Full FM for AI, the benefits

[F/JG300_Gruber]

Hi all 

 

I've been wondering for quite a while now what are the real benefits of having a full detailed Flight Model for the AI ?

I do understand that having the same limitation as a human player is a nice idea. AI will not pull crazy manoeuvers like sometimes on the older sims, and it is fun to see sometimes the AI lose control of the airplane for no reason like human players does. It is also a good marketing selling point. But it seems to me like it comes at a heavy cost :

 

- Harsher load on the CPU to calculate all the aerodynamic parameters of all airplanes.

- Drain of development ressources, as it seems that each AI needs to be taught to fly airplane per airplane (remember the p40 when it came out).

- Severe limit of how many AI server can use.

 

Plus some of AI characteristics fall way short of expectations on some specific areas such as :

 

- Formation flying, especially for bombers and attackers which can't maintain a basic 3 airplane V element, and scatter all around during turns.

- AI still fly a low variety of patterns (go scissors, go continuous turn, go escape climb) that they struggle doing properly, fighting more their plane than the player.

- Flying "style" easily recognizable (constant wing waggle when you get on its 6').

 

I'm maybe too pessimistic but anyways, my feeling is that reverting back to a simplified FM (if ever possible) would allow more time and ressources to tweak the AI behavior and implement more complex logic in their response to improve their fighting skills, not to mention overall flying skills (I'm jealous of CloD 6 bombers tight formations) and hopefully include more of them in MP without throwing the server on its knees. Overall it could bring a better, albeit different, user-end experience to the player at a lower cost.

 

What are your thoughts ?

[unreasonable]

I agree that the current arrangement is sub-optimal, but I see no chance of getting it changed given the low priority of AI issues. 

 

In Il-2 1946 formation flying was, IIRC, solved by having the AI "warp" into formation position once it had steered fairly close to the required position. It would then lose formation if the leader took too sharp a turn. It was a little artificial looking, but at least this meant that your flight could fly in a finger four at a spacing dictated by the player. BoX AI cannot even try to do this.

 

Oddly the BoX AI is much worse than the RoF AI at maintaining formation: I expect the higher speeds are the root of that, but the mission designers are also at fault (both human and the career machine) in giving AI sharp turns. If the direction changes are no more than about 30 degrees the AI, once formed up, can more or less maintain formation and avoid collisions. The planes should be echeloned on the outside of the turn.  You cannot expect AI to make changes of direction in formation that would be impossible for experienced humans.

 

My own opinion is that the fighting skills of the AI are quite good enough. If you play career on iron man the key to survival is seeing the enemy and not getting bounced: which is remarkably difficult to do hour after hour.   

[Ehret]

If we want quality then we have to bear costs. At least bots in the game are fair - they can be criticized but unlike (sigh) WT's ones they don't induce anger.

 

FM for bots is the most ideal approach for an AI crews - the latter may need changes, not the former.

Edited August 6, 2018 by Ehret

[ZachariasX]

5 minutes ago, F/JG300_Gruber said:

I've been wondering for quite a while now what are the real benefits of having a full detailed Flight Model for the AI ?

I think one has to differentiate on the situation.

 

As I understand the workings of BoX, every plane is always on „full real mode“, meaning it has the same realistic flight model as well as engine management, gunner bots etc, unrelated to the situation. An AI plane taxes the CPU in the same way, regardless whether it is alone or is in combat.

 

In case of direct combat, a simplyfied AI flight model will make the AI fly in an unnatural way. For jets, this is not so bad, as they are FBW anyway and they are flight control limited in their maneuvering. In analog slow planes, you indeed take the plane to the edge. Making a mistake there will make the difference in the fight.

 

In DCS, this becomes very evident. The AI Mustang is like in cheat mode, always full power, hardly bleeding speed in turns. Also the Mig15 profits a lot there. They are different aircraft and do things that shouldn‘t be possible like that.

 

Having the hassle to teach the autopilot to fly a complex flight model cannot be an excuse to not do it.

 

I would propose that only planes entering the event horizon (however defined) should switch to complex control and management. Simplified AI vs simplyfied AI is again fair.

 

For example, if one was to fly with a formation of say 300+ bombers, only the ones that are within that event horizon should switch to complex FM, put up a real fight, and the rest can fly somplified and gunners can exchange hit probabilities. You‘re not seeing what happens at the other end of the formation, but if an AI formation attacks AI bombers there, you want an outcome that makes sense. And you shoukd see that outcome once you fly there.

 

This way, you could potentially have 500+ aircraft in one mission, but only maybe 40 or so will be in  „realistic“ mode, the rest will be CPU-friendly.

 

[Ehret]

Just now, ZachariasX said:

This way, you could potentially have 500+ aircraft in one mission, but only maybe 40 or so will be in  „realistic“ mode, the rest will be CPU-friendly.

 

The other way to do this is to lower tick-rate of FM and AI algorithms for bigger, less maneuverable planes. Or, for all planes which aren't engaged at the moment to say 1/10x of the normal tick frequency.

 

[Mac_Messer]

Haven`t put much thought into it, mainly because I haven`t flown online for ages now. I guess it all depends on the engine`s limitations. If relatively few non-static AI units slow down the server to a crawl, that indeed is heavy cost. IL2 series have low pc hardware requirements as one of its bullet points you know. At the same time, nowadays having for example a 64 player coop game is not exactly state-of-the-art technology either.

 

I`d be down for what ZachariasX brought up, that would save some resources. That said, we don`t have the first hand knowledge on how the game engine works. Mine aswell be that making the units placed further away from the player use simplied algorithms don`t make any difference at all.?

[Yogiflight]

I absolutely agree with what Zacharias posted. The AI is simplified anyway in fights AI versus AI. You can see this pretty easy, when a squadmate attacks an enemy, the enemy wil more or less fly straight on. As soon as you attack him, he will start turning, when you turn away and only your squadmate is behind him, he again flies straight on. So why not give him a simplidied Fm, too. It doesn't matter, as there is no real fight anyway.

[Livai]

In other Games it's nearly impossible to catch up to an A.I. controlled plane once it disappears from your screen because the Game logic applies to the CPU  giving them off screen an unfair advantage like simpled FM and Physics whatever -

[-TBC-AeroAce]

I have been thinking about the benefits vs cpu.... cost of the AI using the same FM for a long time. I think at first it was a selling point "look our AI uses the same FM as you , isn't that great" but after playing the game for a long time the costs for cpu load..... are very obvious if you have a lot of AIs.

 

I think that the AI is fine as we have it now (load wise) for fighters but they may want to consider a simplified AI for bombers which are the main cause of slow downs. Maybe if we had a simplified AI for bombers we could have 4 engine heavy bombers but as ever I understand I might not be that simple?

Edited August 6, 2018 by AeroAce

[F/JG300_Gruber]

1 hour ago, unreasonable said:

Oddly the BoX AI is much worse than the RoF AI at maintaining formation: I expect the higher speeds are the root of that, but the mission designers are also at fault (both human and the career machine) in giving AI sharp turns. If the direction changes are no more than about 30 degrees the AI, once formed up, can more or less maintain formation and avoid collisions. The planes should be echeloned on the outside of the turn.  You cannot expect AI to make changes of direction in formation that would be impossible for experienced humans.

 

It's true that waypoints placement should be improved in automatically generated missions. I will try to pay a bit more attention to this in my own. But the root of the problem lies elsewhere I think, look at CloD, in tight formation (even with medium bombers), you can turn quite significantly and the formation will hold solid up to a certain point. (even if it's a bit overdone and sometimes reminds me of a fish bank)

Formation itself still needs a revamp, no matter what is the situation, even after a 100km straight leg, the wingmen will settle 150m behind the leader instead of placing themselves on his 4' and 8'. 

 

1 hour ago, Ehret said:

If we want quality then we have to bear costs. At least bots in the game are fair - they can be criticized but unlike (sigh) WT's ones they don't induce anger.

FM for bots is the most ideal approach for an AI crews - the latter may need changes, not the former.

 

In WT, ennemy AI induce anger, in BoX, wingmen induce anger 

 

1 hour ago, ZachariasX said:

In DCS, this becomes very evident. The AI Mustang is like in cheat mode, always full power, hardly bleeding speed in turns. Also the Mig15 profits a lot there. They are different aircraft and do things that shouldn‘t be possible like that.

 

Well, that may be a bit extreme, I believe there is a middle ground between overly simplified FM and Full Real. If the aircraft can fly outside of the flight enveloppe of the real one, and run at continuous emergency power, obviously something is wrong in the simplification, not in the principle of having a simplified FM.

Maybe if the simplified FM boundaries are set by the practical limits a good human player can get, instead of the theorical figures for a given airframe, it may prevent the possibility of such event ?

 

They are some things like engine management that are, I think, not really hard to calculate and can be kept, but continuous complex airflow response of every single bit of the plane up to the pitot probe for any input the AI may throw on the control column is more than what I believe is necessary to have a realistic reaction in a fight. 

 

Good point about the event horizon though

Edited August 6, 2018 by F/JG300_Gruber

[Quinte]

1 hour ago, ZachariasX said:

In DCS, this becomes very evident. The AI Mustang is like in cheat mode, always full power, hardly bleeding speed in turns. Also the Mig15 profits a lot there. They are different aircraft and do things that shouldn‘t be possible like that.

 

True. But believe it or not, I much prefer DCS's AI when it comes to fighting. At least it's somewhat challenging. In a 1on1, my dora vs AI p-51, I need to focus to win.

 

Currently that's not the case in BoX.

 

Not that it's a simple matter if simplified FM, though. Iirc CLOD has simplified AI, and they're not that good, War Thunder has very simplified AI and they're absolutely ridiculous. I'd love to read Jason's take on this, I assume there are reasons they went this way.

 

As for those claiming AI issues have no priority, I guess you heard that straight from the devs, right ?

[unreasonable]

11 minutes ago, F/JG300_Gruber said:

 

It's true that waypoints placement should be improved in automatically generated missions. I will try to pay a bit more attention to this in my own. But the root of the problem lies elsewhere I think, look at CloD, in tight formation (even with medium bombers), you can turn quite significantly and the formation will hold solid up to a certain point. (even if it's a bit overdone and sometimes reminds me of a fish bank)

Formation itself still needs a revamp, no matter what is the situation, even after a 100km straight leg, the wingmen will settle 150m behind the leader instead of placing themselves on his 4' and 8'. 

 

 

 

Well I expect that would be because CLOD formations, like IL-2 1946, lock the plane into the assigned position.  RoF V formations are positioned at 4 and 8 o'clock and so on, and at times they can hold this very well if as leader you fly straightish and do not fly or climb at maximum speed. I would not be surprised if the developers started out with the same target positions, tried them out and then saw the carnage..... hence the set well back positions so that they can sway from side to side without hitting the leader. Another thing to notice: AI leaders are flying waypoints with a set speed which is often fairly near their rated power cruising speed (especially when lugging a bomb around). Anyone trying to catch them up having fallen a few hundred meters behind has to use combat power - or take many minutes.  So that is another mission design problem that makes everything worse in the forming up stage.  

 

I do not think this is easy to solve, especially for the fighters: making turns in formation is not a simple exercise. For instance, in echelon each plane should maintain it's height, in finger four or vic the outer plane should be higher, the inner lower. That is before you start worrying about crossovers.  

 

When I am leader in career I like to solve this by just flying as a pair. 

11 minutes ago, Quinte said:

 

 

As for those claiming AI issues have no priority, I guess you heard that straight from the devs, right ?

 

We have been told by Jason that he would like to spend more time on AI but cannot as he has no AI programmer.  

[Quinte]

That's sad to hear. I gave up long ago waiting on a flight sim with decent AI anyway, though. That's why MP is where it's at.

[FTC_Riksen]

Gruber is right and simplified within limits is the way to go. The AI in BOS is lightyears behind its predecessor il2 1946.

[F/JG300_Gruber]

22 minutes ago, unreasonable said:

 

Well I expect that would be because CLOD formations, like IL-2 1946, lock the plane into the assigned position.  RoF V formations are positioned at 4 and 8 o'clock and so on, and at times they can hold this very well if as leader you fly straightish and do not fly or climb at maximum speed. I would not be surprised if the developers started out with the same target positions, tried them out and then saw the carnage..... hence the set well back positions so that they can sway from side to side without hitting the leader. Another thing to notice: AI leaders are flying waypoints with a set speed which is often fairly near their rated power cruising speed (especially when lugging a bomb around). Anyone trying to catch them up having fallen a few hundred meters behind has to use combat power - or take many minutes.  So that is another mission design problem that makes everything worse in the forming up stage.  

 

I do not think this is easy to solve, especially for the fighters: making turns in formation is not a simple exercise. For instance, in echelon each plane should maintain it's height, in finger four or vic the outer plane should be higher, the inner lower. That is before you start worrying about crossovers.  

 

 

Believe it or not, it is incredibly hard to make two planes collide in Clod even in 6 planes in a single tight formation. Wingmen are very responsive. 

From what I remember, it looks like they will more or less follow your bank angle and not only your path, hence much less response delay when you start pulling the stick.

 

Also on the BoX waypoint speed, even when reducing the value, wingmen will settle way back their leader. They will just get into this position faster.

When you are the leader, close the throttle to let them catch up and they will also slow down to stay all the way back there. So I believe they still have assigned fixed positions relative to their leaders. I understand the need for a bit more spacing to cope with the increased inertia of a realistic engine management, but the current setting is a bit extreme 

 

Edited August 6, 2018 by F/JG300_Gruber

[SAS_Storebror]

From a Server Operator's point of view let me add that currently the "lots of AI" we can handle on a reasonably well powered dedicated server (speaking of around 2200-2500 single thread passmark points here) is not even remotely in the range of "64" as stated above or anywhere along the numbers IL-2 1946 could handle.

With a dozen active AI ground units, 8 AI fighters and 4 AI bombers, the BoS DServer is living on it's edge.

Let me further add that it's not just AI planes that matter.

The order of impact on the CPU for a dedicated server is like this:

1. AI bombers
2. AI fighters
3. Trains
4. Other AI Vehicles (Tanks, cars etc.)
5. Human Player planes
6. Human Player tanks
7. AAA
8. Other stuff (smoke effects, flares, things like Complex Triggers, Check Zones etc...)

The first half of that list matters to an extent where you really have to think twice about each and every such object you want to put on a map/mission.

 

I'd really be happy if something could be done about that, whatever it is.

From the Patch 3.005c release notes I understand that some multiplayer performance related changes are to be evaluated by beta testers. Let's hope that they really found some significant improvements there and that it's working well without breaking too many other things.

 

Mike

[unreasonable]

3 minutes ago, F/JG300_Gruber said:

 

Believe it or not, it is incredibly hard to make two planes collide in Clod even in 6 planes in a single tight formation. Wingmen are very responsive. 

From what I remember, it looks like they will more or less follow your bank angle and not only your path, hence much less response delay when you start pulling the stick.

 

Also on the BoX waypoint speed, even when reducing the value, wingmen will settle way back their leader. They will just get into this position faster.

When you are the leader, close the throttle to let them catch up and they will also slow down to stay all the way back there. So I believe they still have assigned fixed positions relative to their leaders. I understand the need for a bit more spacing to cope with the increased inertia of a realistic engine management, but the current setting is a bit extreme 

 

 

I have not played much CLOD but I did play a lot of IL-2 46 in SP with AI: I suspect that once the AI is "locked" into the assigned position, the person who is actually flying them is the player, up to a point when they are outside the assigned spot relative to the player at which point the AI wakes up and tries to get them back into the slot.  When you give a formation command in 46 you could see the AI plane fly naturally up to within a couple of meters of it's assigned position, then sharply move the last couple of meters and  lock into position. 

 

I agree they still have assigned positions in BoX, but the AI is still flying the planes all the time.  I have occasionally been able to see three wingmen all lined up in echelon behind while looking over my shoulder in the 109, but only as long as I fly straight and level: with the slightest deviation you are in effect flying using a chaotic version of Malan's stacked trail formation.  TBH it might be easiest just to put them in column all the time and just have faith that they are there.

 

What would be useful would be to make the formation positions mod-able: at least we could then experiment with more realistic formations; and see why we do not have them for ourselves.   

[Brano]

Multithreading is the answer. I remember first running Kingdom Come Deliverance with old 4cores/4threads CPU. This game has lots of NPCs with advanced AI and it put my CPU to its knees in almost every settlement. Switching to 6core/12threads CPU almost trippled my fps as the game can scale and utilise all of them. Imagine having 12 or 16 threads,where couple of them handle physics/FM/DM and the rest could be utilised by AI (8-12 threads). Now imagine one thread could handle 10 AI planes. We would have 80-120 AI available.

 I remember Danil wrote somewhere around DD120 that the engine runs physics and AI on common thread and it would need a significant effort to rewrite it.

[unreasonable]

That would give you more AIs for Storebror's server use, but it would not make a given AI fly any better.  So now we have:

 

1) Rewrite the engine to multithread

2) Rewrite the AI

 

.......

3) Don't hold your breath.

[ZachariasX]

1 hour ago, F/JG300_Gruber said:

Maybe if the simplified FM boundaries are set by the practical limits a good human player can get, instead of the theorical figures for a given airframe, it may prevent the possibility of such event ?

In the end, the computer will always gave to do some sort of trick to be able to competer with a good player. Question is, how obvious you make it that the compuet is tricking you.

 

In DCS, I find it very obvious BUT

1 hour ago, Quinte said:

In a 1on1, my dora vs AI p-51, I need to focus to win.

this is absolutely my impression as well. There is not just a downside to letting the computer to do tricks.

 

However, I would much prefer the AI deciding maneuvers based on basic air combat maneuvring after the book.

 

Regarding AI priorities, I remember the same as unreasonable

52 minutes ago, unreasonable said:

We have been told by Jason that he would like to spend more time on AI but cannot as he has no AI programmer.  

 

A problem is really that fixing the inner workings of a sim don‘t create revenue. Imagine Jason stating that the next update will bring an increased object and player count as well as better AI, nothing else, and this is 40 bucks per module you bought, else you stay with the old version.

Edited August 6, 2018 by ZachariasX

[Brano]

AI needs to be tought basic and advanced flight figures not by tapping ailerons and rudder like crazy. It must be memorised like humans do it = it should be scripted procedure of some sort IMO.

[ZachariasX]

12 minutes ago, Brano said:

I remember Danil wrote somewhere around DD120 that the engine runs physics and AI on common thread and it would need a significant effort to rewrite it.

I am not surprised. It took a long time for Lockheed Martin to take most load out of the main thread (besides asembling the scene) and having that multithreaded. Ten years and four versions. Still, geometry count and scene assembling on the main thread limit max. fps. But you can maintaining fps as soon as you are moving (autogen, moving objects) scales with core count now.

[unreasonable]

7 minutes ago, ZachariasX said:

 

 

A problem is really that fixing the inner workings of a sim don‘t create revenue. Imagine Jason stating that the next update will bring an increased object and player count as well as better AI, nothing else, and this is 40 bucks per module you bought, else you stay with the old version.

 

This is why this game needs to change it's revenue model to a subscription. Otherwise I suspect that it is unlikely that they will ever get off the hamster wheel of more planes, more maps etc to generate just enough revenue to make more planes and maps, but insufficient to fix the underlying flaws.

 

 

[ZachariasX]

Just now, unreasonable said:

 

This is why this game needs to change it's revenue model to a subscription. Otherwise I suspect that it is unlikely that they will ever get off the hamster wheel of more planes, more maps etc to generate just enough revenue to make more planes and maps, but insufficient to fix the underlying flaws.

 

 

Hard to swallow, but from the dev side surely the most convenient way.

[F/JG300_Gruber]

2 minutes ago, ZachariasX said:

However, I would much prefer the AI deciding maneuvers based on basic air combat maneuvring after the book.

 

How the AI responds to the situation is I believe depending on some algorythm used in combat situation. 

With a simplified FM, the programmer could spend more time improving the algorythms instead of teaching the AI how to deal with the complex nature of aerodynamics.

 

I don't think that it is impossible to create a simplified FM that will not put the AI at an unfair technical advantage over the player. It just needs to have the cursors set correctly.

 

 

[ZachariasX]

1 minute ago, F/JG300_Gruber said:

 

How the AI responds to the situation is I believe depending on some algorythm used in combat situation. 

With a simplified FM, the programmer could spend more time improving the algorythms instead of teaching the AI how to deal with the complex nature of aerodynamics.

 

I don't think that it is impossible to create a simplified FM that will not put the AI at an unfair technical advantage over the player. It just needs to have the cursors set correctly.

 

 

Agreed. In the end it will probably have to be a compromise in that direction to give „best results“.

[F/JG300_Gruber]

3 minutes ago, unreasonable said:

 

This is why this game needs to change it's revenue model to a subscription. Otherwise I suspect that it is unlikely that they will ever get off the hamster wheel of more planes, more maps etc to generate just enough revenue to make more planes and maps, but insufficient to fix the underlying flaws.

 

Meeeh, I think you are right, but there is a high risk to lose all the casual players and do more ill than good.

Since they announced BoBp I have my finger crossed that the highly popular planeset will finally make their finances take off and get out of the hamster wheel !

[Livai]

Last flight it happend again I damaged the Ai - leader bomber and all three bombers followed him and crashed all together.

 

5 minutes ago, Brano said:

Multithreading is the answer. I remember first running Kingdom Come Deliverance with old 4cores/4threads CPU. This game has lots of NPCs with advanced AI and it put my CPU to its knees in almost every settlement. Switching to 6core/12threads CPU almost trippled my fps as the game can scale and utilise all of them. Imagine having 12 or 16 threads,where couple of them handle physics/FM/DM and the rest could be utilised by AI (8-12 threads). Now imagine one thread could handle 10 AI planes. We would have 80-120 AI available.

 I remember Danil wrote somewhere around DD120 that the engine runs physics and AI on common thread and it would need a significant effort to rewrite it.

 

I wonder really since Hexa-core and Octa-core Cpus are very cheap and we still hanging around with odd Quad-core support.

If Quadcore can handle 20 AI planes ( what we have now ) so a Hexa-core could handle 30 AI planes and the Octa-core 40 AI planes what is double from them what we have now.........

 

-> Is the game not using LuaScripts? - Using Lua is inherently single-threaded

 

I think as long the Game use LuaScripts we can forget the part with multithreading

 

[SAS_Storebror]

14 minutes ago, Livai said:

we still hanging around with odd Quad-core support.

If Quadcore can handle 20 AI planes ( what we have now ) so a Hexa-core could handle 30 AI planes and the Octa-core 40 AI planes what is double from them what we have now

No sir.

What we have now is strictly single thread.

At least on the dedicated server that is.

The limit is 50 simulations per second (+/- 0.2 or so) and a tick delay of 20 milliseconds (actually it starts getting quirky at around 17-18 milliseconds tick delay).

If you compare those numbers with the CPU load caused by dserver.exe, you will soon notice that exactly at the time when dserver.exe reaches a load of 100% on one core/thread, you reach the magic 20 milliseconds tick delay (and that's a linear figure from 0 milliseconds tick delay = 0 % CPU load, over 10 milliseconds tick delay = 50% CPU load to 20 milliseconds tick delay = 100% CPU load, all adjusted to a single core and thread).

So anything that happens on a dedicated server right now, if it's relevant it happens on one single thread.

 

14 minutes ago, Livai said:

Is the game not using LuaScripts? - Using Lua is inherently single-threaded

I think as long the Game use LuaScripts we can forget the part with multithreading

*cough*

https://lualanes.github.io/lanes/

 

Mike

Edited August 6, 2018 by SAS_Storebror

[Algy-Lacey]

48 minutes ago, Karamazov said:

Simplified FM for AI bombers? Yes please.

 

This was implemented in the game  Battle Of Britain :  Wings Of Victory II, which I never played, but heard / read plenty about it's fantastic AI and intense single player dogfights. I believe that this game didn't have complex AI for the HUGE bomber formations of Heinkel's and Dornier's, I even seem to recall that until the player got really close to the bombers, there wasn't individual AI, just the bombers in a flight of four maybe, or even AI for each squadron (I really can't remember the details).

 

I think that if IL-2 Sturmovik Great Battles were to implement this kind of logic for dealing with masses of bombers (B-17's over the Ruhr anyone?) then we could keep the individual AI for fighters and ground attack aircraft but have a much less detailed AI for Level Bombers.

 

Does anyone else think this might be a good way forward towards having many hundreds of aircraft in a Bomber formation?

 

Quote

ZachariasX

I would propose that only planes entering the event horizon (however defined) should switch to complex control and management. Simplified AI vs simplyfied AI is again fair.

 

For example, if one was to fly with a formation of say 300+ bombers, only the ones that are within that event horizon should switch to complex FM, put up a real fight, and the rest can fly somplified and gunners can exchange hit probabilities. You‘re not seeing what happens at the other end of the formation, but if an AI formation attacks AI bombers there, you want an outcome that makes sense. And you shoukd see that outcome once you fly there.

 

This way, you could potentially have 500+ aircraft in one mission, but only maybe 40 or so will be in  „realistic“ mode, the rest will be CPU-friendly.

 

If I get a few 'upvotes' on this post, I may start a new thread on this idea.

What do y'all think of this idea?

Edited August 6, 2018 by Algy-Lacey
Want to include ZachariasX's post

[Thad]

4 hours ago, Quinte said:

That's sad to hear. I gave up long ago waiting on a flight sim with decent AI anyway, though. That's why MP is where it's at.

Salutations,

 

I agree with the assertion that there are a lot of problems with the games AI.

 

As a single player mission builder, I have often been puzzled by my AI wingmans' behavior. For example, on a ground attack mission versus a vehicle convoy. If I order my wingman to attack the nearest ground target, it normally will and properly. On the other hand, if I don't order it to the AI craft will just slowly spiral down and crash. WHAT the hell is that?

 

Now as portrayed in my first AI example the AI properly attacks the convoy but there are remaining convoy vehicles, I have to order the wingman 'again' to attack the nearest ground target.... and it will. If I don't, the AI will once again spiral down and needlessly crash. This is simply BS AI programming. No AI craft should needlessly fly into the ground like that! Most frustrating.

 

I can just imaging how messed up things become with AI controlled flights made up of three, four or more craft. ?

[Psyrion]

4 hours ago, unreasonable said:

 

This is why this game needs to change it's revenue model to a subscription. Otherwise I suspect that it is unlikely that they will ever get off the hamster wheel of more planes, more maps etc to generate just enough revenue to make more planes and maps, but insufficient to fix the underlying flaws.

 

 

No thanks. Subscription is a bad idea.

[ZachariasX]

4 hours ago, Algy-Lacey said:

If I get a few 'upvotes' on this post, I may start a new thread on this idea.

What do y'all think of this idea?

I think it would be a good idea. In the hope that people literate in programming such routines woukd join the conversation.

 

Indeed, BoB Wings of Victory is probably the reference for AI in air combat.

[unreasonable]

1 hour ago, Thad said:

Salutations,

 

I agree with the assertion that there are a lot of problems with the games AI.

 

As a single player mission builder, I have often been puzzled by my AI wingmans' behavior. For example, on a ground attack mission versus a vehicle convoy. If I order my wingman to attack the nearest ground target, it normally will and properly. On the other hand, if I don't order it to the AI craft will just slowly spiral down and crash. WHAT the hell is that?

 

Now as portrayed in my first AI example the AI properly attacks the convoy but there are remaining convoy vehicles, I have to order the wingman 'again' to attack the nearest ground target.... and it will. If I don't, the AI will once again spiral down and needlessly crash. This is simply BS AI programming. No AI craft should needlessly fly into the ground like that! Most frustrating.

 

I can just imaging how messed up things become with AI controlled flights made up of three, four or more craft. ?

 

Odd - playing career lately and getting ground attack missions (in 109s with me as a wingman, not the flight leader) and the AI all carryout attacks perfectly, circle and repeat until targets destroyed or out of ammo. Then they go home. I have never seen anything similar to your description. Maybe grab a career ground attack mission and see in the editor if you have missed something.

 

On the other hand, what you describe is very similar to what happens if you as leader approach your home field and get the approach clearance.  With an AI leader they all turn on lights and start their normal pattern. With me leading they mill about until I have actually landed and then start the pattern. Very disconcerting to have your wingman fly 2 meters over your head while on final. edit - maybe I have to order RTB.  In my experience the AI controlled flights generally cause less  problems than the player controlled flights, leaving aside the odd silly collision - I know they happened but not that often.

Edited August 6, 2018 by unreasonable

[Psyrion]

41 minutes ago, unreasonable said:

maybe I have to order RTB.  

That's what I tend to do. One thing that bothers me on landing that a plane that is flying its landing patter doesnt react at all to being atacked sadly.

[unreasonable]

57 minutes ago, Psyrion said:

That's what I tend to do. One thing that bothers me on landing that a plane that is flying its landing patter doesnt react at all to being atacked sadly.

 

It is also a good way of getting a vector to base if you have got lost   Easy enough to do on the Moscow map I find.  But yes, as soon as the landing lights are they are helpless. 

 

[SAS_Storebror]

19 hours ago, Psyrion said:

23 hours ago, unreasonable said:

This is why this game needs to change it's revenue model to a subscription.

No thanks. Subscription is a bad idea.

 

I think it's a brilliant idea.

 

Mike

[Yogiflight]

18 hours ago, unreasonable said:

It is also a good way of getting a vector to base if you have got lost   Easy enough to do on the Moscow map I find.  But yes, as soon as the landing lights are they are helpless. 

The other side, as soon as you give this command, your squadmates won't react on anything anymore. They will fly back to base and land, but won't hear any further commands and won't attack any enemies. It is as if they already would have turned on the nav. lights.

[unreasonable]

53 minutes ago, Yogiflight said:

The other side, as soon as you give this command, your squadmates won't react on anything anymore. They will fly back to base and land, but won't hear any further commands and won't attack any enemies. It is as if they already would have turned on the nav. lights.

 

The other disadvantage is that they do not space out properly for the landing. If you give RTB when the planes are close together- say they have kept in good formation for once - they all approach and try to land individually without waiting in turn. I have seen three planes land almost nose to tail and all pile up on the runway.  Not sure if  or  is more appropriate.

 

At least in BoX, unlike RoF, planes ordered to RTB do not routinely crash into hangers.

 

 

[Mac_Messer]

1 hour ago, unreasonable said:

At least in BoX, unlike RoF, planes ordered to RTB do not routinely crash into hangers.

 

 

Well uuuh...you know. As long as they`re not accounted for losses in mission debriefing, it`s ok. I think ?