

Successfully installed 4.11 over vanilla 4.101.
Anyway, a lot of work was done. Thanks to DT.
Tried some QMB and single missions.
Some obsevations:
- former *.trk files not accepted by the new version.
- longer than 10 minutes *.trk files recored differ essentially from that what was observed in the flight.
- CTD at Quick mission without any known reason.
- the possibilities of waypoints - take offs aside (=pair) - do not work in QMB , weird places occupied by AI planes.
Surely these and other bugs will be repaired by DT, I hope.
Ik-3 has roundels on one wing upper surface only, second wing lower surface only. Is it correct ?

[Juri_JS]

I've noticed that an old problem wasn't fixed by the new AI. When doing rocket attacks on ships planes pull up very late and will often hit the ships superstructure. I hope this can be fixed in the future.

[Swo7pes]

I hope this can be fixed in the future.

[FC99]

Originally Posted by Uzin

- former *.trk files not accepted by the new version.
- longer than 10 minutes *.trk files recored differ essentially from that what was observed in the flight.
- the possibilities of waypoints - take offs aside (=pair) - do not work in QMB , weird places occupied by AI planes.

AFAIK TRK files works only on the version they were made on.
TRK are inherently unreliable, they are basically mission played again and minor input from the player like changing the view or things like that can get random numbers out of the synchronization and with completely different result of the mission. NTRK's are much better in that regard.

Is Pair Take Off very important to you in QMB? QMB engine reads the template file and than create new mission file according to the QMB setting, obviously it doesn't take into account some of the new features. We can probably enable that but I'd rather work on things that are more important IMO. But if you want it really bad and I find a time anything is possible.

Originally Posted by Juri_JS

I've noticed that an old problem wasn't fixed by the new AI. When doing rocket attacks on ships planes pull up very late and will often hit the ships superstructure. I hope this can be fixed in the future.

Can you tell me which planes gives you most of the problems?

When I tried to create a ship attack mission for the Mosquito XVIII it showed a strange AI behaviour. It looks as if the planes can't find the correct attack approach to the target and they will just circle around it.

Rgr, will investigate, thanks for sending the problematic mission, that will help us a lot to correct the problem as fast as possible.

Originally Posted by Fall_Pink?

Jet fighters, e.g. me262, have trouble handling engine overheating. Took off with 8 me262's (career center, 1945) with auto pilot on and the engine got overheated. It reduced power after the message appeared, but it was too late and got me a smoking engine.

Taking off in Ta183 with 4 missiles is very hard for AI. At the end of the runway the engine is already overheated and AI throttles back a bit (20 or 30% power), but loses too much altitude and crashes. Others barely escape and fly only a few meter above ground.

Rgr, we will check is this a problem of AI or it is the general problem with jets.

Me163 sometimes switches targets and engages enemy fighters instead of bomber stream (tested with pre-set single mission with me163). Me163 should be mainly a bomber attacker and not engage any fighters. Maybe some defensive manouvres when it encounters fighters, but its focus should be bombers.

It's also prone to miscalculate attacks, likely due the large speed difference with bombers.

Furthermore, it does not disengage itself quickly from the fight to glide back home. Maybe it should disengage, fly back to home base and distance itself as much as it can when only 10% rocket fuel is left.

Are bombers set as target in that mission, AFAIK if bombers are target than plane should attack them first, it could switch the target eventually if it is endangered by fighters.

Miscalculated attacks can look weird but I'd say that it was not easy to do it right in RL too, maybe we can get them better but makeing errors from time to time sounds realistic to me.

AI planes should disengage from combat when fuel is low even now.

Is it possible to add a command like ´disengage´? I now usually use something like get ´rejoin´ or ´cover me´, but that does not seem to work in case of a ground attack in progress. I´d like my wingmen to rejoin immediately.

Secondly, when the command `back to base´ is given, all wingmen follow a route that takes them back to base except the leader which seems to follow the normal waypoints.

Ps. I also see all of my real cpu cores are now involved some how. Is this one of the effects of moving AI code to the DT.dll?

Do we really need new "Disengage" command or it would be enough to enforce rejoining under any circumstances with the current "Rejoin" command?

I'll check RTB command but I need more info, are you talking about your flight or it is about other flights under your command?

Better distribution of the CPU load is not the primary reason for DT.dll but it might help in that regard. In some of our tests performance was approximately the same or even better with 4.11 than same missions in 4.10 despite the lot of new code in 4.11.

Bug: single missions, Hungary, dogfight P51's with Me109G14:

4th wingman of each flight of P51's crashes into the lake because they're too low to the ground. Stacking order of a flight that flies "Reihe" (sorry, only know the German word for it right now...), causes problems for the 3rd and 4th wingman. I'd suggest to use a formation with less vertical spacing when lead plane flies at very low altitude. It would lead to fewer crashes of these wingmen.

Can you give me more details, was that during the cruise or it was some sort of ground attack, I'm not sure what "Reihe" is but from your description this could be "Line" formation which is not standard for P-51.

Originally Posted by pupo162

ATM AI is so much interesting than in previous releases. its more creative, has more maneuvers etc but it lacks the 101 of every fighter pilot: they dont know how to combat turn!

We are aware of that and some other issues with AI ACM. You have to understand that we are doing this in our free time and there is only so many hours in a day. We know that current AI is far from perfect but we thought that it is better to release "improved" AI than to wait another year for a "perfect" one.

Anyway, thanks for input, it's appreciated and we will do our best to improve AI turning ( I hate what they do too so chances are good that this will be better in next patch. )

Originally Posted by slm

I attacked 3 enemy planes, TBD-1. I got some hits to a plane that was probably the leader of the group. Smoke was coming out of the damaged plane and it started slowly losing altitude.

the bug: its wingman kept following the damaged plane towards the ground and both AI planes crashed.

Another old problem, we are aware of that and it's high on priority list.

Originally Posted by Treetop64

Really diggin' the new patch, but one persistent problem with the AI continues:

I'm still getting hammered by the AA fire over my own airfield when defending it.

EDIT - Sorry, one more thing later observed. AI formation keeping is much worse now than in previous versions:

AFAIK we didn't change anything regarding AAA, they should not shoot at enemy when friendly planes are close but it is possible that something is not working right. Still, we need to give them a chance to be as stupid as humans, people have been killed by friendly AAA(G.Preddy for example)

AI is making "check six" maneuvers from time to time maybe this can be the reason for some of the problems with keeping the formation.

Originally Posted by MadCat242

I experienced an odd ai behaviour:
QMB 2 bf109 against 4 ai novice il-4. Crimea map.
- 1st pass head-on: damaged no.4. (lost one of his engines -> crew bailed)
- 2nd attack from 6 o clock : I hit the wingleader. Several hits on one wing, damaged his tailplane, smoking engine -> bomber rolls over, going into a steep dive, slowly recovers (guess due to increased speed) and crew bails.
-> entire AI flight (no. 2 and 3) flipped over as well and followed their leader...
The moment No1 started bailing the other 2 went back to level flight and headed north.

Is that on purpose? Did the ai "thought" their leader is leading them into that dive? Blindly following orders? Or did the ai just didn't switch to "hes going down, no need to follow. No.2 is the new boss"?

This is same problem reported by slm too. It is the old one and we are working on it, hard part is to decide when to declare the Leader plane as not combat capable and assign the lead to someone else.

Thanks everybody for your input, keep them coming, this can only help us to get the game even better.

FC

[Juri_JS]

Originally Posted by FC99

Can you tell me which planes gives you most of the problems?

All aircraft types with air-to-ground rockets are affected by the late pull up issue during ship attacks and also all large caliber aircraft that are able to do strafing attacks on ships. I have uploaded a mission to demonstrate the problem and to make the testing easier for you.

http://www.axis-and-allies-paintwork...attacktest.zip

Originally Posted by FC99

Rgr, will investigate, thanks for sending the problematic mission, that will help us a lot to correct the problem as fast as possible.

There seems to be a problem with the ship attack routine of both the Mosquito XVIII and the rocket equipped Mosquito VI. When bombs are used to attack ships everything is working fine. My guess is that it has something to do with the fact, that the Mosquito VI only had bomb loadouts when it was initially released and a correct rocket attack routine for ship attacks was never added, which also affects the Mosquito XVIII because it uses the same routine - but that is just a hypothesis, let's see what your investigation will find.

[slm]

Originally Posted by FC99

This is same problem reported by slm too. It is the old one and we are working on it, hard part is to decide when to declare the Leader plane as not combat capable and assign the lead to someone else.
FC

From what I've read, when their plane got hit so there was serious damage pilots often knew it soon and announced it to other pilots **long before using parachute**. Of course the pilot doesn't always know how far his plane can continue flying, but I think it would be great if this "unable to continue mission" could be simulated at least in some cases. Especially when bombers are flying in formation towards target and the lead plane is hit and thus cannot keep planned speed/altitude anymore.

[Fall_Pink?]

Quote:
“Me163 sometimes switches targets and engages enemy fighters instead of bomber stream (tested with pre-set single mission with me163). Me163 should be mainly a bomber attacker and not engage any fighters. Maybe some defensive maneuvers when it encounters fighters, but its focus should be bombers. It's also prone to miscalculate attacks, likely due the large speed difference with bombers. Furthermore, it does not disengage itself quickly from the fight to glide back home. Maybe it should disengage, fly back to home base and distance itself as much as it can when only 10% rocket fuel is left.”

FC > Are bombers set as target in that mission, AFAIK if bombers are target than plane should attack them first, it could switch the target eventually if it is endangered by fighters. Miscalculated attacks can look weird but I'd say that it was not easy to do it right in RL too, maybe we can get them better but making errors from time to time sounds realistic to me. AI planes should disengage from combat when fuel is low even now.

Me > Bombers are set as target for Me163 in that mission I think. You're right, it attacked the bombers first and made 2 (very) wrong passes without firing anything and then started to dogfight with the Mustangs. Mustangs did indeed fire upon the Me163, so maybe that's the logic behind it all. From an AI point of view it sounds logical, but nevertheless, it's not very realistic for a Me163 to do The Me163 is the odd one out here I think, because gliding back is of course intended. It needs to disengage from the fight to get out of reach of fighters for a safe gliding position.

My main concern here is I think real high speed fighters like early jets (and Me163) need a bit different attack AI most of the time to really exploit their high speed/energy advantage. Jets have so much energy that they cannot make these quick and sudden turns and therefore often miscalculate when they attack relatively slow targets like bombers. Boom and zoom from far away and extend is the preferred tactic most of the time. I think jets generally require a more relaxed and more gentle AI when it comes to attacking. It would definitely look more real that way.

Quote:
Is it possible to add a command like ´disengage´? I now usually use something like get ´rejoin´ or ´cover me´, but that does not seem to work in case of a ground attack in progress. I´d like my wing men to rejoin immediately. Secondly, when the command `back to base´ is given, all wingmen follow a route that takes them back to base except the leader which seems to follow the normal waypoints. Ps. I also see all of my real cpu cores are now involved some how. Is this one of the effects of moving AI code to the DT.dll?”

FC > Do we really need new "Disengage" command or it would be enough to enforce rejoining under any circumstances with the current "Rejoin" command? I'll check RTB command but I need more info, are you talking about your flight or it is about other flights under your command?
Better distribution of the CPU load is not the primary reason for DT.dll but it might help in that regard. In some of our tests performance was approximately the same or even better with 4.11 than same missions in 4.10 despite the lot of new code in 4.11.

Me -> It was my flight under my command (4 F4U's attacking Palau with bombs), a DGEN mission. If the “rejoin” command also implies "disengage" that would also be okay of course

Quote:
“Bug: single missions, Hungary, dogfight P51's with Me109G14:
4th wingman of each flight of P51's crashes into the lake because they're too low to the ground. Stacking order of a flight that flies "Reihe" (sorry, only know the German word for it right now...), causes problems for the 3rd and 4th wingman. I'd suggest to use a formation with less vertical spacing when lead plane flies at very low altitude. It would lead to fewer crashes of these wingmen.”

FC -> Can you give me more details, was that during the cruise or it was some sort of ground attack, I'm not sure what "Reihe" is but from your description this could be "Line" formation which is not standard for P-51.

Me -> Reihe = line astern. Background: P51's are about to strafe a Hungary position near lake Balaton and fly across the lake. The 4th and last plane in the line is almost leaving a wake in the water and eventually crashes. Of two flights of P51, the 4th one crashed. Line astern position during a ground attack uses a stacking where each wingman flies somewhat below its leader. The 4th one in the flight flies near ground level when the leader flies ~ 100 meters above ground level. The line astern formation during ground attack leads to unnecessary AI crashes if each plane flies beneath its leader. It could be prevented if programmed differently. Let's say: don't use this stacking order when the leader flies at 100 or 200 meters.

Don't get me wrong, 4.11 is a very good patch 4.11 is a great improvement when compared with 4.10.1 when it comes to AI

Regards,
FP

[Bearcat]

I like most of the AI improvements .. but one thing that still sticks in my craw is the fact that AI will still keep flying after getting flamed. I can see if it is an engine fire and the plane goes into a dive .. but a wing fire .. the pilot should bail... or crash ... or bail and then crash .. but he definitely should not continue flying and even shooting..

[IceFire]

Originally Posted by Bearcat

I like most of the AI improvements .. but one thing that still sticks in my craw is the fact that AI will still keep flying after getting flamed. I can see if it is an engine fire and the plane goes into a dive .. but a wing fire .. the pilot should bail... or crash ... or bail and then crash .. but he definitely should not continue flying and even shooting..

There may have to be some subtlety there. Not sure how the AI works but if I were flying a multi engine medium or heavy bomber I may stick with the plane with a fuel fire on a wing and see if it will go out. If so the plane may be flyable back to base. On a Ki-84 with a wing fire I would bail immediately.

I suspect that was also the case with many crews during WWII having read so many reports of B-17s coming back with incredible damage and sometimes suffering fires on the way back. Not all of those ended at all in a happy way but some of them did make it back...

So ... we may not want the AI to jump at the first sign of fire. Some sort of judgement calculation maybe? No idea how it specifically works.

[MOG_Hammer]

Originally Posted by Bearcat

I like most of the AI improvements .. but one thing that still sticks in my craw is the fact that AI will still keep flying after getting flamed. I can see if it is an engine fire and the plane goes into a dive .. but a wing fire .. the pilot should bail... or crash ... or bail and then crash .. but he definitely should not continue flying and even shooting..

If the mission writer unckecks the parachute box in the full mission builder for any particular AI flight, there is no parachute, thus no bailing out of the pilot. Besides, if it was a P-38, it can be brought home with a flaming wing for about 20 km. Sometimes yes the AI will continue to fight back, because it is the only valid option it has. Besides, do you always bail the moment you are burning in your crate?