AI

[mav1]

Happy to hear the AI is being improved, but will this stop the ai leaving units behind doing nothing when its on the attack. In one battle I played, I was attacked by personnel carriers and infantry, which I easy defeated with sherman tanks. In the end I won the battle easily, when i looked at the map at the end of the battle. I noticed 2 sturmIII's and 1 PzIV hadn't moved in the battle. Maybe the crew's of the tanks were having a picnic while there comrades were dieying.

This has happehed several times in battle.

[timu24]

Were there any farmhouses nearby? Maybe they were fratinizing with the locals (if you get my drift)

Tim

[mav1]

There were no farmhouses nearby. Maybe the crew were fratinizing amongst themselves? (if you get my drift)

[SteveP]

Keep in mind that the AI is basically a "path-finding" method. That is, it's looking for covered routes that get its units closer to the flags. AFV create a greater problem for the AI, because they don't have as many terrain options for cover (for example, they can't go into rough terrain to protect themselves. For them, it's mostly woods and, even more important, reverse slopes. If the AI does see any good ways to move AFVs forward, while minimizing the time spent in the open, there's a good chance it won't.

That may or may not explain what you saw. I don't know, for example, why the carriers moved up, but the AI may consider speed and profile in deciding what is safe to move or not.

[mav1]

The poor path-finding method of the ai gives a big advantage to a player. As a player I have no hesitation in moving forward my troops, cover or no cover if the moment demands it.

[mav1]

The game I suppose lacks a more human element on the part of the ai. If the crew of the tanks had heard about the massacre taking place, they would have rushed to the rescue, cover or no cover. But then again this is balanced by the player not giving a damn about his own troops unless it interferes wth him/her winning the battle. That's why having a campaign is important, in adding to the realism of the game, because then losing troops will count. It will be interesting how bfc will cope with creating a campaign. It will be hard to get it right.

In battle it is important to get more troops than your opponent in a critical area. If the ai is failing in using all its troops, it could miss the bigger picture of the battle being lost.

I hope the new path creation plan in combat mission x2 for the player will leave a random element for the ai. Otherwise as the scenario creater, I will know were the attack is going to come from.

Hopefully this will stop tiger tanks travelling through a cementry to get to it's objective rather than using a road near by.

Thanks for your input SteveP, its good to get someone elses view on things.

[SteveP]

From my reading of posts in this forum, I have gotten the impression that a lot of players have unrealistic expectations about what the AI "should" be able to do, and/or have exaggerated ideas of what the AI is able to to do already (for example, thinking that the AI can do calculated ambushes).

BFC has invested most of their AI programming skill in development of the TacAI. The TacAI is designed to make all of the decisions about where and when to direct fire, and handles all of the reactions of the units when under fire or when threatened. It does not deal with movement (except for covered panics and routs). IMHO, the TacAI makes relatively few "stupid AI tricks," but that as you go up the ladder in the AI system, you start to run into more of those tricks. My understanding and expectation is that in CM2, the TacAI will be even more complex and powerful, but that there will be little change to the other components of the AI -- other than, I gather, some ability to respond to scripted "values" that designers can build into scenarios.

I think it is a good thing, if you are going to play the AI, to have a pretty detailed understanding of how it works, even though that may give you an advantage. Otherwise, you won't know if you are winning because you've learned good tactics, or because you are knowingly or unknowingly taking advantage of the AI's limitations.

[Cpl Steiner]

Originally posted by SteveP:

From my reading of posts in this forum, I have gotten the impression that a lot of players have unrealistic expectations about what the AI "should" be able to do, and/or have exaggerated ideas of what the AI is able to to do already (for example, thinking that the AI can do calculated ambushes).

SteveP,

I think you are correct in what you say, and that in CM:SF, the computer will probably still have to take the defending side in most battles, as the defending side doesn't need much strategic AI beyond an understanding of how to set up a decent defensive position (for non-fixed setups only).

Having said that, I would have thought you could still set up a viable CM:SF scenario in which the Syrians ambush a US column, with the player taking the US side and the computer taking the ambushing Syrian side. The Syrians would have fixed setup zoned dictated by the scenario designer, and would just have to wipe out the US column from their starting positions or die trying. As the US side, the player would have to get out of the ambush or destroy the ambushers. I can imagine this would make for quite an exciting scenario.

[stikkypixie]

oops

[ March 11, 2006, 10:54 AM: Message edited by: stikkypixie ]

[SteveP]

Originally posted by Cpl Steiner:

Having said that, I would have thought you could still set up a viable CM:SF scenario in which the Syrians ambush a US column, with the player taking the US side and the computer taking the ambushing Syrian side. The Syrians would have fixed setup zoned dictated by the scenario designer, and would just have to wipe out the US column from their starting positions or die trying. As the US side, the player would have to get out of the ambush or destroy the ambushers. I can imagine this would make for quite an exciting scenario. [/QB]

Clever scenario designers who understand the limitations of the AI have been able to create situations that resemble ambushes. Moreover, a lot of players (and I suspect a lot of scenario designers) don't realize that the AI automatically hides all its units at the beginning of a battle when it is on defense. It also apparently has the ability to unhide them in the middle of a turn (something a player can't do). It does this on a semi-random basis. It's a kind of pseudo-ambush effect that may have been intended for this purpose by BFC. I don't like it, because it ends up being a stupid AI trick most of the time, but we'll see what they decide to do in CM2.

Clever scenario designers have also been able to create situations that help the AI to attack or do meeting engagements. The best trick is to reduce the amount of movement the AI needs to get into battle. It's the movement toward contact phase that screws things up. That may be why in QBs the map is relatively shallow (more so than one would like if playing the defensive side). With a shallow map, the AI doesn't have to move as far.

[mav1]

I agree with you SteveP, that how well the AI work's is down to the scenario designer. A few of the designers in the scenerio's included in the game got it wrong. But I wonder if you give more time to the AI to think, could you improve the AI. If during the time the player is thinking of his/her move. The AI uses the time to think of its move (dont know if this is done already). This approach is done on a game called Galactic Civilation, its supposedly got good AI because of it. Dont know if its relevent, but its a suggestion.

[mav1]

SteveP, what do you think of BfC's approach to CMx2. What I was hopeing for was an improve CM on a grander scale, with a campaign {to replace operations}. But they have gone for smaller action, based on urban warfare {I think}.

Also what are your ideas on how to make a campaign for combat mission. I would like to compare your ideas on how to do a campaign with my ideas.

[SteveP]

Originally posted by mav1:

I agree with you SteveP, that how well the AI work's is down to the scenario designer. A few of the designers in the scenerio's included in the game got it wrong. But I wonder if you give more time to the AI to think, could you improve the AI. If during the time the player is thinking of his/her move. The AI uses the time to think of its move (dont know if this is done already). This approach is done on a game called Galactic Civilation, its supposedly got good AI because of it. Dont know if its relevent, but its a suggestion.

I don't think it helps much to compare CM's AI to other games, because different game designers approach the problem with different design strategies. My understanding of CM's AI is that it tries to get an optimal result (using "fuzzy logic") from an accumulation of values that are either set in the code or are situational. BFC has to make improvements to that incrementally, in order to know if any one change is working successfully within a reasonable range of acceptable outcomes. IMHO, the performance of the AI could be significantly better just by making two improvements: recognizing command relationships and keeping units in command radius to the extent possible (which may be in the works based on things that Steve has said in other threads), and getting the AI to use other movements commands besides move and run (or fast for AFVs). Beyond that, I'm not sure what is realistic to hope for.

[SteveP]

Originally posted by mav1:

SteveP, what do you think of BfC's approach to CMx2. What I was hopeing for was an improve CM on a grander scale, with a campaign {to replace operations}. But they have gone for smaller action, based on urban warfare {I think}.

Also what are your ideas on how to make a campaign for combat mission. I would like to compare your ideas on how to do a campaign with my ideas.

If you did a search you'd find that this is a very old debate in this forum. There are a wide variety of personal preferences among the people who play CM, and BFC can't please them all. My own view is that I'm happy to see BFC working toward getting a game that produces very good results when played at a relative small scale (also it's a lot easier to play). Once they do that, then I think it makes sense to look at how best to scale it up.

[Cpl Steiner]

If you ask me, the best thing you could do with the AI would be to teach it how to coordinate two groups of units. It would then be able to recognize the concept of "fix" and "flank" - i.e. if one of your groups is engaged, keep that group where it is and try to work around the enemy's flank with another group. Something as simple as that would help enormously.