Passive attacking AI

[BornGinger]

2 hours ago, RepsolCBR said:

It has been a known fact for many years that the AI is incapable of conducting a good attack.

This thread is not about whether the AI is able to attack in a good way. What wham is mentioning in the first post is that the AI-attacking forces don't move from their setup positions to begin the attack but stay there as if they have a supply problem or as if they all have the runs.

I noticed the same with a quick battle yesterday and today. Although there is an attack plan for the AI-opponent the AI troops never starts to move but stay in their setup positions the whole two hours.

When I checked that quick battle in scenario author mode a couple of times the AI-troops sometimes began to move out to start the attack but more often didn't.

[BornGinger]

I think I might have found the reason to why the AI-attacking forces never began to move from the setup positions in this quick battle map. There were two other AI-plans with setup positions but without any move orders added to them. Those plans weren't set to "never use" which caused the game to load them with the result that the AI-troops just stood and waited.

I have now removed the setup positions in those two AI-plans and also set them to "never use". It should work now.

Thanks for your time.

[Glubokii Boy]

36 minutes ago, BornGinger said:

This thread is not about whether the AI is able to attack in a good way. What wham is mentioning in the first post is that the AI-attacking forces don't move from their setup positions to begin the attack but stay there as if they have a supply problem or as if they all have the runs.

But they did move...not much granted...but they did move...

In a not all impressive way as I understand.

This is part of the problem with the current AI programing...especially on the attack...

The unpredictability if the AI behavior...

I currently don't have that game so i'm unable to check the AI planning out.....

What is the size of the AI groups ? What are their initial orders (as set in the editor) ? What does the timing options look like for the initial orders.

Having to large AI groups can lead to some really wierd AI behavior...the moral status of the troops if they take fire can lead to rather long delays and messe up attacks...

The larger the AI group the bigger the problem...

 

[Glubokii Boy]

If not even scenario designers assosiated with BFC and tasked with designing the stock scenarios can get the AI attack to work even half decently....

There has to be something wrong here

[ASL Veteran]

Okay hang on a second here.  All any scenario designer can do is create an AI plan.  As to whether it is challenging or not is going to depend on the situation and the player.  In Drive Them Out the scenario is very difficult for the attacking force - even for a player attacking and I did the best that I could do but the Americans frequently get pinned down and the MG sections and the tank often have trouble spotting the enemy so they don't fire very much in support so it is what it is.  The scenario designer can't adjust the attack based upon circumstances or adapt on the fly.  A scenario designer can only do so much.  So as long as the Americans are moving then I've done what I can.  If the player stops the attack with minimal difficulty then I don't know what to tell you.  Just getting the plan to the state it was in took many many hours of work along with some trial and error and that's the best I could come up with.  It is what it is.  Not every scenario is going to be that easy for a defender though as different situations will yield different results.  If the American AI in Drive Them Out actually captured an objective then I consider that a win.   

Oh, and yes, I was originally answering the question of why the Americans were lined up not moving and not why does the AI plan not challenge me enough so whatever was going on initially seems to have been taken care of.

[chuckdyke]

I had a go myself as defender opened fire by establishing fire arcs. The response was a typical silly attack by the AI. It stopped attacking after a while I pressed the cease fire button, and their morale made them ineffective as an attacking force. @RepsolCBR I completely agree with him. To win this as defender come out of your hides and counter attack. The Germans don't have artillery or armor which makes that option hard. I read the scenario and there were no orders which may function as trigger for the AI. Apart from the heavy machine guns in the back.

[Glubokii Boy]

ASL Veteran....my comment wasn't an attack on you...sorry if it came out that way ! ...

It was ment to show the limitations of the editor...and not any shottcommings on your part.

I apprisiate your work...as well as that of the other stock designers....thanks

But your comment also highlights one other designgoal of BFC that might need some considderation...

- the fact that ALL stock scenarios have to be playable from both sides.

The simple fact is that a scenario that has primarely been designed to provide the player with a good challange as the attacker can't do the same with the player as the defender...the AI will be far to underpowered to handle such a task....such in your scenario above.

Ones again...it is not your fult ! It's a gameengine limitation...

[chuckdyke]

8 hours ago, ASL Veteran said:

Okay hang on a second here.  All any scenario designer can do is create an AI plan.  As to whether it is challenging or not is going to depend on the situation and the player.  In Drive Them Out the scenario is very difficult for the attacking force - even for a player attacking and I did the best that I could do but the Americans frequently get pinned down and the MG sections and the tank often have trouble spotting the enemy so they don't fire very much in support so it is what it is.  The scenario designer can't adjust the attack based upon circumstances or adapt on the fly.  A scenario designer can only do so much.  So as long as the Americans are moving then I've done what I can.  If the player stops the attack with minimal difficulty then I don't know what to tell you.  Just getting the plan to the state it was in took many many hours of work along with some trial and error and that's the best I could come up with.  It is what it is.  Not every scenario is going to be that easy for a defender though as different situations will yield different results.  If the American AI in Drive Them Out actually captured an objective then I consider that a win.   

Oh, and yes, I was originally answering the question of why the Americans were lined up not moving and not why does the AI plan not challenge me enough so whatever was going on initially seems to have been taken care of.

The Troop Leading Procedure could be rescripted. Warning Order: Expect an attack any moment. Stop and repulse it, organize patrols, and destroy targets of opportunity. It will be a very playable game as a defender. Thanks for your effort. I had a 2nd go as a defender and looked for intel and unit identification. Then you discover the HMG's in the buildings. We can have Repulse and Pursue as the scenario. Give the defender a touch objective is something to be looked at.

[wham]

While I understand the challenges of designing AI plans as well as the fact that Combat Mission is less a game and more a simulation of sometimes unfair combat conditions, for some reason I expected the base game scenarios to be balanced out to such a degree that they would provide a challenge no matter which side you played on. Once I noticed this was not always the case, I got curious and opened this thread. So far I see three issues:

1. The attacking force and scenario combined often make it so that even if the defending played does not play and just skips all their turns, the defender can score a major victory or tactical victory. This means that even if the defender tries to actively play on their side, they can only really do worse than if they didn't play, which feels pointless and tells me the scenario is likely not worth playing as a defender. My project to catalogue what maps this occurs on is just a curious effort to see how common this is.

2. Occasionally seeing attacking AI units not move some of its units at all.

In the past I've seen entire squads, especially heavy weapons teams, staying back and not moving. In this round of testing I've mostly discovered things like transport trucks, single HQ's etc not moving. In "The Muddy Affair" one of the American tanks had not moved from start, and for some reason the AI had also not dismounted any of the units on top of it, which I think amounted to a light mortar and 2 MMG's.

3. During testing I've also discovered a third problem: the AI seems to run out of planned actions 15, 20 or even 30 turns before the mission time ends. I found that, unless the defender counterattacks, the attacking AI just sits in place toward the end of the scenario, not making any attempt to reach further objectives or to inflict casualties on defender, despite not holding any objectives, or just holding one objective of, say, six.

I'm not going to start preaching on how the game is "literally unplayable" because of this, or whine on how we "need better AI". These things are not something easily fixed, and in far less complex wargames than Combat Mission, other developers regularly have to stoop down to having the AI cheat in order to be challenging. Spawning more troops for the AI to just throw into the fray isn't really a feasible idea in Combat Mission, and so the massive challenge falls on the scenario designers to do the best they can with the AI plan. Somehow I've just had this fantastical notion of Combat Mission having way better and more competent AI than its competitors since the days of Combat Mission 2 where I joined the series as a teenager. Now I've learned more of the limitations of the AI, so I guess I feel a bit underwhelmed.

I've still got 5 of the largest scenarios left to test, but so far there have been only 3 of the very smallest scenarios where the AI was able to score a win as the attacker (I didn't test meeting engagements). It seems that as soon as the AI reaches the first one or two of its objectives, it either runs out of plan to follow, or the chaos and casualties make it unable to even make an attempt, resulting in problem 3 described above.

[chuckdyke]

26 minutes ago, wham said:

While I understand the challenges of designing AI plans as well as the fact that Combat Mission is less a game and more a simulation of sometimes unfair combat conditions, for some reason I expected the base game scenarios to be balanced out to such a degree that they would provide a challenge no matter which side you played on. Once I noticed this was not always the case, I got curious and opened this thread. So far I see three issues:

1. The attacking force and scenario combined often make it so that even if the defending played does not play and just skips all their turns, the defender can score a major victory or tactical victory. This means that even if the defender tries to actively play on their side, they can only really do worse than if they didn't play, which feels pointless and tells me the scenario is likely not worth playing as a defender. My project to catalogue what maps this occurs on is just a curious effort to see how common this is.

2. Occasionally seeing attacking AI units not move some of its units at all.

In the past I've seen entire squads, especially heavy weapons teams, staying back and not moving. In this round of testing I've mostly discovered things like transport trucks, single HQ's etc not moving. In "The Muddy Affair" one of the American tanks had not moved from start, and for some reason the AI had also not dismounted any of the units on top of it, which I think amounted to a light mortar and 2 MMG's.

3. During testing I've also discovered a third problem: the AI seems to run out of planned actions 15, 20 or even 30 turns before the mission time ends. I found that, unless the defender counterattacks, the attacking AI just sits in place toward the end of the scenario, not making any attempt to reach further objectives or to inflict casualties on defender, despite not holding any objectives, or just holding one objective of, say, six.

I'm not going to start preaching on how the game is "literally unplayable" because of this, or whine on how we "need better AI". These things are not something easily fixed, and in far less complex wargames than Combat Mission, other developers regularly have to stoop down to having the AI cheat in order to be challenging. Spawning more troops for the AI to just throw into the fray isn't really a feasible idea in Combat Mission, and so the massive challenge falls on the scenario designers to do the best they can with the AI plan. Somehow I've just had this fantastical notion of Combat Mission having way better and more competent AI than its competitors since the days of Combat Mission 2 where I joined the series as a teenager. Now I've learned more of the limitations of the AI, so I guess I feel a bit underwhelmed.

I've still got 5 of the largest scenarios left to test, but so far there have been only 3 of the very smallest scenarios where the AI was able to score a win as the attacker (I didn't test meeting engagements). It seems that as soon as the AI reaches the first one or two of its objectives, it either runs out of plan to follow, or the chaos and casualties make it unable to even make an attempt, resulting in problem 3 described above.

You need to counterattack and give the defender a touch objective. Even in real life you just don't sit at your position. Replay and give yourself some more challenges if it is too easy. My parameters in Shock Force: one body bag is one too many. NATO loses an MBT and the newspapers shout it from the rooftops. In Final Blitzkrieg battles give yourself the parameter when you defend as Germans the Americans must surrender. The editor writes down the parameters you can change them. 

[BornGinger]

14 hours ago, RepsolCBR said:

Having to large AI groups can lead to some really wierd AI behavior.. The larger the AI group the bigger the problem.

That's the reason to why it's good to use all, or most of, the 16 AI-groups in quick battles (or to save a quick battle map we got with the game under another but similar name and split up the AI-groups in the battle plans into smaller ones so there are more of them). It's a bit more work to give all those 16 groups orders but might make the quick battles more fun.

Quote

In "The Muddy Affair" one of the American tanks had not moved from start, and for some reason the AI had also not dismounted any of the units on top of it, which I think amounted to a light mortar and 2 MMG's.

I had the same happening to me in that scenario. But that tank doesn't just stand without moving. If you click on the tank you will see that it has been bogged down in the mud.

I have complained in other threads about how easy it is for vehicles to get bogged down in a little splash of mud and suggested that there could be different kinds of mud; deep mud, for vehicles to get stuck in, and shallow mud, where vehicles don't get stuck.

 Another thing is that when AI-vehicles do get stuck there is no function to make infantry leave those vehicles automatically when the vehicles haven't been moving for a set amount of time.

I hope that BF will make some changes to mud and how the infantry behaves when the vehicle they're in, or on, get bogged down.

Edited January 14, 2021 by BornGinger

[Glubokii Boy]

32 minutes ago, BornGinger said:

That's the reason to why it's good to use all, or most of, the 16 AI-groups in quick battles 

It most certainly is...as it is in scenarios...I always do 

But 16 AI groups it not enough. Thats one of the points that we 'more scripting tools'-guys have been saying for years now...

It might be enough for a reinforced company on the attack at best or a battalion on the defence...but if the AI attacks with a reinforced company then the player can not be allowed much more then a platoon or maybe  two to command (but that would be pushing it)...playing with a single platoon might be a bit on the small side to make for a fun scenario...

Increasing the amount of AI groups to atleast 32 would make even attacking battalions a possibility.

 

[Glubokii Boy]

Wham....

I belive that most of the issiues you are seeing comes from the fact that the majority (if not all) of these scenarios has from the start been designed to be played as the attacker. The reason there is a player defensive option comes from BFCs demands that all scenarios should be playable from both sides....but i'm guessing that most of the designers have added this option simply because they had to...to mest the demands from Battlefront.

These player defensive scenarios have not been designed with a full commitment to design good defensive fighting....they are simply a neccsary biproduct of the original player attack scenario....

To be able to judge what the attacking AI is currently able to do you will need to play a scenario that has been purposly designed to be played as the defender...

In these scenarios all efforts have been made to make the AI attack as good as the current gamerngine will allow.

I don't remember if CMFB has any such senarios included though...

 

[Glubokii Boy]

5 hours ago, chuckdyke said:

You need to counterattack and give the defender a touch objective. Even in real life you just don't sit at your position. 

No...You do not need to counterattack. The goal of the attacker (AI) is to achive its objective...not to sit an wait for some flipping counterattack.

If the attacking AI troops are not moving forward and the reason for this is the lack of a player (defender) counterattack...that would be a very strange scenariodesign.

 

 

[chuckdyke]

You can give both sides objectives. Read the Battle of Long Tan an NVA Regiment vs an Australian Company (D Company 6RAR). The defense was concluded with an Australian counterattack. M113 APC's combined with NZ 105 mm Artillery. Waiting in the rain for the next NVA attack was poor tactics. Do something unexpected has surprising results.  Nothing strange of having a counterattack in your scenario. It made that game a little bit more playable I just did it for fun. Observed numerous cases of buddy aid and started probing their positions it made it less boring. I don't take one iota notice of the victory conditions and made it a seek and destroy mission. People buy the game to have some fun. If some games lack in design, change the parameters. After all, I bought the games with my own money. The AI can't attack it doesn't give area fire and artillery support is without the input of a forward observer. I can make my bed on top of the church tower and the AI leaves me alone and call strikes all over the place. I am very happy with combat mission when I play it with my mates. The AI is just practice. 

[Glubokii Boy]

13 minutes ago, chuckdyke said:

You can give both sides objectives. Read the Battle of Long Tan an NVA Regiment vs an Australian Company (D Company 6RAR). The defense was concluded with an Australian counterattack. M113 APC's combined with NZ 105 mm Artillery. Waiting in the rain for the next NVA attack was poor tactics. Do something unexpected has surprising results.  Nothing strange of having a counterattack in your scenario. 

No it is nothing strange with counterattacking per say...but to have an AI attacking force rely on (wait for) a player counterattack before proceeding would be strange imo...

As the attacker you do not voluntarely want to give the initiative to the defender and then having to react to that. It is the defender that is supposed to react to your actions...not the other way around.

 

[wham]

I attached an excel sheet with a list of mission and their outcomes when played as the defender, but not giving units any orders.

I did not test meeting engagements, and two other missions kept crashing for me, so I ended up giving up on testing them.

Out of 18 scenarios tested:

• Attacker was able to score some kind of victory in 5 scenarios, mostly smaller ones
• Attacker managed a draw in 2 scenarios, though I feel these were mostly due to the defender having objectives that required them to take on a more active role
• Attacker lost against a passive defender in 11 scenarios, regularly granting the defender major victories

The following scenarios seem to be somewhat suited for playing as a defender, though with an active player most will likely prove easy to win.

• Die Patrouille
• Trouble with Siegfried
• Lanzerath Ridge
• Hung up and Chenneux
• Wax Museum

I discovered that the mission I had mostly had trouble with in the past was Hung up and Chenneux. An entire heavy weapons platoon of the attacking side fails to move from its starting position, sitting in a neat line at the attackers edge of the map and never joining the battle. This behaviour repeated every time I played this scenario, before and during this particular testing.

CombatMissionAnalysis.xlsx

[Glubokii Boy]

Thanks for running this test....intresting results.

But the simple truth is as I have stated above. Designing good player defensive scenarios is difficult...even if the goal is to design a dedicated player defensive scenario (no option to play as the attacker).

The tools to do so are simply to limited right now. 

To design a player defence version of an initially designed offensiven scenario is...not possible ! Not a good one atleast. The demands on the AI will be far above its capabilities. This fact is known by the stock scenario designers and consequently they don't have the intrest, time or energy to commit fully to include a GOOD defensive version of their scenarios. Simply because it can not be done...

These defensive versions (for the most part) i belive have very simple AI plans and have seen very limited playtesting i'm betting....simply because there is no reason to do othervise...

They will not work and very few will ever play them with the goal of getting a challeging fight...so why invest lots and lots of time developing them.

They are included simply because they have to be...as required by BFC.

 

 

Edited January 14, 2021 by RepsolCBR

[chuckdyke]

10 hours ago, RepsolCBR said:

No it is nothing strange with counterattacking per say...but to have an AI attacking force rely on (wait for) a player counterattack before proceeding would be strange imo...

As the attacker you do not voluntarely want to give the initiative to the defender and then having to react to that. It is the defender that is supposed to react to your actions...not the other way around.

 

You play the game as defender to win is easy press the cease fire button and you end up with a 'Major Victory. In the case of an attacker, you end up with a Major Defeat. If you like to end up with a Total Victory you need to look for enemy forces. I played and after 10 turns I pressed the cease fire button. I kept all my objectives so Major Victory. But later I realized as defender you can make it more interesting. Set your parameter to "Total Victory" and you end up with an enjoyable scenario even as defender. The purpose of wargaming in the military is to design a model which can be used in real life. I thank @whamfor making me look deeper into the issue. I know now to play as defender can be enjoyable. You play your game on your computer and giving yourselves different parameters is fine. In this game I saw the M4 Sherman just standing somewhere. Organized a tank hunter party to destroy it. The original attack party doing buddy aid in the forest capture them for their intel. Just a few suggestions you can do as defender which is standard procedure. 

[Glubokii Boy]

1 hour ago, chuckdyke said:

You play the game as defender to win is easy press the cease fire button and you end up with a 'Major Victory.  

Prematurely pressning the ceasefire button in a defensive scenario is not a win...it's cheating ...

Anyway....the point i'm trying to make is not primarely about tactics as the defender....but about the current difficulty in designing an AI attacking scenario...

And the fact that you as a player needs to tweak these scenarios into more of a meeting engagemang or even yet another player attack scenario in order to enjoy them seems to me to be yet another indicator to the fact that these player defensive scenarios are not working....

 

[chuckdyke]

1 minute ago, RepsolCBR said:

Prematurely pressning the ceasefire button in a defensive scenario is not a win...it's cheating ...

Anyway....the point i'm trying to make is not primarely about tactics as the defender....but about the current difficulty in designing an AI attacking scenario...

And the fact that you as a player needs to tweak these scenarios into more of a meeting engagemang or even yet another player attack scenario in order to enjoy them seems to me to be yet another indicator to the fact that these player defensive scenarios are not working....

 

 

 

 

If I get bored, I press the cease fire button nothing to do with cheating it takes another player before you can cheat. Defending involves counter attacking if you think you win by defeating the AI you're wrong. You merely have come up with a model which may work against a human player. Win as an attacker challenge another player who is familiar of the scenario and replay it as a quick battle on the same map. Both players switch sides. I never use words like cheating, you may offend people. The AI can't attack anyone who has owned the games for a few years knows it. Defending is a little better but not much better. I am very happy that I play mostly on Hot-Seat against the AI I would have stopped years ago. In this scenario I used the cease fire button because the AI stopped attacking. Same as you would do against a human player, against a human player you can say no I want you to surrender. If you want the AI to surrender you need to counterattack and destroy his forces. At least in this scenario, nothing but respect for the designer he has done nothing wrong. It is meant as to attack only but they are obliged to have the defend option. Future replies won't be answered. 

[BornGinger]

6 hours ago, chuckdyke said:

If I get bored, I press the cease fire button nothing to do with cheating it takes another player before you can cheat.

It seems to me that you should avoid playing the scenarios and quick battles against the AI and spend most of your time on TheFewGoodMen and TheBlitz so you can play against human opponents.

Or you could edit the scenarios and change the experience of the defenders to be conscript or green and save the scenario with another name. Maybe fighting the AI will get more challenging and interesting then.

Edited January 15, 2021 by BornGinger

[chuckdyke]

6 minutes ago, BornGinger said:

It seems to me that you should avoid playing the scenarios and quick battles against the AI and spend most of your time on TheFewGoodMen and TheBlitz so you can play against human opponents.

 

Is it your hobby to tell other people what they should do? I have 2 human players where I live, and we visit each other to play on Hot-Seat. I play only Quick Battles against a human opponent. Scenarios are great to test tactics. A Few Good Men you play by Email and must disable the security of your PC. No thanks not for me. The AI is not hard to beat as attacker or defender. To get your best Intel? learn the editor and you know the way your digital opponent thinks. You can use the Eifel Tower in Paris figuratively speaking as your spotting and communication post. The AI even if it has all the assets, it will leave you alone as long as you stick to spotting and communicating. It doesn't know area fire. Happy gaming and do me a favor, and don't bother me with your advice what I should or should not do. 

[Glubokii Boy]

7 hours ago, chuckdyke said:

Future replies won't be answered. 

Answer or no answer...i will just end by mentioning that any of my replies have not been directed at you personally...but generaly at any player.

And while discussing the quality/challange of the AI i don't think that a statement like this...

9 hours ago, chuckdyke said:

You play the game as defender to win is easy press the cease fire button and you end up with a 'Major Victory. 

might be a fair description of the AI quality. If a player does that he is not even giving the AI the chans to win. Offering a human player a cease fire is one thing because he can reject it. The AI rarely will...Doing so would be cheaing the scenario and the result can never be considdered a 'major victory'. But you are correct though that i and i guess most others would do the same thing if we get bored. But a victory ? NO...

37 minutes ago, BornGinger said:

It seems to me that you should avoid playing the scenarios and quick battles against the AI and spend most of your time on TheFewGoodMen and TheBlitz so you can play against human opponents.

That is obviously the right way to do it if you want a real challange but H2H may not be an option for every player....for various reasons.

[wham]

While I see the merit of setting myself extra goals as the defender, I feel this is akin to modding the scenarios myself, which is not what I'm looking for when purchasing a game. I feel this issue could pretty much be remedied by admitting to the shortcomings of the scenarios and engine and AI, and adding a note either in the game description, manual or scenario descriptions that all scenarios are best played as attacker unless the mission is designed and tested to be challenging to play as the defender as well.

Multiplayer is a different matter, and I feel nearly all scenarios are perfectly playable as either side with an opposing human player, but as much as I would love to have one, I remain stuck playing with the AI due to the nature of the game making it difficult to approach or maintain focus in for a lot of people. 120 turns, with each turn taking half an hour to an hour or more to set up, is just too much time to invest for my friends.

Additionally, in some scenarios there are already such objectives in place in the mission design, with several touch objectives set for the defender far outside of their starting area, forcing the defender to push out of a secure position to capture forward positions before the advancing attacker can reach them. At least in non-historical scenarios, I feel this might be a suitable way to force the defender to be active as well, making it so that the defender cannot simply win by sitting still and being passive. Likewise, in many scenarios a key part of the defenders forces arrive later as reinforcements, but even in these scenarios the attacking AI currently loses due to a failure to advance past initial objectives, meaning that even with the defenders reinforcements sitting at the edge of map and never engaging the attacker, the defender ends up winning. If anything can be done in the current engine and level of AI, I think these tools might be key to creating interesting attack/defend scenarios that are playable as either side against the AI. However, this is speculation on my part, as I've never really studied the AI planning tools in any depth.