Panzer Corps AI discussion

[Rudankort]

Hello All!

This was discussed before, but I thought I'd create an "official" thread for this and keep it sticky for a while. Now that there is a new round of AI changes planned for Africa, let us discuss what changes are needed in the AI department, and how to prioritize them. Especially important is to put together a full list of little things the AI does stupidly (like firing artillery after assault attacks or moving it in trucks next to the enemy), because these can be really annoying, and are likely possible to fix.

Some past topics on the subject (feel free to post more links):
viewtopic.php?f=121&t=30198
viewtopic.php?f=121&t=27353
viewtopic.php?f=121&t=26067

A separate question is, what new options for configuring the AI are needed in the Editor to enable better user-created single player scenarios.

[bebro]

Aside from things already mentioned as arty or mounted movement some ideas:

1. Maybe improve the AI's use of air units, I mean esp. that AI aircraft "freeze up" when they lost all airfields. It would be nice if they either continued fighting to the last drop of fuel, or (for scn design) to have the option to withdraw them off-map by trigger in such cases.

2. Still air - maybe fighters could be have an AI behaviour setting that makes a certain task priority for them (air superiority: fight other planes first; or - escort own planes, or - ground support). Could be expanded for bombers in various ways (attack ground/naval first, etc.)

3. Ground: better AI use of bridge engineers.

[El_Condoro]

Let's go!

AI
1. AI units set to Hold, Active should not leave entrenchment to attack likely targets.
2. Likewise, Hold, Active units should attack adjacent targets if there is a good chance of success. e.g. an artillery unit in transport is adjacent
3. AI ground units should not attack naval units
4. AI fighters should target player fighters as priorities and only soft ground targets [Edit]: bebro's 2. suggestion is probably better
5. AI recons are ironically the ones that seem to get the most surprise contacts! Recons should reconnoitre before they move.
6. AI bridging units need to know how to use their ability and not move away from a river crossing when set to do so as part of a coordinated attack
7. Engineers can lay mines, tank traps, blow bridges, create airfields (Actually, this one is probably more for player units than the AI)
8. [Edit] Make better use of terrain e.g. keep tanks out of close terrain, use infantry for close terrain attacks etc

Editor
1. Ability to set the targets for AI units set to Attacker
2. Ability to set waypoints using zones - e.g. All units with "2SS" in their name move to Zone 1 on Turn 3; then Zone 2 on Turn 6 etc. or All bombers with "IX TAC" in their name attack Zone 3 on Turn 2 etc etc
3. Ability to REMOVE AI (and non-core player) units from a scenario according to criteria. e.g Attached units, bombers etc.
4. Ability to set maximum size of unit strength independently of editing the equipment file. e.g. If the designer wants Tigers to be no more than 5 strength.
5. Elite and SE units available to both sides - not just Side 1/Axis
6. Starting XP for units in scenarios to work for campaign scenarios, too.

These are not AI...
a. Ability to set towns/cities as NOT supply points or conversely ability to set any hex as a supply point (all others are not by default)
b. River terrain type that does not freeze over e.g. the Rhine.
c. Ground conditions can be set for the whole scenario e.g. it is always the snow terrain at Wacht am Rhein even after 20 turns of clear (exaggeration)

There are a lot of Editor enhancement requests here, too.

[dks]

have not read all the threads but something to me that sticks out is the AI does (not) do is finish the kill. I have had many units live due to the AI chasing other targets when my unit was the easy kill if the AI pressed the issue.

[robman]

Relative to human vs. human play, the AI does not take advantage of opportunities to seize back captured but un(der)occupied objective hexes seized earlier in the game. AI units in bypassed non-objective hexes usually sit there the whole game, when they could try to infiltrate behind the player's "lines" and wreak havoc by seizing or threatening previously captured objectives. Granted, this may be by design. But it would certainly change gameplay if reserves had to be left toward the rear in anticipation of infiltration of this sort.

[Ballermann]

Ai-Paratroopers don´t jump?
viewtopic.php?f=121&t=30624

[deducter]

Basic techniques:
1. Improve Artillery use. A simple solution might be to force the AI to fire all artillery prior to ground combat.

In terms of advanced techniques, there are a lot more to think about. For instance, say there's a player infantry with entrenchment 3 in a city. The AI has 2 artillery and 2 infantry. Instead of bombarding twice with artillery then attacking with infantry, it may be better to bombard once with artillery, then attack with infantry, then bombard again and attack with the other infantry to keep the player's infantry in close terrain to take more losses.

2. The AI should check whether an air attack or bombardment is covered by a fighter or AA. It is currently possible to "ambush" AI planes by hiding a fighter or AA out of the AI's spotting range. The AI will move a plane up and attack. A player can move a plane to attack, pause before attacking to spot the surrounding hexes. If the player sees AA or a plane, he might not attack to save his plane losses.

3. The AI should cycle through units again after all movement is finished, to see if any attacks are still worth making. For instance, say there's a Panzer IVH with 3 hexes open. The AI moves up a T-34, does not attack because it doesn't like the odds. Moves up another T-34, again doesn't attack. Moves up a IS-2, likes the odds, and attacks, inflicting heavy losses on the Panzer IV. Now the other 2 T-34s can attack, probably with much better odds. Currently, once the AI moves a unit and doesn't attack with it, the unit is finished.

4. AI's recon units should use their phased move ability. Currently the AI treats recon units like tanks. Make the AI know how to use recon units to ignore ZOC and hit a player's unit in the rear lines. Or better yet, allow the AI to actually scout by moving its recon first and retreating after spotting the player's units. Useful for helping the AI avoid ambushes.

5. The AI should know how to switch units. Soviet mobile artillery often have a switch mode, and the soviet 85 mm AA gun now has an AT mode too.

6. Make the AI take into account ammo supply of the player's units. For instance, the player sends a Tiger II with 2 ammo left out in front of their army. The AI has 2 infantry and 2 tanks available for attack. The AI should attack twice with the infantry, possibly losing both, but then the Tiger II is out of ammo, and the AI can attack twice with its tanks to inflict some losses on the Tiger II. This can even be done if the AI has 4 tanks. Sacrifice 2 of them to force the Tiger II out of ammo, then attacking with the other 2 to inflict some losses.

Advanced Techniques

1. Let the AI spread out its units to interdict the player. The AI could benefit immensely by understanding how ZOC works. Also the Ai should avoid clumping up all of its tanks so the player can force easy surrenders.

2. Take into account surrender/retreat with making its moves. This is a complicated topic, but even programming something as simple as the AI spotting the player clumping up a huge mass of units and then forcing some surrenders at the front would be nice. Other techniques including forcing retreats onto rivers to take advantage of the combat bonus, forcing retreats of units into close terrain for infantry to destroy.

3. Take advantage of mass attack, especially against infantry in close terrain and also in bad weather.

4. Understand that sometimes it is best to position units in the rear or on the flanks for an encirclement attack. Don't attack a Tiger II head on, instead surround it, or ignore it and go after the weaker units. Don't drive up T-34s next to the unit unless it is out of ammo or nearly so. This is particularly relevant for the Soviet because their equipment tends to be worse than the Germans.

5. Shoot through fighter cover to target a vulnerable plane. Currently the AI might do this on Manstein difficulty, by accident, or if the player's escorting fighter is too weak relative to the AI's fighter. Make this a "conscious" decision, for instance, send a I-16 to attack a Ju 87 protected by a Bf 109, then send in two Yak-7s to finish the job.

6. Purchase "battlegroups" of infantry/artillery/tanks, along with fighters/bombers. This could also reduce the IS-2 spam common in many late war scenarios. These forces can be used as a mobile reserve for defense or to reinforce an attack. Not particularly relevant for the current DLCs as most of the units are pre-placed, but would be nice for future DLCs.

Other notes
It might be interesting to have an AI mode that uses advanced MP techniques. It would probably make the game as unplayable as the ultimate difficulty, and I guess programming such an AI is impractical and too time-consuming. But yeah, even having the AI be capable of performing some advanced techniques could be even to warrant its own separate difficulty level.

[parusski]

have not read all the threads but something to me that sticks out is the AI does (not) do is finish the kill. I have had many units live due to the AI chasing other targets when my unit was the easy kill if the AI pressed the issue.

Ok, I bounced over from Matrix. I agree the AI needs to finish the job.

[El_Condoro]

^ deducter's tips for the AI are very good for players, too! Just an observation...

[Kerensky]

TLDR on topic:
If I could change the AI I would... (I'll try not to be redundant and repeat what has already been mentioned, such as AI using switch mode)

1. Not allow units in transport (or Artillery) to move adjacent to enemy ground units UNLESS they cannot see the unit. (AI trucks SHOULD drive up and even get ambushed by player units hidden in Fog of War or hidden by bad weather)
2. AI should pay a little more attention to its supply status. A 1 or 0 ammo unit should probably not advance towards the enemy, it should want to resupply or even move away from the enemy if it cannot resupply. Currently, a 0 ammo AI unit on attacker will sit face to face with the player, becoming something of a roadblock that can be ignored.

3.
Also:
Currently, the order of operations goes something like:
Activate, move, fire all recon units.
Activate, move, fire all tank units.
Activate, move, fire all infantry units.
Activate, move, fire all artillery units.
Activate, move, fire all anti-tank units.
And so on. There is some variation to these behaviors (sometimes artillery will fire on a target before infantry attacks it) but these tend to be the exception not the rule. For example if you put 3 AI tanks and 3 AI infantry in range of a player controlled victory hex city, the 3 tanks will all activate first and move adjacent to the city (they may or may not fire depending on how good of a tank they are) and THEN the 3 infantry will get their turn to act, which leaves them blocked and unable to reach the city hex.

Sometimes this behavior is great and we want tanks to lead the charge as advance units. Sometimes though, we don't, such as the city example above.
Solution: When we configure units in the editor, add an OPTIONAL parameter called 'priority'. Adjusting priority, which is a simple number from... 1-10, moves up or down the priority in which the AI will active this particular unit.
If priority field is ignored, AI will active in the current order. Any number 1-4 will activate in advance of the typical AI order. A priority 1 unit will be the first thing the AI picks to move/attack with. Any number 6-10 means these units move even later than the normal AI decision making progress. A priority 10 unit will move absolutely dead last.


4. Add kamikaze setting to AI parameters. Similar in every way to attacker, but kamikaze will even attack when the odds are absolutely NOT in their favor. 10-0 attack? Do it anyways. A few of these units would be nice to keep good pressure on a player, and also fight against AI traffic jams.








Off topic thoughts:
After so many, many scenarios, I've grown rather fond of the AI for the most part. I'm not particularly sure that a more threatening and cutthroat AI is in the best interest of the game either, especially when it already exists.
For example, the AI and dealing with escorted tactical bombers. As mentioned, this pretty much never happens. If you escort your tacticals, the AI fighters ignore them because the attack odds are so dismal. Sometimes the AI will attack anyways though, but this is mostly because of Manstein difficulty setting. I don't see a problem.
Players on the average difficulty settings are rewarded for good escorting of their tactical bombers. Players on the highest difficulty setting run the risk of losing them even though they put together escorts.
If the AI intentionally soaked off fighters to absorb interception and then pounced and crushed player tactical bombers, I'm just imaging how players would feel losing their veteran units to the textbook example of 'zerging' and how they would ask 'why bother having escorts.

Also, if you create an AI that is immune to being ambushed and intercepted, all it creates is the illusion of a cheating AI. A player who understands sight rules and carefully places a unit in sight range of the enemy but keeps the escort out of sight should be rewarded with an intercept or ambush. I'm pretty sure if the AI never got ambushed and never ate a bad interception, players unfamiliar with the game would think it's cheating with some sort of Fog of War cheat. The fact the current AI does not cheat is exactly why it ends up in these bad situations. So you see the obvious problem with changing this.

Bottom line, a lot of suggestions seem to want the AI to play more like a human player. I don't think this is good at all. AI play and multiplayer are two entirely different modes of game play. When you play against the AI, you can expect to have a K:D (Kill:Death) ratio of 10:1, or maybe even better especially on the lower difficulty settings. You will never see that in a multiplayer game unless one side is completely crushing the other (the worst and most unfun kind of multiplayer experience).

Multiplayer is the place to go if you want an adversary who is witty, crafty, deadly, and ruthless. Singleplayer's priority is campaign play, and campaign play is fundamentally based on two concepts.
1. You should win, and win a LOT (or your campaign won't continue for lack of prestige/because the game is over).
2. Your units need to survive for extended periods of time.

Both of those tenants are in total conflict with the multiplayer setting of 1 piece battles, but that's fine. Why make singleplayer more like multiplayer? Just play multiplayer. It's like there are two games in one.

[El_Condoro]

I don't want to detract from the main focus of this thread but I would like to offer arguments related to what Kerensky has said. I am happy to move this to a new thread.

Bottom line, a lot of suggestions seem to want the AI to play more like a human player. I don't think this is good at all. AI play and multiplayer are two entirely different modes of game play. When you play against the AI, you can expect to have a K:D (Kill:Death) ratio of 1:10, or maybe even better especially on the lower difficulty settings. You will never see that in a multiplayer game unless one side is completely crushing the other (the worst and most unfun kind of multiplayer experience).

The current DLCs fit reasonably well with the concept of an AI that makes mistakes, overcomes the player with numbers and needs huge numbers to have any chance of success (overcome its flaws). It would sit reasonably well with a West Front DLC where the player was the Germans and the AI had huge materiel supremacy. But its tactical ability would not.

Where it definitely doesn't work well is when the player should be up against a numerically inferior foe who fights hard. e.g. A scenario where the numerically superior Allied human player is trying to break out of the Normandy bocage. The historical Germans held them up for weeks with better tactics, better (but less) equipment and smart use of terrain. The current AI would put its SS Panthers IN the bocage to be destroyed!

So, in short, the weak AI can be justified when the player is the Germans (of course, it is called Panzer Corps!) but is harder to justify when the player is Allies or Russian. Since an "Allied pack" is forthcoming in Q4 I think an improved AI is essential to make the AI Germans competitive without needing to have many more units than they really had.

Multiplayer is the place to go if you want an adversary who is witty, crafty, deadly, and ruthless. Singleplayer's priority is campaign play, and campaign play is fundamentally based on two concepts.
1. You should win, and win a LOT (or your campaign won't continue for lack of prestige/because the game is over).
2. Your units need to survive for extended periods of time.

Both of those tenants are in total conflict with the multiplayer setting of 1 piece battles, but that's fine. Why make singleplayer more like multiplayer? Just play multiplayer. It's like there are two games in one.

A recent poll showed single-player is far and away more popular than MP. As much as I love a good MP game the pace does not suit everyone. I would like to be able to sit down to an hour or so of playing a decent player - the improved AI - as well as wait a day for my next MP turn.

Not related to the above, another reason I'd like an improved AI is so there can be less AI units on a map. As a scenario designer I would like to be able to field a strong AI force that isn't a horde and then nerf it to meet the player's likely force, rather than constantly buffing it. That comes from making Allied player scenarios, though.

Anyway, as I said at the start, if this is too OT I will move it to a new thread...

[El_Condoro]

Is it possible to have 2 AIs? The reason I ask is that a new AI would require all existing scenarios to be redesigned or at least checked for balance.

The current AI could be "sealed".
A new AI (exe) created.
The scenario editor would have a option button to select which AI exe to use for a scenario (or a whole campaign).
The player could just play a scenario or campaign without worrying about the AI or - optionally- swap to one or the other.

[deducter]

My suggestions under "basic techniques" numbers 1-5 are all things that are part of the game's mechanics that a human player can very easily do. Any of those things require minimal effort. One good thing about the PzC AI is that it doesn't cheat, except on Manstein, which you can think of as "loading the dice" in favor of the AI. (Although it is more like that it gives the AI more dice to play with.) Hence, I am strongly in favor of equalizing the AI's capabilities so that it can take full advantage of very simple game mechanics that every player does. Even something as simple as adding a second cycle to the AI's unit move list would tremendously increase the difficulty. That said, maybe you can add a dice roll to the AI to determine whether it will do those things correctly. Bombarding a unit with a bomber before checking whether there's AA cover or not? Roll a dice. On Colonel the AI might fail half the time, on General maybe 20% of the time, to simulate the mistakes a player make.

I think a more "human-like" AI is also unplayable for the campaigns, but would be interesting to have as a bonus difficulty. It would be much, much harder than Manstein, and I'm sure the forums would be flooded with complaints that the AI has too many units for campaign play, that their paratrooper drops are hax, or that there's no way a human player could pull off encirclements. I actually think this AI could be very good for skirmish maps or even MP play, for a nice preview for MP. Which, btw, I was hoping there'd be much more support for in the expansion. Replays??? Ladders???

[Chris10]

A separate question is, what new options for configuring the AI are needed in the Editor to enable better user-created single player scenarios.

Apart from a lot of things already mentioned there are a few things coming spontaneously to my mind
GENERIC:
Expand equipment file with another column TRANSPORTS where the allowed transports for that particular equipment (max. 4 ?) could be assigned with their ID. That would allow for great variety in transport and more tactic depth and historical accuracy.

EDITOR/AI
1.
Export basic AI settings to .csv or .txt where people can set priorities,unit purchase preferences, preference targets (example: fighter attack fighter priority 80%) etc.
2.
Expand script commands and checks with more commands,conditions and events and export to .txt. Its easier to write scripts (or generate) for big scenarios instead of using the editor.
Of course some docudemons would be nice .
Some new commands.conditions and events...just brainstorming
a.
b. remove unit xxxxxx,class xx (removes unit from scenario)
c. country XX (country ID) alliance 0 or 1 (makes country change sides (Romania 1944)
e. occupy hex xxx,yyy (gives hex to country x independently from its current flagg)
f. attack hex/zone
g. remove country XX (removes country and all its units from scenario) >flaggs to neutral

GAME:
a. New equipment available window also to show up for allied countrys and SE units
b. In-Game pop up window in order to display text/picture for news,briefing or whatever...triggered by condition (occupy hex,specific date or any other trigger to be specified) with ON/OFF option
c. Make AI as smart as possible (Recon,Bridge Eng,Mass Attacks etc) with ON/OFF option or tie to existing difficulty levels...

SPECIFIC:
If PzC really freezes or crashes when reaching 1GB RAM usage then my guess would be a stack overflow...
Stack Data is set to default limit of 1GB no matter if 32bit or 64bit system...
Is this really appropiate nowadays ?
There are some more tools in the box...cmon...> Static allocated Data > Dynamic Data >
At least the game should be able to adress the usual 1,75 GB for a 32bit application running in a 32bit OS...
various options now:
fix this by reorganizing your code
or
increase stack size ?
In that case: > linker /stack option.
Properties - Configuration Properties - Linker - System - Stack Reserve Size

[ivanov]

I'd be careful with making the AI too powerful, because it could make most of the campaigns simply unplayable. Of course a new AI would require a different approach to the campaign/scenario design.

Two AI behaviours that should be fixed IMO are:

1. AI units set to static/passive defenders schould be able to attack the adjacent enemy units when the odds are favorable.

2. AI should protect it's arty and don't place it next to the players units.

Other changes mentioned already like the correct AI use of the arty suppression or dealing with the fighter escorts, would dramatically increase players casualties and core unit loses. I can imagine that most of the people wouldn't be too pleased with it. The current AI works mostly fine in the Eastern Front DLC's and simultes well the way Red Army was waging the war there. The Western Front would be a completelly different story, as the Germans were always taking higher loses than the Allies. So desiging for example Western DLC's would be a reall challenge. Let's see how the Afrike Corps will work out...

So in general, some fixes and tweaks of the AI's bizzare behaviour are needed, but I wouldn't go to far with it.

"Be careful what you wish for you just might get it".

[Wildthing]

I'd be careful with making the AI too powerful, because it could make most of the campaigns simply unplayable. Of course a new AI would require a different approach to the campaign/scenario design.

Two AI behaviours that should be fixed IMO are:

1. AI units set to static/passive defenders schould be able to attack the adjacent enemy units when the odds are favorable.

2. AI should protect it's arty and don't place it next to the players units.

[...]

So in general, some fixes and tweaks of the AI's bizzare behaviour are needed, but I wouldn't go to far with it.

"Be careful what you wish for you just might get it".

Yes Ivanov, you are completely right! For example: with the masses of enemy fighters, if used like humans do, the player can not stop them from getting all bombers within 2-3 turns. And an AI without any big mistakes is very hard to match, in every level of difficulty.

[VPaulus]

"Be careful what you wish for you just might get it".

Very wise. I don't want a perfect AI too. Again, this is not a chess game.

[Rudankort]

Great discussion so far, thanks guys. I fully understand that some changes in the AI can break existing content, and will watch out for this aspect very closely. If any changes are so big, they make the base game unplayable, they will be separated into an additional AI level which will be used in Africa, but won't be used by default in main and grand campaigns. It will be possible to enable this additional level manually, which will be good for veteran players, while the rest will just ignore this.

And yes, I fully agree that the AI in such a game should not be chess-like. Not that it is even remotely possible to implement this kind of AI in Panzer Corps, but still...

[huertgenwald]

Sorry, i'd like to propose an UI wish, again .

If i want to repair a unit like this one:



neither repair options work "perfectly"

a) cheap repair would cost me a star (overstrength)
b) will be "too costly", i can afford to drop to 200 exp. in the worst case.

But i can't do that, since elite repair fills up my tank to strength 10.

If i could fill up two ranks by "smart repair" like this :



i would be able to fill up to 10 cheaply !

The one thing which needs changing is that elite repair only increments in single steps, as it does from strength 10 on upwards.

TiA

[billmv44]

I'll be happy if we get some incremental improvements that eliminate the obvious foolish mistakes the AI makes. A smarter AI, yes! A brilliant AI - please, no!

Since the game is going to be around for a long time, there will be further opportunities for tuning the AI in the future.

The game is great now. Let's be conservative in making changes.