Dev Diary #08: Adaptive AI in Scenario Design in Four Easy Steps
With CS Middle East, one of the totally new concepts introduced to The Campaign Series was the Adaptive AI mechanism.
Berto wrote an excellent Coder Diary of it at the time, check it out: Coder Diary #30 -- The Adaptive A/I
[*] FYI: All previous Coder and Developer Diaries are listed in the Mods And Their Use post in the Mods and Scenarios folder.
To recap, with added emphasis by me:
ORIGINAL: berto
Coder Diary #30 -- The Adaptive A/I, part 1
[This Coder Diary is an adaptation, an abridgement, of the AdaptiveAI PDF included in the Middle East game's manual folder.]
The Campaign Series A/I -- One size does not fit all!
Arabs did not fight the same way as Israelis. Jordanians outclassed the Egyptians, who outclassed the Syrians. Israel's combat doctrines changed from 1948 to 1956 to 1967 to 1973 and beyond. The cocky Israelis at the beginning of the Yom Kippur War learned to give their revitalized Egyptian foes newfound respect, and learned to change their ways. Adapt or die.
From nation to nation, from war to war, from year to year, even from one battle to the next -- they all differed, so much evolved.
But in the traditional Campaign Series, as in most (all?) war games, there is one and only one A/I. Locked down tight. Mostly inaccessible to the scenario designer, much less to the player.
Until now...
We have devised a new customizable A/I System. Customizable by side, by nation, by scenario. Open to the scenario designer's tweaking. Also the modder's.
We call this new system: "Adaptive A/I." As in: An A/I that adapts to the side, the nation, the era, the individual scenario. (Not as in: Adapts real-time to changing game play circumstances. But that too will be implemented in future updates.)
The purposes of the Adaptive A/I are:
[*]To coach the A/I in ways and things it is too dumb to figure out by itself.
[*]To achieve game play "balance".
[*]Not so much to improve the A/I play as to better simulate actual combatant and scenario conditions. To achieve greater "realism" and "historical fidelity".
[*]To add randomness.
[*]Thereby increasing replayability.
These cross purposes are often at odds with one another. Some Adaptive A/I parameters will make the A/I opponent "better"; others will make it "worse". It all depends. Some players will like the increased uncertainty; others will not. It all depends. That's why the Adaptive A/I is optional. Select it or not as you wish.
AI parameters that can be tweaked per scenario, per side, per nation? In this blog entry, let us seehow it all played out with the hypothetical Crisis in Sirte Libya'85 scenarios.
First, let us revisit what Adaptive AI does. Even for H2H scenarios, AI does a lot of things, as one might guess at looking at the Adaptive AI groupings.
This, from the latest CS:ME 2.0 Beta init.ai file:
[*] 0-29 movement
[*] 30-49 opfire (opportunity fire)
[*] 50-59 IF (indirect fire)
[*] 60-69 combat, direct fire, assault
[*] 70-79 fatigue, morale, disruption, leadership, command
[*] 80-89 visibility, spotting
[*] 90-99 supply, ammo
[*] 100-109 activity
[*] 110-119 misc
[*] 120-129 opfire (opportunity fire)
[*] 130-139 SAMs
[*] 140-149 helos
So for me, pitting Gaddafi's finest versus the British Quick Reaction Force, in general, I wanted to achieve at least these two things.
[*] Adapt AI calculations to this particular set of scenarios. Sirte 1985 is a specific set of battles with all sorts of modern gear, and I want to have some unique features.
[*] Also, I want to beef up Libyan combat behavior a bit, as they are supposed to be the elite formations of Gaddafi's arsenal.