CS AI

[Hobbes]

Hi folks, just having a look at the scenario I made for Mike's Korean War mod. It's been 5 years since I wrote a scenario for CS and I never bothered much about the AI as it always seemed fairly poor - so I made scenarios for human v human play. Looking at it afresh after making scenarios for another game it seems that Talonsoft and Matrix might have missed a trick here.

Surely the global and individual Op fire settings could be set for the AI and human player at the scenario design stage rather than having to wait for the scenario to start? I would think it would help the AI immeasurably - and the human player could change his settings once the game is underway.

Another big improvement would be a similar method to instruct the AI to
have certain units hold a position at all costs (at least until forced to retreat). If a unit doesn't have this order it follows the scenario mission type as usual.

For more complexity units like indirect fire or HQ types would be less likely to move unless certain conditions are met. A bit late in the day for this I know.

Sorry if this has been discussed before - I have been away for some time!

Cheers, Chris

[Kool Kat]

Chris:

The dismal and pathetic AI has been a criticism of CS since the early Talonsoft days... been consistently ranked as one of the "top 5" items that need drastic improvement in CS... and the "scorn" of numerous past threads?

The "best" way to "overcome" HAL is simply don't play it?

Thankfully, since we are a ladder gaming club... with the majority of play being H2H (human versus human)... HAL's limitations and shortcomings can be "overlooked" by most players? :)

[Hobbes]

It seems a real shame to me. Talonsoft obviously put a lot of effort into the game but it seems strange to me that a few easy changes were not made that would improve the AI no end.

I watched the first turn of my Korean scenario and was dismayed to see units in perfectly good positions leave them to make ridiculous attacks. MG and mortar units charging towards enemy lines. The mind boggles at what the programmers were thinking. A few days work and the AI would have been so much better.

Cheers, Chris

[Larry Reese]

You can get around the AI's limitations in my opinion, at least to an extent, with some clever scenario drafting, but then, in the end, look at how old the system is and how difficult it is to actually program a convincing AI for any game. The sheer number of random variables in this game make the task almost incredibly daunting, or not impossible without major resources to commit to such development (at least in my opinion) and I just don't see anyone making that kind of commitment.

I never played it, but years ago there was a Midway game (if I remember correctly, I know it was naval warfare in the pacific) where the AI actually used different human programed strategies and attack plans based on your selected difficulty level (of course, re-playablity was pretty limited once you'd figured out what the strategy was for each level, but still a novel idea and interesting take on defeating AI limitations. Must have resembled a huge decision tree to my way of thinking.

Anyway, just rambling.

LR

[Huib Versloot]

Hi Chris,

There are several ways to make the AI perform better that were never utilized by the designers from Talonsoft. Probably because they had a lot less time than the 10+ years we had to ponder about the game and its behaviour.

A good example of a reasonably good performing AI is WF "**A Key Position" (Loon op Zand.scn), but also "**The Old Guys Got It Today" (Dairomont.scn) I made those especially to see how the AI behaves and to see if it is possible to make a scenario that has historical oob's (so no extra advantage in strength for the AI side) and no fixed units (except HQs). The two latter being the instruments most used in the old stock AI scenarios.

The secret of succes is in the placement of objectives; they are in a line from rear to front. If you only have objectives at the front, the AI will keep running towards them like an idiot even is this position is hopelessly lost.
Another point is that the AI is best as the defending side, but it can do som offensive things too. For MW there will be a scenario within a LCG where the AI will be on the offensive.
When the AI is defender make sure the AI setting in the header is very low. Default it says 100, make sure to put it to +/- 10 so that the AI won't take hilarious initiatives.
Setting individual OP fire helps during the Human players turn, during the AI turn the AI will still shoot at unfavorable targets unfortunately.
Lastly Do fix the HQ units otherwise the AI will its spend its supply capacity by moving all the time.
During testing of AI scenarios, make sure to set the game that you can see all the things the computer is doing. That way you can respond in your design to what you don't want the AI to do.

With the current AI it is possible to make challenging scenarios. The main reason I hardly make them, is because of the lack of audience for it.

Click here to see a nice AAR from someone who played "**A Key Position" against AI.


Huib

[Hobbes]

Interesting post Huib. I never thought about playing around with the placement of objectives with the AI in mind. A problem with fixing units in place for the AI is that it would cause problems if a human player took that side. I suppose a scenario could be made for human v human and a second version for play v AI - just to fix certain units (a fix if controlled by AI parameter would be good).

The few scenarios I have made have periods where a side should defend and periods when it should attack - the worst sort of scenarios for an AI opponent. (but also easily improved if it were possible to change the mission type during the game - even better with a slight random factor).

The shame is that they didn't have to spend ages improving the actual intelligence of the AI - they just needed to give scenario designers a few simple options to help the AI.
Cheers, Chris