[AI] Observations, Wishes & Proposed Solutions

[HexCode]

INTENT & UTILITY
===================

This topic will be dedicated to posts and, hopefully, follow-up discussions directly aimed at identifying and practically dealing with PGF AI's various "ways".

Although a subject's genesis may be traced back to some [DEV], even [EPH], discussion, the objective here is to come up with reasonably solid findings applicable across the board.

There's no reason to adhere to some strict, sequential coverage of subjects. Not all subjects are equally tractable. However, when posting about a particular subject, it's always helpful to specifically entitle follow-up posts so as to make it easier for participants to follow the particular discussion.

The topic's title starts with "[AI]". Essentially, the topic is akin to a [DEV] one, the only difference being that, instead of some particular mod, generalizable knowledge is being developed, comprising a diverse set of AI-specific behaviors and content designers' responses to them.

Readers who aren't intimately familiar with PGF AI behavior's ins and outs may want to spend some time looking in here first:

[VP] AI Module Behavior
viewtopic.php?f=100&t=532


Note: There's a good chance that some of this topic's contents will be incorporated here:

[VM] AI Module: Behavioral Improvements
viewtopic.php?f=100&t=599

in due course.

[Lettos]

INTENT & UTILITY
===================

This topic will be dedicated to posts and, hopefully, follow-up discussions directly aimed at identifying and practically dealing with PGF AI's various "ways".

Although a subject's genesis may be traced back to some [DEV], even [EPH], discussion, the objective here is to come up with reasonably solid findings applicable across the board.

There's no reason to adhere to some strict, sequential coverage of subjects. Not all subjects are equally tractable. However, when posting about a particular subject, it's always helpful to specifically entitle follow-up posts so as to make it easier for participants to follow the particular discussion.

The topic's title starts with "[AI]". Essentially, the topic is akin to a [DEV] one, the only difference being that, instead of some particular mod, generalizable knowledge is being developed, comprising a diverse set of AI-specific behaviors and content designers' responses to them.

Readers who aren't intimately familiar with PGF AI behavior's ins and outs may want to spend some time looking in here first:

[VP] AI Module Behavior
viewtopic.php?f=100&t=532

I'll be honest with you. It was your reasoning about the reasonableness and behavior of the AI that was the determining factor for me when I made the decision to register on the forum.

It's not interesting to live an uninteresting life. And here is such a field of activity - a whole wise cunning AI!

You wrote it all right. It's a good start. Then I did a little research on this AI beast. Didn't discover anything new or good. On the contrary, this beast appeared before us in all its terrible nakedness.

Which I did in the last few days between gardening and other chores:
Played the Stalingrad scenario in PGF as Allied in defense against the dumb AI once more and tried to write down all my actions, dividing them by unit class. The first four turns are pure defense tactics. I have 3 A4 sheets of handwritten notes, and I need some time to publish them.
It's an interesting algorithm of actions
I'm not talking about offensive, and the hardest part is the transition from defense to offensive or vice versa. This will then be spread out over 24 sheets of A4
The most curious thing is that all the defensive actions are quite easy to implement in the program code. It would be nice if somebody would like to test this algorithm a lot of times in different scenarios, improve and add various tricks, and start creating a more or less "hard" AI

[HexCode]

Observations

PGF's AI Module is akin to a deaf player...

It's not interesting to live an uninteresting life . . . I did a little research on this AI beast. Didn't discover anything new or good. On the contrary, this beast appeared before us in all its terrible nakedness.

Agreed !

Wishes

... all the defensive actions are quite easy to implement in the program code.

So far, so good...

Proposed, Practical Solutions ?

I see none on the horizon. PGF's source code isn't "publicly" accessible. If I were a betting man, I'd predict that it'll never be...

[Lettos]

Four simple improvements to this ... Absolute Idiot (AI)

1) Using WAW as an example. Radoye has already taken the important step of moving the self-propelled guns, aka sometimes Assault guns, into the ATG class.
But they are rather uncomfortable there, in ATG. They hit without loss, but unfortunately... last in half-turn.

I'm only interested in land-based scenarios. AI at sea - see his full name...
In land scenarios, convert all Assault guns to Destroyer class 12. After that, these guns finally start performing their striking function on the battlefield!
By the way, you can even assign them an ENT level at the beginning of the scenario. However, if they move and then freeze again for a few turns, they get no ENT boost. I'm not too worried about that, though. It's much more interesting to have a strong offensive opponent against you.

All enemy ground units in the scenario should probably just copy the Hard Attack to column Naval Attack parameter.

I'm a bit unsure whether this should be done at all. It's hard to tell from experimentation whether the nature of combat encounters between units has changed. It seems to me now, after a few experiments, that it's not even necessary to change the attack parameter, since the target type remains the same, Hard/Soft and not Naval.
If the Destroyer class can have ENT at the beginning of the scenario, then ... he's really a different type of target after changing Class, or not?

2) Move the large caliber ATY to the Capital Ships class 13. All calibers from 150mm and above, and from 150mm all long range. They were never engaged in direct infantry support on the battlefield anyway. Long-range accuracy did not allow them to do so, as accuracy within 50 meters is important on the battlefield.

There is even a counter-battery fight going on!

As for the loss of ENT, I don't think you need to worry. Or how do you imagine any serious ENT and camouflage of a metal thing like this in field positions: https://en.wikipedia.org/wiki/21_cm_M%C3%B6rser_18
In my opinion, ENT is, within the PGF game model, the category of those weapons that can be cunningly hidden ... behind that bush in the trench over there.

3) Move some fighters to the Tac Bombers class. Conventionally, these would be the same fighters as the Fighters class, at the same prices, with the same parameters, just with an additional marking in the name like "Hunters" (HUN, KIL etc.)
This would give the AI a chance to attack part of the fighters first, and only then move the bombers. Since the AI hardly ever escorts its bombers anyway, there won't be any fundamental changes here. However, the AI's behaviour will become more dangerous for the player.
By the way, you could give the Hunters a bit more FUEL and MA.

4) The AI now puts its trucks and armored vehicles with infantry under attack by tanks at the infantry movement stage. Infantry moves last and enters all possible vacant hexes after own tank and artillery strikes. I am tired of killing them several at a turn in the Stalingrad, Kiev, Moscow scenarios! Why give infantry transports, which in the realities of PGF AI deliver them as quickly as possible not to the victorious hex, but to the afterlife?
If you take away the transports from the infantry, at least they have a chance to fight and survive.

Question is - in Axis scenarios such as North Africa, Anzio, etc., do the Allied infantry by any chance like to ride under the Tigers? In these scenarios, is the problem not relevant? I think it should be relevant.

By the way, have you ever thought about how long it takes an infantryman to leave an armoured personnel carrier? Why not make the infantry unit "the same but on armoured personnel carriers" at once, and which doesn't have the option of having Organic Transport?
An armoured personnel carrier is not just an "iron" truck, whatever SSI thinks about it.
MVT=6-7 for an infantry unit after an upgrade, and go ahead, with no transport mode!

And it will be easier for the AI to imitate reality...

[HexCode]

Observations

# Lettos # has done extensive empirical testing to determine how PGF's AI Module goes about performing various "acts" with the units it leads. Focusing on movement and attacks, he concluded that:

1) The AI Module "acts" on behalf of its units on a per Unit Class priority basis. Once it finishes "acting" within the context of a particular Unit Class, it moves on to "acting" on behalf of the units of the Unit Class which is assigned the next lower priority. At the end of each unit's movement phase the AI may or not initiate an attack. Does it ever attack prior to actually moving, though ?

2) Transported units "act" on the basis of their particular Unit Class.

3) The following priorities apply to the THREE (3) Unit Super-Classes:

Air
Naval
Ground

4) The following priorities apply within the FOUR (4) Air Unit Classes:

Tactical Bomber
Level Bomber
Fighter
Air Transport (self contained)

5) The following priorities apply within the FIVE (5) Naval Unit Classes:

Destroyer
Submarine
Capital Ship
Aircraft Carrier
Naval Transport (self contained)

6) The following priorities apply within the NINE (9) Ground Unit Classes:

Reconnaissance
Tank
Anti-Aircraft
Artillery
Air Defense
Infantry
Anti-Tank

Structure (i.e., Forts)
Organic Transport (self contained)

? ? ?

Wishes & Proposed Solutions

There's plenty of time for that.

[Lettos]

At the end of each unit's movement phase the AI may or not initiate an attack. Does it ever attack prior to actually moving, though ?

Yes, AI can attack first and only then move. Just checked in the Stalingrad scenario in AI vs AI mode.

It is interesting to observe that movement after an attack occurs either to one's rear, obviously for resupply, or one hex "forward", taking the place of an enemy unit destroyed by the attack.
Attacking and moving forward 5-6 hexes after an attack is something the AI does not understand. All rushes forward to a full MA unit take place before the attack. Big difference from the Blitzkrieg logic...

[Lettos]

Continuing and elaborating on the theme

Four simple improvements to this ... Absolute Idiot (AI)

1) Using WAW as an example. Radoye has already taken the important step of moving the self-propelled guns, aka sometimes Assault guns, into the ATG class.
But they are rather uncomfortable there, in ATG. They hit without loss, but unfortunately... last in half-turn.

I'm only interested in land-based scenarios. AI at sea - see his full name...
In land scenarios, convert all Assault guns to Destroyer class 12. After that, these guns finally start performing their striking function on the battlefield!
By the way, you can even assign them an ENT level at the beginning of the scenario. However, if they move and then freeze again for a few turns, they get no ENT boost. I'm not too worried about that, though. It's much more interesting to have a strong offensive opponent against you.

All enemy ground units in the scenario should probably just copy the Hard Attack to column Naval Attack parameter.

I'm a bit unsure whether this should be done at all. It's hard to tell from experimentation whether the nature of combat encounters between units has changed. It seems to me now, after a few experiments, that it's not even necessary to change the attack parameter, since the target type remains the same, Hard/Soft and not Naval.
If the Destroyer class can have ENT at the beginning of the scenario, then ... he's really a different type of target after changing Class, or not?

You could keep it simple and put Assault Guns R=1 in the Tanks class.
We know from sandbox experiments that units with a lower MA in their class will move first.
If Assault Guns have MA=3-4, this ensures that they will strike before the faster moving tanks.

3) Move some fighters to the Tac Bombers class. Conventionally, these would be the same fighters as the Fighters class, at the same prices, with the same parameters, just with an additional marking in the name like "Hunters" (HUN, KIL etc.)
This would give the AI a chance to attack part of the fighters first, and only then move the bombers. Since the AI hardly ever escorts its bombers anyway, there won't be any fundamental changes here. However, the AI's behaviour will become more dangerous for the player.
By the way, you could give the Hunters a bit more FUEL and MA.

Knowing that the units with the smaller MA will be the first to move, there is not much benefit from the above suggestion. Or it's just another partial patch based on the "it can't get any worse, but it can get better" principle. At least there is a chance that some Fighters will hit before Level Bombers and some Tac Bombers.

4) The AI now puts its trucks and armored vehicles with infantry under attack by tanks at the infantry movement stage. Infantry moves last and enters all possible vacant hexes after own tank and artillery strikes. I am tired of killing them several at a turn in the Stalingrad, Kiev, Moscow scenarios! Why give infantry transports, which in the realities of PGF AI deliver them as quickly as possible not to the victorious hex, but to the afterlife?
If you take away the transports from the infantry, at least they have a chance to fight and survive.

Tested this idea in practice.

Scenario Stalingrad WAW.
AI starting PP = 1000. Core limit = 1 unit, i.e. no new units for AI.
Allied PP = 0 at start and 0 per turn. Core limit = 20, i.e. no any new units.

Human wins against AI(AXIS) Advanced in 9-10 moves.
I removed all those trucks and armoured transports from both sides, and assigned units MA and MT according to the transport they had.
The changes affected four classes: INF, ATY, ATG, AD.
For ATY, I made MA=4 Wheeled for the experiment.

I also moved all light tanks and armoured vehicles to the infantry class, and ATG to the tank class.

In order of movement it turned out like this:

OLD - NEW.
Recon - Assault Guns.
Tank - ATG and Medium and Heavy tanks
AA - AA
ATY - ATY
AD - AD
INF with "embedded" transport, Light tanks and Recons
ATG - empty

The AI on the offensive was still an idiot, but no longer an absolute one. I had to fight him for 13-14 turns. By the way, at one point my front line was almost broken, and I had no reserves in the rear at all. But the front line held. I would like to point out that in the scenario with the original Equipment file I had no fear of the AI breaking through at all.
So, thanks to some injections, the AI could become much more intelligent, by about 50-70%.

More observations about this battlefield, and about the embedded transports, in the following posts.

[Radoye]

I was considering adding a Panzergrenadiere type of mechanized infantry unit where the organic transport is embedded into the unit itself (also similar units for other nations where appropriate). These will likely replace the "HW" types of units. Right now this is on hold because there are no suitable icons available but i can make those.

On the other hand, i am not convinced about doing the same for AT / ATY / AD because it will erase any distinction between towed and self propelled units.

[Lettos]

I was considering adding a Panzergrenadiere type of mechanized infantry unit where the organic transport is embedded into the unit itself (also similar units for other nations where appropriate). These will likely replace the "HW" types of units. Right now this is on hold because there are no suitable icons available but i can make those.

Yes. Again we come to the idea of a complex unit.
One armoured personnel carrier
https://en.wikipedia.org/wiki/Sd.Kfz._251
had two MG34/42 machine guns. Per battalion - about 50-60 armoured personnel carriers. That's 120 machine guns, an awful lot of power!

Speaking of infantry trucks... well, the infantry didn't sit in the trucks all night until morning, did they? Jumping out of a truck, that's pretty quick, isn't it?
It's just that someone at SSI wanted to play with nice toy cars. But they're all grown up now, aren't they? Why do we need these cars, moreover, serving exactly one turn in the AI?

If infantry is transferred to trucks, it gets more mobility, it does not gain an additional bonus to attack, but it loses a little in Defense. Trucks have to be guarded. You can't drive a truck close to the battlefield. And if you do, you risk being ambushed.

If the infantry gets armored vehicles, they gain more mobility, gain a very significant bonus to Soft Attack and Air Attack and Air Defense, and also gain a bonus to Ground Defense.

There is plenty of information on the cost of armoured personnel carriers and trucks on the internet. It is not difficult to calculate how much it cost to upgrade one battalion's armoured personnel carriers compared to its small arms, and it is also fair to compare the cost of trucks and armoured personnel carriers. I will write about this later.

Added later
I wrote here:
Re: [DEV] Cost of weapons
viewtopic.php?f=95&t=467&start=150#p10246
Re: [DEV] Cost of weapons - INF Battalion
viewtopic.php?f=95&t=467&start=150#p10248

On the other hand, i am not convinced about doing the same for AT / ATY / AD because it will erase any distinction between towed and self propelled units.

Self-propelled gun has much more Ground and Air Defense, and much more INI compared to Towed ATY.

Artillery movement in the SSI model looks like this: transport the guns to their destination, and go to sleep. All guns stay attached to trucks and all-terrain vehicles. Tomorrow we'll unhook them from the transports and shoot!

In AI Offensive scenario even ATY with embedded transport does not live long either. When testing Stalingrad, only one gun out of four survived until turn 11. But this is already a very good result!
And if we leave the Towed+Transport scheme, AXIS are guaranteed to be without artillery by turn 3.
Tired of looking at it

[Lettos]

In the Offensive scenario, the AI does not replenish units with elite replacements. Only regular ones, regardless of how much STR a unit is replenished. This behavior is demonstrated by the AI at all three levels.

This is a very serious argument for thinking carefully about creating Offensive AI Scenarios. If the offensive survives a hit within the first 3-4 turns, and the AI has no PP and no ability to buy new N-star units, then further defense becomes a very simple matter.

[Lettos]

Observations

In the Offensive scenario, the AI does not replenish units with elite replacements. Only regular ones, regardless of how much STR a unit is replenished. This behavior is demonstrated by the AI at all three levels.

Does this also mean that PGF's AI Module totally ignores how many PPs it has at its disposal ? In other words, would it still abstain from granting its under-strength units Elite Replacements even if it were... swimming in a gazillion PPs ?

I'm not quite sure about the gazillion, but I think the AI will spit on it with his idiot spit as much as he did on the thousand.

Do a simple experiment. STALINGRAD WAW scenario. AI AXIS. Give it a few gazillions thousands on 1 turn in file 028.pgscn.
Adjust Spotting of one of Your unit , for example KV-1/42, to 70.
You don't have to play as Allied. Just don't move your units for three-four turns, or hit there just a couple of times so the AI has something to replenish. And see what the AI will turn his units assigned to the rear for replacements.

Another oddity. The AI Offensive does not replenish its units during first two turns. It's not depends from how much cities AI have occupied.

It looks like the AI has some dumb rule like: hit two-three turns, then move to rear to replenish. Accordingly, the Air unit reaches its airfield on the third, and more often on the fourth turn. That's where these half-dead after AD attacks units should be caught and hit with a fly swatter from inexperienced Fighters.

In Defense scenarios, the AI is much smarter. It replenishes units from the first turn, and uses elite reinforcements.

[Lettos]

Another oddity. The AI Offensive does not replenish its units during first two turns. It's not depends from how much cities AI have occupied.

Misconception!

I finally saw one unit get reinforcements from the AI on the second turn.
It was an AD unit, STR2>STR7. Usually the AI does not like to replenish units not to "10" at once. But here it happened. The AI had 1 million PP, but made Replacements non-elite.

[Lettos]

Observations

Usually the AI does not like to replenish units not to "10" at once. But here it happened. The AI had 1 million PP...

Let me make sure I perfectly understand this. Are you saying that PGF's AI Module replaces Strength Factors (SFs) à la carte ? If so, it violates the play system's provisions regarding such actions; at least the ones that apply to human players...

Basically, an under-strength unit not subject to enemy unit adjacency or terrain restrictions (e.g., desert) must be brought up to SF TEN (10) in one fell swoop, provided there are sufficient PPs to accomplish this. Are you absolutely sure about your experimentation findings ? Alternatively, I may be misunderstanding you.

You get it right. I was the one who was inaccurate.
In this case, the AI sent reinforcements to the AD unit and the Allied unit was in the adjacent hex. Therefore the STR increased only to 7. Usually the AI does not send units if he cannot replenish them to 10 at once. But in this case, the AI for some reason decided to replenish a unit that could not be replenished to 10.

[Lettos]

Observations

Ok, given a PGF AI Module's attack posture, when it comes to replacing its unit Strength Factors (SFs), it does so with Regular Replacements even in instances where it does have plenty of PPs to grant those units Elite Replacements instead.

Wishes

Can something be done about this oddity ?

Proposed Solutions

From the Library:

Attack Posture: The only "remedy" I know of is both crude as well as totally historically counterfactual. Namely, the AI Module directs the fates of many, many prepositioned units to begin with. Invariably, these units are being assigned grossly inflated Experience Levels and Strength Factors... A "slugfest" ensues; at least for a while !

Forget about representational historicity. Start off experimenting with AI-led units having TWENTY (20) SFs and Experience Level FIVE (5). See how long the "slugfest" meaningfully lasts and calibrate "things" accordingly. It's all about brute force and little else...

Yes, that's right!

There can be no historicity in Offensive AI scenarios.
Except as in the memoirs of many Wehrmacht officers: only hordes of scary Allied, preferably with brutal Asian faces.

To give these hordes something meaningful in the AI's execution, in my opinion, you should give prestige not a little bit each turn, but in large portions once in several turns. Then the AI won't be able to use it up and buy a lot of scary units. This is roughly how it was done in the SSI Kursk PG1 scenario for Allied. In this case, the fact that Allied were Defending side does not affect the very method of solving the problem.
SSI knew their AI, saw him still in the egg shell, and that's why they made such batch stimulation of the AI.
It would be nice for us to adopt their method, too!

[HexCode]

Ok, let's forget historicity. Therefore, one is forced to design custom content for "intended play quality effect". I see two general approaches here; not mutually exclusive:

Approach #1

Shower PGF's AI Module with lots of PPs. This will enable it to keep on purchasing new units and to be pushing them forward relentlessly.

Approach #2

Grant the AI Module side's prepositioned units Experience Level, say, FIVE (5) status and Strength Factor, say, TWENTY (20) status. As play progresses, their losses will be automatically replaced by Regular Replacements. This means that their lethality will be diminishing over time. Combining the preceding suggestion with:

a) Appropriate PP availability (not necessarily large);

b) Ensuring, perhaps, that newly purchased units sport a more "reasonable" Experience Level;

c) Tinkering with file EQUIPMENT.PGEQP so as to render newly purchased unit type possibilities highly targeted;

d) Indirectly controlling the number of newly purchased units available to the AI Module;

Might increase play quality a bit. Yeah, a bit !

[HexCode]

Observations

PGF's AI Module:

1) Upon losing Air Super-Class units, it NEVER purchases new such units.

2) Upon scenario's commencement, even if it's given a Defensive Posture, it immediately rushes its Air Super-Class units forward under both "Advanced" and "Intermediate" Level settings. Mind you, under an "Intermediate" Level setting, it sometimes retreats one or two such units and keeps them "around" for later action...

Wishes

My young nephew dukes it out with the AI Module if and only if the latter:

--- Is given a Defensive Posture; and
--- Plays at the "Intermediate" Level.

Now, he's gotten sick and tired of eliminating the AI Module's air assets right off the bat, thereby achieving relatively unchallenged air superiority over the battlefield early on. He asked me if "something" could be done about this "problem"...

Proposed Solution

I've decided to give my nephew a... hornets' nest !

A) Armed with the contents of

Dual-Mode, Composite Units
viewtopic.php?f=100&t=540#p8965

DMCU Mount / Dismount & Movement Capabilities
viewtopic.php?f=100&t=543#p12052

"Anchored" Garrison Units
viewtopic.php?f=100&t=540#p8971

I "anchored" some of the AI Module's Fighter Class units over its Airfield hexes which either are objective hexes themselves or are adjacent to objective hexes. To boot, armed with the contents of

Prepositioned Units: "Unorthodox" Entrenchments
viewtopic.php?f=100&t=540#p12142

I gave those units Entrenchment Levels of FIVE (5). Oh, before I forget, I also placed "respectable" Air Defense units directly below the units.... My nephew has been telling me that the emerging challenge is on a par with his attempts to... bed extremely "hard-to-get" females !

Fly in the Ointment: Beware, "anchoring" fails in all instances where the Air Super-Class unit is adjacent to Body of Water, Port or Embarkation City hexes.

[HexCode]

Observations

When given a Defensive Posture, PGF's AI Module just does NOT "understand" how to delay a human player's rapid advance all along a reasonably broad front. It just "corrals" every available friendly unit around and moves it in the immediate vicinity of some friendly objective hex.

Wishes

My young nephew dukes it out with the AI Module if and only if the latter:

--- Is given a Defensive Posture; and
--- Plays at the "Intermediate" Level.

Once again, he's gotten sick and tired of rushing forward his armored and mounted units without encountering any credible opposition. He asked me if "something" could be done about this "problem"...

Proposed Solution

Here's my rather simple, practical solution. I prepositioned a sufficient number of AI-controlled, static units on the map fully aimed at blocking the most obvious avenues of enemy advancement. These units often belong to the Anti-Tank Class, can inflict serious damage on enemy armored and mounted units, and are granted Entrenchment Level NINE (9) status right from the get go. Basically, if the human player were to attempt to eliminate too, too many of them, he'd be wasting lots of time while the scenario clock would be mercilessly ticking away... So, the human player will be forced to engage in efficient, bypassing maneuvers thereby significantly enriching his play experience which would otherwise amount to serially and monotonously / tediously laying siege to and capturing objective hexes invariably ringed by numerous enemy units.

[Mustang19]

Observations

When given a Defensive Posture, PGF's AI Module just does NOT "understand" how to delay a human player's rapid advance all along a reasonably broad front. It just "corrals" every available friendly unit around and moves it in the immediate vicinity of some friendly objective hex.

Wishes

My young nephew dukes it out with the AI Module if and only if the latter:

--- Is given a Defensive Posture; and
--- Plays at the "Intermediate" Level.

Once again, he's gotten sick and tired of rushing forward his armored and mounted units without encountering any credible opposition. He asked me if "something" could be done about this "problem"...

Proposed Solution

Here's my rather simple, practical solution. I prepositioned a sufficient number of AI-controlled, static units on the map fully aimed at blocking the most obvious avenues of enemy advancement. These units often belong to the Anti-Tank Class, can inflict serious damage on enemy armored and mounted units, and are granted Entrenchment Level NINE (9) status right from the get go. Basically, if the human player were to attempt to eliminate too, too many of them, he'd be wasting lots of time while the scenario clock would be mercilessly ticking away... So, the human player will be forced to engage in efficient, bypassing maneuvers thereby significantly enriching his play experience which would otherwise amount to serially and monotonously / tediously laying siege to and capturing objective hexes invariably ringed by numerous enemy units.

The ring behavior is what makes the game brutally hard and borderline impossible. I'm not complaining though.

If the ai was "hard" it would construct a impassable wall of 60 point AA and AT guns. The AI only goal is to slow the player, it's prestige works differently. From AI perspective immobile defensive units are worth much more than to the player.

The original PG had some sort of budgeting where AI balanced purchases, if it just lumps everything into a raw prestige value then the spam behavior occurs.

[HexCode]

The original PG had some sort of budgeting where AI balanced purchases, if it just lumps everything into a raw prestige value then the spam behavior occurs.

PGF's Developer / Programmer was fully aware of SSI's Prestige Management algorithms; Prima's PG Official Strategy Guide documents the underlying math in great detail... So, for whatever reason, PGF's Developer / Programmer introduced his very own Prestige Management algorithms. Interestingly enough, it appears that his "development buddies" never gave a hoot, let alone, object to the emerging "improvement"; at least not "publicly"...

The key idea behind SSI's Prestige Management algorithms is to periodically reset the AI's prestige to ZERO (0). On the other hand, PGF's Prestige Management algorithms are strictly, algebraically, cumulatively additive...

[HexCode]

Absolute Prerequisite

[AI] Offensive "Slugfests": Two Approaches
viewtopic.php?f=95&t=597#p10284

Addendum

There's a kind of scenario where the attacking AI need not enjoy ridiculously overwhelming numerical or qualitative superiority over the human player's forces and STILL be somewhat challenging for the player. However, it calls for a content designer to move away from SSI-like historical themes and specificities and embrace PGF's play system as a "pure", ahistorical strategy wargame.

The requisite scenario map should feature body of water or clear / rough terrain expanses containing very few (if any) terrain "obstacles" such as islands, mountains etc. All one needs to do is grant the AI forces moderate quantitative or qualitative superiority over the human player's forces. Judicious placement of objective hexes (and airfields) along the map's certain edges will "instruct" the AI to meaningfully attack by thrusting its units in a "reasonable" direction.

Remember: PGF's AI module does very few "things" reasonably well. To this effect, content designers may wish to adopt a targeted, minimalist approach to scenario construction and just limit any underlying complexity to the human player's ensemble of usable features and degrees of play freedom. And yeah, adherence to historical specificities (i.e., SSI-like content) mightily degrades the AI module's battlefield performance.

[Radoye]

To compensate for all that, I'd greatly increase their fuel capacities to ensure such air units keep on flying around until the very end without needing to refuel.

What happens if you set fuel to 0, like with air transports?

[HexCode]

What happens if you set fuel to 0, like with air transports?

Aha, I see you're quite alert on a... Sunday ! Yes, this is the elegant, global, technical solution. However, this is a pilot... discussion. Let me push "things" a bit forward. The less elegant solution might be the one to adopt under certain specific design circumstances. For example, I can visualize some scenario where the AI doesn't get a completely free pass here. In other words, at some point during play, AI-controlled air-units may start plunging to their doom; by design...

One thing is pretty obvious here:

PGF's AI module does very few "things" reasonably well. To this effect, content designers may wish to adopt a targeted, minimalist approach to scenario construction and just limit any underlying complexity to the human player's ensemble of usable features and degrees of play freedom.

SO, I'm looking forward to exposing human players to quite a bit of complexity. Extreme asymmetry is the order of the day...

[Lettos]

2) Upon scenario's commencement, even if it's given a Defensive Posture, it immediately rushes its Air Super-Class units forward under both "Advanced" and "Intermediate" Level settings. Mind you, under an "Intermediate" Level setting, it sometimes retreats one or two such units and keeps them "around" for later action...

I have a desire to play an entire campaign with AI Basic. At least 8-10 scenarios to see the difference in AI behavior in the air.

I "anchored" some of the AI Module's Fighter Class units over its Airfield hexes which either are objective hexes themselves or are adjacent to objective hexes. To boot, armed with the contents of

Prepositioned Units: "Unorthodox" Entrenchments
viewtopic.php?f=100&t=540#p12142

I gave those units Entrenchment Levels of FIVE (5). Oh, before I forget, I also placed "respectable" Air Defense units directly below the units.... My nephew has been telling me that the emerging challenge is on a par with his attempts to... bed extremely "hard-to-get" females !

Fly in the Ointment: Beware, "anchoring" fails in all instances where the Air Super-Class unit is adjacent to Body of Water, Port or Embarkation City hexes.

You can make Fighter-class unit MVT=1 and FUEL=1 or 2. You get such a "dog on a chain", guarding a small area around the airfield.

Let me give you an example of what I mean "for effect". Very often, AI-controlled air units run out of fuel and plunge to their doom. Relatedly, they don't "recognize" friendly Aircraft Carrier Class units. To compensate for all that, I'd greatly increase their fuel capacities to ensure such air units keep on flying around until the very end without needing to refuel.

I too have come to the conclusion that to balance air warfare in a scenario requires giving AI a handicap in both the number of air units and their EXP and STR, and FUEL too. A sort of "AI stupidity factor" of about 1.5.
And it is necessary to check by practice: unlimited AI FUEL can lead to creation of eternal "blockade rings" in the air.

[Lettos]

Defensive scenario.

Observations of AI actions.

First turn: somewhere far out in the ocean there is a group of AI ships and transports.

1) Regardless of whether the AI sees a group of player's ships somewhere far out on the flank (out of reach of his ships in one turn), he will move his ships in a straight line to his nearest Victory Hex on the coast to defend it. There will be no ship actions against the player's fleet. But the AI may send its air force in the direction of the player's fleet.
If the player removes the ships from the AI spotting area, the AI aviation will not search for the "missing" ships, but will return in the direction of its airfield.
Intermediate conclusion: if we want more aggressive AI actions, maybe it makes sense to increase the Spotting AI ships?

2) The AI doesn't understand the concept of concentrating forces in a single group at all, and absolutely does not understand the task of guarding transports.
Conclusion: different speed of ships, from 4 to 8 - it is beautiful, but harmful for modeling in the scenario of AI squadron movement. The player, already smart, gets to destroy AI ships one by one. The speed of all AI surface ships, except for coastal gunboats, should be the same in order to maintain the naval order that the scenario author has in mind.

3) If there are AI troop transports at sea, they head for the nearest VH (obviously a port), keeping as close to shore as possible if the route allows. A port inherently has some bordering land, and some sea hexes. If the port's sea hexes are vacant, the transports will occupy them, closing the AI's favorite defensive ring of 6 units around the VH, and will never in this scenario be able to unload.
Solution to the problem: at the start of the scenario, place stationary ships on the sea hexes near the port. It can be a Cargo vessel with zero attack parameters, or some ship, for example, as a stationary air defense battery.
Separating the port from the sea with a strip of Neutral Hexes also brings results. There is a disadvantage here - an AI ship will not be able to enter the port to defend it. However, this disadvantage is rather controversial. Do you really like a torpedo boat or submarine defending a port against a tank attack? Or when a company of cyclists attacks a battleship in port? Or a situation where ships are huddled around an anti-aircraft gun standing in a port or near it, creating a nearly indestructible group of targets?
Why in some cases would ships not be one hex away from that anti-aircraft gun?

[Lettos]

Observations

# Lettos # has done extensive empirical testing to determine how PGF's AI Module goes about performing various "acts" with the units it leads. Focusing on movement and attacks, he concluded that:

1) The AI Module "acts" on behalf of its units on a per Unit Class priority basis. Once it finishes "acting" within the context of a particular Unit Class, it moves on to "acting" on behalf of the units of the Unit Class which is assigned the next lower priority. At the end of each unit's movement phase the AI may or not initiate an attack. Does it ever attack prior to actually moving, though ?

2) Transported units "act" on the basis of their particular Unit Class.

3) The following priorities apply to the THREE (3) Unit Super-Classes:

Air
Naval
Ground

4) The following priorities apply within the FOUR (4) Air Unit Classes:

Tactical Bomber
Level Bomber
Fighter
Air Transport (self contained)

5) The following priorities apply within the FIVE (5) Naval Unit Classes:

Destroyer
Submarine
Capital Ship
Aircraft Carrier
Naval Transport (self contained)

6) The following priorities apply within the NINE (9) Ground Unit Classes:

Reconnaissance
Tank
Anti-Aircraft
Artillery
Air Defense
Infantry
Anti-Tank

Structure (i.e., Forts)
Organic Transport (self contained)

? ? ?

Wishes & Proposed Solutions

There's plenty of time for that.

So, two and a half years have passed, and it's timely to revisit the topic of prioritizing movement and attack.

Fort attacks and moves first of the ground units. Attack first, then movement. Just like ARTY.
Fort can move by itself (having the appropriate MVT TYPE and MVT parameters), or have Organic transport.
Fort can replenish AMMO and Fuel on the appropriate Terrain type.

Further interesting.

Fort does not get an EXP increase from attacks.
Fort can get replacements in two ways. The first way is to switch to Organic transport mode. The second way is to get replacements only in the city.

Yeah, it's a bit of a weird class, especially if you use it on the move. But you can use it.
In fact, you can make a Large caliber Heavy ARTY with Class FORT.

ARTY EXP does not increase as fast as units with Attack Range = 0. The most I've seen is two (or is it three?) stars for my ARTY in the Washington scenario. But not four or five stars.

Let's say you can approach the EXP problem from a different angle. Where will the combat experience come from for an ARTY firing at the enemy from a range of 15-20km? As they bring shells, so they will bring shells. As they used the calculation tables for shooting, so they will use them.
Their combat is carrying, counting, and cleaning and lubing the guns. I don't see a big problem with EXP not growing.

The problem with getting Replacements is the player will get used to that order. What AI will do with replacements? We need to test it.

The fact that the INF will not have a bonus (+2 I think) when attacking heavy, slow-moving and poorly defended ARTY can be compensated for by reducing GD and also INI.

There is one very benefit here, and it's a big one. This ARTY will shoot first when it is controlled by AI.
The second benefit is that large caliber artillery will no longer do things like Friendly fire. In reality, it never did for physics and geometry reasons.

In UI purchase.htm you can add FORT class as available for purchase. Rename it to Heavy ARTY. And change the text in the EXE file, so that in the game when clicking on a unit (ALT+ENTER) it will be called not "Fortification", but "Fort and ARTY".
Bytes 00080324 - 0008033C,
"00 46 00 6F 00 72 00 74 00 00 20 00 61 00 6E 00 64 00 20 00 41 00 52 00 54 00 59 00 00 00 00".
You can also write in text in the ASCII or Unicode field, if the editor allows.

[Radoye]

Actually, rather than heavy ATY, and considering certain oddities spotted with "anchor" transport units, i think this would be great to model armored trains...

[Lettos]

Actually, rather than heavy ATY, and considering certain oddities spotted with "anchor" transport units, i think this would be great to model armored trains...

You're right. Armored trains are also heavy artillery in their essence. We can put armored trains in this class 7 as well.
One does not interfere with the other.

[HexCode]

Intended Audience: # Lettos # & # Radoye #

PGF-SSI features Dual Mode Composite Units (DMCUs), of course. The AI Module does "flip" some of them during play. Examples:

A) Mounting / Dismounting ground units possessing Organic Transports.
B) Landing ground units embarked on Naval Transports.

On the other hand, the Module never air lands / drops ground units embarked on Air Transports.

PGF-CDP looks at the AI Module's DMCU "flipping" capabilities very carefully. Empirically seeking expected behavioral certainty is a prerequisite for deciding whether to put such Unit Types under the AI Module's control (or not).

Important: The AI Module's demonstrable inability to "flip" some DMCU Type does not necessarily imply that the Unit Type should not be put under the AI Module's control. The content designer may still field such units provided he has no... illusions as to what "exactly" he is handing over to the AI Module...

[Lettos]

Single unit Fort-train, Class 7, assigned MVT TYPE " - 4 " (minus 4)
Icon Block house in test:

[Lettos]

While experimenting with the AI's defensive behaviour, oddities in the AI's behaviour during its counter-attacks became apparent.

We know that if AI Offensive, its tactics and strategy are very simple. AI sends everything that can move forward.
In defense, AI often counter-attacks player units that are in its Spotting zone.

And sometimes AI suddenly makes strange rushes forward into an unseen space, where it bumps into a player's unit with Rugged defense, and suffers huge losses. What logic compels the AI to counterattack in this particular way? The last thing I'd want to think is that the AI suddenly "thought". There is no thought process inherent in AI.

I've run the tests.

Initial situation


AI Tank parameters:
-------------------------SA---HA---GD--INI--Range---Spotting----MVT---Cost
1) Tetrarch I------------3---9-----7----7-------0---------2------------7----144
2) Matilda II------------3---9-----14---7-------0---------2------------3-----96
3) Valentine Mk VIII--5---11----13---9-------0---------2------------4----252
4) Sherman V----------12---11---12---9-------0---------2------------5----288
5) Stuart V -------------5---10----9----8-------0---------2------------6----180
6) Centaur III----------12---11---10---8-------0---------2------------7----240

6 tanks can go on a counter-attack. I expect that from the AI. And, which tanks do you think the AI sent to attack?

Only four tanks. Two AI tanks moved to the rear.



It took tests and effort to figure out why this was happening.

As a result, there is a clear conclusion:
AI will move into the hidden zone to the Player's VH first unit which have MVT equal or more than distance from its location hex to hex adjaсent to attacked VH.
In this test, Sherman MVT=5 moved first towards the player's VH.



If AI will see hidden units, or have a unit with Spotting=10 in a given test, or even see the entire map, nothing will change.

The logic behind the decision is simple: if a priority target can be reached in one turn, the AI moves the unit to the target. It looks like VH is the highest priority target, and AI does not consider possible losses when moving to it.
It's logic within just one turn! There is no logical algorithm that Matilda (MVT=3) and Valentine (MVT=4) will attack and reach the goal in two turns. Therefore, the AI has moved them to the rear.
You can see that this logic is radically different from the logic of AI behaviour in Offensive mode.

Let's say a scenario designer wants to create a powerful AI counter-attack. Prepare frozen units with a calculated negative Fuel value so that they activate on the appropriate turn. Give the AI a lot of prestige on that turn, so that the AI buys something in addition to the frozen units.

How will AI move and prepare units for a counter-attack? On the "+ first" turn after unfreezing and buying AI will move units to its VH. On the "+ second" turn AI will start sorting units into slow and fast units. Maybe some fast units will attack on the "+ second" turn. On the "+ third" turn some more units will attack. In total, AI will send about half of all possible units to attack.
A ring of six units and a unit in the VH city AI will not move at all.
And AI will not move units from a purchase location (non-VH) or freeze location directly to the enemy.
Only through its VH first.



Does AI behaviour change if you set "At least N VH" in the AI victory conditions? NO.
Will AI behaviour change if you set "Mandatory VH" in the AI victory condition? NO.

It was while researching the Mandatory VH theme that I encountered some strange AI behaviour.
It looks like the victory conditions for AI are not reflected in any way in the software algorithm of its logic.

AI behaviour in counter attacks with Defense mode "At least N VH" (and Mandatory VH) is no different from Defense mode "Opposite side fails to win" and AI=Advanced. AI=Basic&Intermediate result in extremely passive AI behaviour, and cannot result in any counter attacks.

For AI, only one criterion is clear: MVT.

A few practical tricks to create an effective counter-attack in a scenario:

- there should be a distance = 2 hexes + MVT of the units used in the counter-attack from the AI VH to the VH that the player has captured and against which the counter-attack should take place.
It is pointless to use units with less MVT for counter-attack. They will just cluster around the VH AI.

- There should be a non-VH objective (city or port) near the VH controlled by the AI, where the AI will make its purchases.
It's pointless to expect the AI to free up hexes around its VH for purchases. He's not trained to do that at all.

- Prestige for AI should be given a large amount on the appropriate turn.
It is pointless to give AI prestige in small amounts each turn and expect an effective counter-attack. AI is not trained to accumulate prestige.

It should work!

[Lettos]

Continuing with the theme [AI] AI counter-attacks viewtopic.php?f=95&t=597#p17959

It has already become clear that we can manipulate AI. Experiments and practical games show that AI has the VH objective as the highest priority of all objectives. In this case, neither the cost of the VH nor its place on the map is at all significant for AI as a criterion for decision making.

All of this has already been noted by Hexcode a long time ago:
[VP] AI Module Behaviour > OBJECTIVES: OVERVIEW
viewtopic.php?f=100&t=532#p8877

By placing VH objectives on the map at an appropriate distance from each other, it is possible to create even an almost continuous systematic AI attack along the VH chain in Defense mode, and it will not be a completely mindless rush as in Offensive mode.

But how to implement this in a scenario? Of course, if it's in a densely populated area, you can somehow select appropriate cities and assign them VH status. But what about in the desert? And, besides, increasing the number of VHs will lead to breaking the PP balance of the scenario (if there is such a balance there, of course). To realise the idea, we need to understand what tools to use and how.

1) FPGE scenario editor has a limit of 20 VH in a scenario. PGF does not seem to have a limit. At least 26 VH does not cause PGF crash. In FPGE, if there are more than 20 VH in the scenario, clicking on VICT causes the program crash. If the VICT button is not touched and the scenario is re-saved, FPGE will reduce the number of VHs to 20 on its own.
Therefore, editing more than 20 VHs should be done only in a text editor. Nothing complicated - just write down the cities we want to make VHs and add them to the list of VHs.

2) And edit their cost. Let's say, make it equal to "1".This will not affect the game balance of PP.
And here is the main point - it is required to physically (visually) distinguish such technical VHs from traditional real ones.

Assume that another City tile can be created.



It should be clear to the player that after capturing this VH, he will not get 80PP as usual.
Let's tell him that at the start of the campaign!



Now you can give VH status to three palm trees and two camels near a well in the desert.

New tile Set, download link: https://www.mediafire.com/file/tmg0e0v5 ... W.zip/file
New set is same as used now in WAW and compatible with vanilla PGF. Three tiles added: Sailing ship (specialized to have airfields in Ocean), single island to have river terrain in Ocean, and new "technical VH" city. Row 18th.

[Lettos]

Lettos wrote: Fort does not get an EXP increase from attacks.

Wrong! It's a mistake! Forts get EXP whether they have Organic transport or not!
How could I have made such a mistake...?
I guess I should get outdoors more often and not think about a dozen topics at once

Actually, rather than heavy ATY, and considering certain oddities spotted with "anchor" transport units, i think this would be great to model armored trains...

There is another way to do the same thing, which is to move Heavy ARTY to the FORT class.
You can make a unit of class "13" Capital ship, and all other parameters as ARTY. If this unit has Organic transport, it loses the specific feature of the Capital ship class to strike back.

Class 13, unlike FORT, if it has Organic transport, can receive Elite and non-elite replacements on other terrain types (e.g. Clear) if not loaded on transport. But only not up to STR=10, but at STR = "+1".
Strange features of classes that are not intended by the EXE creator to use Organic transports...

One more nuance about moving Heavy ARTY to the FORT or Capital Ship class. These ARTYs must be very limited in their own movement (without loading on transports), because they can fire both at the beginning and at the end of the movement. However, there is no particular contradiction here if this unit can only move 1 hex on roads (as in the new MVT table). The conversion time from marching to combat position for the 17 K18 gun took 20 minutes in reality...

When making trains, you have to remember that under AI control they will only go in the direction of the player's VH, or towards their VH. These trains will be governed by the "One way ticket" rule and will not roll around the map in different directions.

[Lettos]

I "anchored" some of the AI Module's Fighter Class units over its Airfield hexes which either are objective hexes themselves or are adjacent to objective hexes. To boot, armed with the contents of

Prepositioned Units: "Unorthodox" Entrenchments
viewtopic.php?f=100&t=540#p12142

I gave those units Entrenchment Levels of FIVE (5). Oh, before I forget, I also placed "respectable" Air Defense units directly below the units.... My nephew has been telling me that the emerging challenge is on a par with his attempts to... bed extremely "hard-to-get" females !

Fly in the Ointment: Beware, "anchoring" fails in all instances where the Air Super-Class unit is adjacent to Body of Water, Port or Embarkation City hexes.

I like the idea of covering AD units from air attack! I don't understand why AD units should be suppressed by LB air strikes. Their personnel are specially trained to fight against such air attacks!

There is a question about the anchored Fighter - where does it go if the player captures an airfield hex? Even if the MVT parameter is set very high compared to the FUEL of such an anchored fighter, it will still interfere with one next turn until it disappears from the map. But the next turn can be very critical in some scenarios. Any ideas?

[Lettos]

Sometimes, despite the player's heroic fighters, AI Tac Bombers still break through to their targets.
What will they bomb, what are their priorities?

A test site has been made. 10 units of each Ground class (except ships), same AA, AD, INI parameters, same cost. AI bombs have parameters that allow them to attack all units except Anti aircraft and Air Defense without losses.
INF, Tank, Recon, ATG, ATY, AA, AD, Fort, INF + Land transport (cost =1) (landed on transports before start test)

AI has 90 bombers on the map to have enough to attack all player units.

There's a lot of nuance to this issue. You can spend more than one exciting evening experimenting and changing parameters.

So far, based on the results of the first experiments can be said for sure some observations:

- in the first priority of targets for AI is ARTY. Even if the unit's cost is only "1" and not 60 like the other units. Even if ATY's AD parameter is higher than other units of other classes.
- Tank, Recon, ATG, Fort, Land Transport classes AI (all of them have AA=0 in tests) attacks by some complicated selection algorithm. It simply alternates attacks for each class, choosing only a few targets out of 10. At the same time, each test showed some preference of one class over another.
- It seems that infantry without transports has a lower priority
- AI doesn't like to attack targets with increased ENT. If the AI sees an equivalent target with a lower ENT, the AI will choose it.
- If a unit has a higher AA value, the AI attacks it last.
- Targets that are protected by the player's AD units are last on the AI's priority list.

Hey, wait! I saw AI attacking my AA and AD units! And there were infantry right next to them!

Yeah. But that's what you think infantry was there. To the AI it wasn't. Because at the moment when AI moved its bomber, it didn't see any other targets.

AI is single-tasking. It acts according to the principle - "if I move, I attack according to the algorithm of target selection. If I can't attack, I still move in the direction where the player's units are already there" (that's why AI moves its Air Units to hexes next to the player's AA and AD). Just because the AI is slow to attack units on hexes next to the player's AD does not mean the AI can't leave its bomber there.

The magical word Spotting means a lot to the AI mind! Only the AI forgot to learn how to create a Spotting zone for itself.

As for the priorities in the AI-spotted zone, we need to keep experimenting.
I can't imagine at this point what difference changing the AA, AD, Cost and ENT of a player's units will make to the AI's goal priorities. It is possible that the prioritization algorithm is so complex (several variables) that incomplete experiments could lead to erroneous conclusions. Or the conclusions will have some degree of error.

But I think even incomplete results would give more clarity to the question : what do we do to AI behaviour when we change unit parameters in eqp?

[Lettos]

Based on the results of further experiments, the list of AI bombing priorities was refined.
The basic condition of the test is still the same: same AD=6, AA=0, Cost=60.
In all tests INI and SA and HA parameters of the target do not affect the AI selection. At least I haven't noticed any changes.
There's no difference, have target its Target Type Soft or Hard. At least changing the target type of the Tank, Fort class didn't change anything in the tests.
The only thing that matters is Class and the presence/absence of Land transport.

1. ATY. Always attacked first. Except when the AD parameter is set so high that bombers ignore this target. In practice, this means that only the rare armoured self-propelled ARTY will not be attacked.

2. Fort. Often second in order. In some cases shuffled with 3-7.
3. INF + Land transport. Often third in order. In some cases shuffled with 4-7, in rare cases with Forts.
4-7. ATG+Land transport, ATG, Recon, Tank. The tank is more often in seventh seat, Recon - on sixth. But not always. Often happens shuffled attacks to all these Classes 4-7.

8. INF without Land transport. (Almost always)

9. AD (always)
10. AA (always)

I started to change the Cost parameters.
Positions 4-7 are shuffled somehow when Cost changes by at least the value "90", i.e. +50%.
A change to "120" (+100%) results in a very significant shuffling.
A change to "180" (+200%) or more moves any Class from positions 2-8 "up" on the list to position 2, after ATY!
Well, let them switch positions. This is good and creates a randomness effect with a large assortment of units.

Changing the unit cost alone won't solve anything in problem. Yes, AI selects more expensive targets within the same class. Yes, within the narrow confines of positions 2-3-4-7, shuffling can be achieved. But so far it looks from the tests that the cost parameter is not the main determining factor.

When creating new units, as well as moving them from a traditional class to another class, it is important to consider the AI's priorities when selecting targets. For example, moving some ARTY units to the FORT class will not change the overall picture. And moving Recon to the TANK class won't spoil anything either.

For example, when we create an Armored INF unit, we take away transports. Infantry without transports will be the last on the list to get hit from above on their helmets.
To make it not a super-unit with immunity against bombardment, but similar to regular infantry, we have to raise the cost of a created unit by at least 100% compared to a regular INF unit.

But besides COST, there are also AA and AD parameters! The cumulative effect of changing all three parameters at the same time is possible!

Tests show that the AA parameter of the attacked unit is taken into account by our Artificial Intelligence in a completely non-intelligent way. Simply put, contrary to initial assumptions, it is not taken into account at all.
That probably explains the AI aircraft attacks of the Player's AD and AA units.

Fortunately, the AD parameter did not fail!

It affects the priority of the target approximately as follows (according to subjective observations): reducing the AD parameter by "1" is equivalent to increasing the cost of a unit by 100%.
And there's only simple addition. I didn't notice any multiplication.

For example, reducing AD from 6 to 4 - i.e. by "2" is equivalent to increasing unit cost by 200% or equivalent to changing AD 6 -> 5 and simultaneously COST +100%.
This change of both parameters at once is also guaranteed to move INF without land transport to position 2 in the target priority list.

[Lettos]

A continuation of the theme about bomber attacks.

Does the COST value of the player's units affect the AI's choice in absolute numbers? Does it make a difference if all units cost 12 PP or 480?
No difference. We are dealing with the theory of relativity

Then, where is the median average value of the cost? AI takes it from the price of the cheapest unit, the most expensive, or? Yes, AI calculates the cost from the sum of all units.

It has not yet been determined whether the value of all the player's units on the map matters in this median calculation or whether the value of units of one class matters more than the value of units of other classes.

Observations: sometimes the AI will be the first to attack some other non-ARTY unit. That does not change the further course of events. But there is some small random. And that fact is great for a scenario designer!

The AI considers the anti-aircraft class to be the most unattractive and undesirable to attack. Reducing the cost, adding Land Transport (in all experiments Land Transport cost = 0) and reducing AD for anti-aircrafts does not change anything in this shark's tastes. ARTY is tasty, Anti-aircrafts are untasty. This is an axiom for AI.

It has already become clear that by changing the prices of the Player's units (not only and not so much CORE, but specifically AUX) in a scenario, you can significantly change the course of the scenario. Make crap units cheaper - and the Player's tanks get a bump on their turrets! Put some expensive Air Defence guns somewhere in the rear - and AI will change preferences when attacking. Isn't that why SSI programmers sometimes put expensive 8.8 cm anti-aircraft guns, which are useless in the scenario, on airfields in the rear of the player?

Soon these experiments will be over. Perhaps. Unless you want to answer the question - does it matter only for TacBombers, or for land units too?

[Lettos]

1) It looks like units located on adjacent VH hexes become more palatable to the AI.
And out of several identical units as a target for attack, the AI will prefer to attack first the unit that is not only on the adjacent hex, but also close to the VH within 1 hex.

2) In the tests, it was observed that:
Fighters with SA=1 do not attack Land units with AD=6 and ENT=2 or more. Attacks on an AD=6 ENT=1 unit are rare. But they do.
Bombers with SA=4 do not attack Land units with AD=9 and ENT=3 or more. Etc.
Although sometimes such attacks can be effective, causing 1-2 STR losses.

Seems the general rule for an attack from air Soft or Hard target to take place is close to this:

Target (AD+ENT) - Attacker SA(or HA) = 6 or less

[Lettos]

Thinking about the Air Carrier and at least a partial solution to the problem
viewtopic.php?f=95&p=18226&sid=1d90c04e ... a6d#p18226

ended with the conclusion:

Unfortunately, a dived submarine in PGF won't be able to sink such a pseudo aircraft carrier. Nor will it be able to sink an AA Cruiser.... That's a disadvantage.

As it turns out, if you create the right Dual Purpose unit, the situation where a pseudo Air Carrier cannot be attacked by a submarine will be rare enough.

Dual Purpose unit:
Anti Aircraft unit with parameters:
SA, HA = 0.
AA, AD, GD, NA = designer's choice
Very important! MVT=0.

Assign it Organic Transport, one of the three ship classes. The parameters SA, HA, AA, AA, AD, AD, GD, NA are at least +1 higher than the AA unit. MVT=6 (or about the same as the other ships in the squadron).
This Organic Transport looks better to the AI than the AA unit that uses this transport. I don't know what exactly is the determining criterion for AI when choosing a mode. Maybe it's only single MVT?
But it can be seen that the AI in the tests only uses AA mode to attack the player's air units, and then almost always switches to Organic Transport mode.
Sure, we need to test it in a real scenario.

If Organic Transport is going to be Class 12, then Organic Transport starts hunting for submarines. This is not necessary.
If Organic Transport is going to be Class 13, AI Organic Transport may start selecting as a higher priority target the player's ships rather than aircraft. I don't think this is necessary. Except maybe assigning the unit a lower Naval Attack parameter than Air Attack.

If Organic Transport is going to have a Class 14, then we know the AI doesn't attack with an aircraft carrier. You can freely assign any Naval and Air Attack parameters you want. Interestingly, such an Organic Transport Class 14 does not lose the aircraft carrier function and allows Resupply for aircraft.

Conventionally, we can consider that when such a pseudo-aircraft carrier (called Chimera? ) is in AA mode, a submarine is afraid to attack it - it will be seen from aircraft at torpedo attack depth.

========

More about aircraft carriers, only already on land. Such a land-based aircraft carrier with a small airfield icon (like terrain, but one runway instead of two is enough) is a small airfield for one aircraft unit. Airfield can be transported on transporters - it is conventionally a unit of airfield builders who find a level ground, quickly set up a runway and bring the necessary AMMO and FUEL.
Such an airfield can even be purchased as an AUX unit..... in desert scenarios such a unit may come in handy.

[Radoye]

Yeah i was thinking about improvised airfields... transport is an INF class unit (technically a construction battalion with stats similar to bridge engineers only without bridging ability, with their transports integrated like motorized / mechanized units i recently added to WaW) and transported is an Aircraft Carrier class unit with 0 towed movement, same attack / defense stats like transport (the construction battalion on guards duty)...

[Lettos]

Yeah i was thinking about improvised airfields... transport is an INF class unit (technically a construction battalion with stats similar to bridge engineers only without bridging ability, with their transports integrated like motorized / mechanized units i recently added to WaW) and transported is an Aircraft Carrier class unit with 0 towed movement, same attack / defense stats like transport (the construction battalion on guards duty)...

Fine!
Yes, mobile airfield should have MVT =0 and MVT type... in the traditional universe it's Towed, in my universe it's more like MVT Type = Leg (otherwise you can't build an airfield on certain types of Terrain where Towed has no input). These are modder tricks with movement types, which I will describe in great detail soon.

Transportation of mobile Airfields can be performed by usual Land transport. Something powerful, like SdkfZ.9. Parameters SA/HA, GD - yes, almost zero, like a transport. Any enemy ground unit captures(i.e. destroys) this airfield. But AD parameter of mobile airfield can be raised. If enemy bombers arrive and hit the airfield, the pits can be backfilled by airfield builders and the airfield will continue to exist.

Shall we include it in the modder's manual?

[Lettos]

For those readers who are not familiar with the topic, I suggest reading excerpts from Prima's Official Strategy Guide first, as well as some very interesting observations and sometimes ironic comments from the esteemed HexCode!

[VP] AI Module Behavior
viewtopic.php?f=100&t=532#p8874

While developing and researching the topic of naval combat, I had a question about AI logic and how we perceive it in the game.
"Oh, he didn't take a chance!", "Oh, he took the unit to the wrong place!", "Instead of hitting, he ran away!" and so on.
We know these things, don't we? (That's not the main issue yet!)

AI's actions are guided by the logic of benefit. Benefit from attacking a player's unit, benefit from defending objectives, and so on.

And here I am currently researching the war at sea (in relation to the PGF model, of course!). And the main question arises - if the parameters of units even in the vanilla version of Panzer General are set so that they have no benefit to go on the attack, what do we expect from AI?

There is a cognitive dissonance here, when we see a powerful unit, for example AI-Led Battleship, and it does not attack anyone, except the weakest units such as Destroyers.
"It's in its algorithm!".
Yes and no at the same time. The rules of decision making are written in the algorithm, and in those rules there is already a benefit from the decisions as the AI creator understood it.
But if we (or anyone else before us) deployed in scenario (and eqp file too!) beautiful units with very strange parameters in terms of what they look like in an action-benefit AI model, the AI will just - no, not go crazy - it will just logically refuse to consider a deal with these units as beneficial.

Yes, I have actual examples as well. They will be discussed in detail in the "Admiral's tools" topics in
thread [ADV] Advanced Play System - Questions & Commentary
Starting from 3rd page : viewtopic.php?f=95&t=470&start=100

[Lettos]

Observations

# Lettos # has done extensive empirical testing to determine how PGF's AI Module goes about performing various "acts" with the units it leads. Focusing on movement and attacks, he concluded that:

1) The AI Module "acts" on behalf of its units on a per Unit Class priority basis. Once it finishes "acting" within the context of a particular Unit Class, it moves on to "acting" on behalf of the units of the Unit Class which is assigned the next lower priority. At the end of each unit's movement phase the AI may or not initiate an attack. Does it ever attack prior to actually moving, though ?

2) Transported units "act" on the basis of their particular Unit Class.

3) The following priorities apply to the THREE (3) Unit Super-Classes:

Air
Naval
Ground

4) The following priorities apply within the FOUR (4) Air Unit Classes:

Tactical Bomber
Level Bomber
Fighter
Air Transport (self contained)

5) The following priorities apply within the FIVE (5) Naval Unit Classes:

Destroyer
Submarine
Capital Ship
Aircraft Carrier
Naval Transport (self contained)

6) The following priorities apply within the NINE (9) Ground Unit Classes:

Reconnaissance
Tank
Anti-Aircraft
Artillery
Air Defense
Infantry
Anti-Tank

Structure (i.e., Forts)
Organic Transport (self contained)

? ? ?

Wishes & Proposed Solutions

There's plenty of time for that.

So, two and a half years have passed, and it's timely to revisit the topic of prioritizing movement and attack.

Fort attacks and moves first of the ground units. Attack first, then movement. Just like ARTY.
Fort can move by itself (having the appropriate MVT TYPE and MVT parameters), or have Organic transport.
Fort can replenish AMMO and Fuel on the appropriate Terrain type.

Further interesting.

Fort does not get an EXP increase from attacks.
Fort can get replacements in two ways. The first way is to switch to Organic transport mode. The second way is to get replacements only in the city.

Yeah, it's a bit of a weird class, especially if you use it on the move. But you can use it.
In fact, you can make a Large caliber Heavy ARTY with Class FORT.

ARTY EXP does not increase as fast as units with Attack Range = 0. The most I've seen is two (or is it three?) stars for my ARTY in the Washington scenario. But not four or five stars.

Let's say you can approach the EXP problem from a different angle. Where will the combat experience come from for an ARTY firing at the enemy from a range of 15-20km? As they bring shells, so they will bring shells. As they used the calculation tables for shooting, so they will use them.
Their combat is carrying, counting, and cleaning and lubing the guns. I don't see a big problem with EXP not growing.

The problem with getting Replacements is the player will get used to that order. What AI will do with replacements? We need to test it.

The fact that the INF will not have a bonus (+2 I think) when attacking heavy, slow-moving and poorly defended ARTY can be compensated for by reducing GD and also INI.

There is one very benefit here, and it's a big one. This ARTY will shoot first when it is controlled by AI.
The second benefit is that large caliber artillery will no longer do things like Friendly fire. In reality, it never did for physics and geometry reasons.

In UI purchase.htm you can add FORT class as available for purchase. Rename it to Heavy ARTY. And change the text in the EXE file, so that in the game when clicking on a unit (ALT+ENTER) it will be called not "Fortification", but "Fort and ARTY".
Bytes 00080324 - 0008033C,
"00 46 00 6F 00 72 00 74 00 00 20 00 61 00 6E 00 64 00 20 00 41 00 52 00 54 00 59 00 00 00 00".
You can also write in text in the ASCII or Unicode field, if the editor allows.

lettos wrote: Fort attacks and moves first of the ground units. Attack first, then movement. Just like ARTY.

Unfortunately, WRONG! Forts can attack before movement phase and also after it!
However, with the very limited ability of the Heavy ARTY unit to move without transportation, this is not a significant problem.
The pros of the solution still outweigh the cons.

[Lettos]

Based on the results of further experiments, the list of AI bombing priorities was refined.
The basic condition of the test is still the same: same AD=6, AA=0, Cost=60.
In all tests INI and SA and HA parameters of the target do not affect the AI selection. At least I haven't noticed any changes.
There's no difference, have target its Target Type Soft or Hard. At least changing the target type of the Tank, Fort class didn't change anything in the tests.
The only thing that matters is Class and the presence/absence of Land transport.

1. ATY. Always attacked first. Except when the AD parameter is set so high that bombers ignore this target. In practice, this means that only the rare armoured self-propelled ARTY will not be attacked.

2. Fort. Often second in order. In some cases shuffled with 3-7.
3. INF + Land transport. Often third in order. In some cases shuffled with 4-7, in rare cases with Forts.
4-7. ATG+Land transport, ATG, Recon, Tank. The tank is more often in seventh seat, Recon - on sixth. But not always. Often happens shuffled attacks to all these Classes 4-7.

8. INF without Land transport. (Almost always)

9. AD (always)
10. AA (always)

That's right. But there's one nuance. (Tests are somehow different from a real combat environment, no matter how hard the experimenter tries. Although... then it all comes out during the games!)
If in a real scenario the AI has different Air Units that can attack different visible units of the player, the AI makes a decision depending on its benefit. And a faster bomber may fly to attack first (because it has better SA/HA parameters than a slower bomber, and the benefit of the attack will be greater).

It should be clearly understood that the attack order of AI classes does not change. What changes is the priority within a class.
It is very likely and logical that this is the algorithm for determining the benefit of an attack within each AI class. At least during the attack of AI-led tanks on the player's units the same thing was observed as in the case of TacBombers.

[Lettos]

... otherwise, if the cube is taken away, strange things will happen!

Several AI-led identical units but recorded under different numbers in eqp. Assume they are destroyers of the same class. Which of them will move first if AI sees no benefit from the attack?
The answer is surprisingly pragmatic - the one with the lowest eqp number will move first.

How is that right... someone from a set of identical brothers should move first, right? But a question to the programmer - why didn't you buy a child a dice?
Each unit rolls a die, a movement priority is determined.... and intrigue is created. Well, well, well, who's dreaming of primitive things to make the game more interesting?

And if there are five identical AI units on the map, who will move first (we don't think about the benefit of the attack, as it brings its own priorities!)?
Have you probably already drawn your attention to the strange "names" of units in PGF? "1st Tiger", "2nd Tiger" etc...?
Yes, that's right! 1st Tiger will move first! 2nd Tiger will move second, 10th Tiger will move tenth.

And what determines the order in which these mystical numbers are assigned? The chaotic actions of the scenario designer at FPGE? Yes and no. Yes, there is some correlation between the actions in FPGE and the ordinal number of the combat unit... and there is, just without understanding the nature of the process.
But the best way to think about it is that the order in which numbers are assigned is determined by... by the place in pgscn where the unit record is. Move it up and down in the list, and you'll see that the mystical battalion/division/company/regiment number has changed ordinal number!

And then we look at this shirt and resent it for being so predictable. Of course it is!

By the way, one of the options for creating some initial unpredictability in a specially prepared scenario is to change the lines of unit records in pgscn...
Funny, simple, clever and silly all at the same time. But it'll work!

Again! Why didn't the programmer buy his brainchild a dice?

[Lettos]

Since the days of Panzer General, there are clear rules for organizing the process of naval combat:
Fog of war - On
Spotting = 2, maximum 3.
Fire range = 0 up to 5, even 6.

These rules should make for an interesting naval battle. And yes, they probably do, but... only in H2H mode.
AI doesn't know how to do reconnaissance. It lives by the principle of where it came, what it saw, I shoot at.
Accordingly, if AI didn't go anywhere, he didn't see anything.
Set his ships close to coastal line, and wait for them to be shot one by one out of the impenetrable fog of war.
Even if the AI-led battleship automatically strikes back while the player is attacking the AI, it's just a one-time action. The player can safely leave his ships in the fog of war, regroup here, three hexes away from the enemy battleship, and strike again a turn later.

Of course, the first thing that is possible to do is to give AI some specific units with a high spotting value.
The game is already getting more interesting!

And the second - here's a question to everyone who can report something on the topic: have there been any modding experiments with Fire Range and Spotting parameters in the sense of their equality?

For example, if Battleship has Fire range = 5, then it spotting set to 5 too?
Otherwise, it's illogical to have a lone Battleship standing in the middle of the ocean. It can see two hexes, but can shoot at the fog of war two and a half times farther. How's that? In fact, from the mast of a battleship 35 meters high (Iowa) you can see a tall enemy ship at a distance greater than the maximum range of the main caliber guns.

Can AI actions in naval combat be improved in this way?

[Radoye]

Distance to horizon from 35m height is ~20 km. Battleships main guns maximum range is ~40km.

(also, keep in mind there's no way to limit visibility to sea only, if spotting is increased capital ships would be able to see long distance inland, even over mountains )

[Lettos]

Distance to horizon from 35m height is ~20 km. Battleships main guns maximum range is ~40km.

(also, keep in mind there's no way to limit visibility to sea only, if spotting is increased capital ships would be able to see long distance inland, even over mountains )

The horizon line is zero meters high. But we are looking from a height of 35 meters at another tower 35 meters high. At least we see a large ship's hull and superstructures, totaling 15-20 meters in height. We will notice a thin black line on the horizon, which will turn into a mountain of steel as the ships get closer . Distance to the thin line (visible enemy deck) from the observer's position on the Iowa tower = 37 kilometers. You can already open fire and see the results of the hits. This explains why the battleships' guns fired at a range of 35-42 kilometers. There was no point in firing at longer ranges.

The top of the mast of another ship (another such observer, only on a Japanese battleship) can be seen as early as 42 kilometers. But this is in good clear weather and in the absence of strong waves.

Yes, on land, a ship with spotting=5 will see both the mountains and beyond them..... so now a destroyer and a submarine can see three hexes! And we're used to it.
If spotting=4-5 for cruisers and battleships will change naval combat for the better, then we will get used to 5 viewable hexes of mountains.

[Lettos]

If you're asking does PacGen have purely naval scenarios, the answer is yes, both in vanilla and in custom made (i believe there's one in Steve Strayer's Plan Z campaign just like you ask - however, it has been a while since i ran PacGen, it doesn't go along well with modern PCs). But, the PacGen unit eqp parameters don't copy to PG / PGF 1:1. There are more unit classes, target types, attack / defense stats, it is possible to specify base unit strength and many more. There are also day / night turns, each with their own rules, different terrain types including deep and shallow ocean, impassable reefs... So the mechanics of the game especially regarding naval combat are very different from PGF. What we have in PGF is a small subset of what can be done in PacGen, and while it is possible to fairly simply copy things from PGF to PacGen (you can fit 1/2 liter of water in a 1l bottle) the other way is significantly more complex...

Excellent mechanics that PGF can only dream of!

But the mechanics are only part of the game. There's a triangle of three component parts here: mechanics, AI actions, and the predictability/randomness of each scenario.
The mechanics generate sales - it's very good material for advertising. So many new features!!!
AI few people will notice at first, then AI will be scolded.
Indeed, what is the point of the most perfect combat mechanics, if the combat does not happen in any even remotely similar to reality in the purchased game!
But even if the mechanics and AI are great (the combination of the two is already fantastic!), the... how many times can you play the same thing? We need a mechanism for unpredictability in starting conditions, in weather, in equipment, in the human factor. A set of random events, which in life sometimes led to a completely unpredictable result of the battle. One shell that hit the right place (or vice versa, not necessary) - and there is no more powerful ship....

Sometimes game makers were tempted to make mechanics + random, and then blame all AI shortcomings on random. Sooner or later smart gamers understood everything, and such tricks did not lead to game fame...

It's a shame that Pacific General has stopped developing.

By the way, this

different terrain types including deep and shallow ocean, impassable reefs...

can be done in PGF even tomorrow.
Make an ocean out of two Terrain types - Ocean + some other type. For example, River. How to separate real rivers from the ocean, I've already written about it.
Make two types of movement, existing Naval (modified) + another one. We have a completely unused type Seep Movement. We can have a specific MVT Type for big ships, and for small ships.
We can have different conditions of movement of ships at different depths during a storm (I have already written my starting ideas about how to make sea storms from unnecessary Frozen conditions too).

And I would like to get the AI to change the direction vector of the squadron not only depending on the starting position of its ships (it works!), but also in the process of the game. I have some ideas about it, I need to test it!

[Radoye]

By the way, this

different terrain types including deep and shallow ocean, impassable reefs...

can be done in PGF even tomorrow.

I know but there's more to it, not just additional naval terrain types and movements. There are new choices and possibilities on land as well.

Where in PGF if we want to make naval combat more nuanced we must sacrifice something else in return.

PacGen showed great promise but was sadly rushed onto market in an unfinished state. I guess SSI wanted to focus more on PG2?

Unit icons stand out poorly against terrain graphics, AI is weak (when set to attack, behaves as PG1 AI set to defend, when set to defend does pretty much nothing)...

Shame!

[Lettos]

Code fragment from an EXE file:



Funny question - why does PGF need trigonometric functions?

The answer is that the AI uses them when determining the direction vector of the attack. The direction vector is calculated based on the player's VH on the map and the AI unit's position. And from practice we can see that there are many additional conditions that determine the attack priorities.
But tangents with cosines are also present. AI knows about "Euclid..."