Upgrading the AI

[michaelguth]

Phil Barker, (who's he?), once wrote that only if your light troops are defeated should you send in the phalanx.

I find this to be a critical problem for the AI in FOG. The AI routinely allows its light cavalry to be charged by light infantry when either a. the light cavalry can be charged by stronger troops on the next turn or b. the light cavalry can be hit in the flank or rear on the same or next turn. AI light infantry sometimes behaves similarly.

I think the AI would be much stronger if it screened for the situations above before deciding to take a charge at the halt with its light troops.

A second issue is deployment. The AI often deploys in groups around commanders. This often leads to one or both flanks having heavy troops separted by many hexes from the center group leading to certain defeat. I think in most cases the AI would be better served by taking all heavy troops and lining them up in the center or opposite the enemy camp. The AI could then run a screen to determine where it could gain an advantage over opposing heavy troops and rearrange its main battleline troops as it advanced to maximize the number of favorable matchups. Deploying with large gaps in the battleline, often with heavy troops and cavalry being posted in difficult terrain, is a recipe for defeat in detail against a determined advance.

[TheGrayMouser]

I dont think I have ever seen a game that features a random battle generator be able to program an Ai to either buy a quality army (as a human would buy) or deploy in a sensible manner, with the exception of a WW2 game or so, although in those types of games the initial deployment is less critical. I have seen the pattern you describle of little battle groups clustered around leaders, which are easily picked off in detail by the player...

[Scarz]

The AI would improve by 100% if it just deployed in a line and either never moved, or when it did, it just stayed in a long line. Watching the AI I can not even understand how its behavior has been set, it makes random weird moves, at time sending units backwards or to the sides never to join the battle.

[MesaDon]

Could it be that the AI is incredibly brilliant that our human brains can not understand or keep up with it? Does it plan at a deeper strategic and tactical level leaving us shocked and awed?

[TheGrayMouser]

Could it be that the AI is incredibly brilliant that our human brains can not understand or keep up with it? Does it plan at a deeper strategic and tactical level leaving us shocked and awed?

Maybe those devious AI's are tricking us into thinking we players are tactical geniouses as we stomp all over them, then when the machines make their true bid for conquest we will have underestimated them....

[Igorputski]

Well Conquest Medieval has a pretty robust ai. Great battles series is another at least the army lines stay pretty much intact in them. Tin Soldiers Caesar also has a very good ai but it's advantage during the whole campaign is numbers of idiotic barbarians vs the might armies of Rome. I also like the ai in HPS Gallic Wars but the ai in this game is piss poor pitiful even with all the handicaps of numbers. They need to rewrite it for single players as I'm not going to pour money into expansions with a piss poor pitiful ai like this.

[TimW]

Take Command 2nd Manassas has a pretty good AI - it's main weakness is being over-aggressive, which it shares with FoG. I've seen many FoG AI armies start in a pretty good defensive position (holding high ground, flanks well covered), then advance in to the middle of the map and throw that advantage away. In fact, I've never seen the AI even try to fight defensively. On the other hand, it has actually managed to form at least a partial line a few times since the last update and has got much better at handling its' light foot.

As far as deployment goes, I agree that it tends to be all over the place - in one game the AI Bosporans had a map that was basically clear on their side apart from the left flank, where about 1/3 of the map width and half the map depth was woods. Guess where the AI deployed pretty much all of its' lancer and bow cavalry and considerable amounts of heavy foot? By the time the cavalry had got out of the woods the battle was over. The foot in the woods seemed mostly to be going (literally) in circles, one unit following another round and round. Could be the AI isn't up to handling a scattered and mobile opposition very well, but even so, its' deployment was daft from the start. Maybe an AI "drilled" to think a little like the armies it's "commanding" would toughen it up a bit - basic stuff as has been suggested, like actually forming a pike phalanx rather than three mixed commands any two of which can be overwhelmed.

On the other hand, at least FoG's designers haven't resorted to tired old AI methods like outnumbering the human player, having an army that's in some other way much more powerful, or setting objectives that have to be taken very quickly or you lose and then defending them with overwhelming force.

In DAG battles the AI is generally beatable without too much trouble (unless it uses a Judean army with lots of light foot that easily outrun horses...).

[SRW1962]

In DAG battles the AI is generally beatable without too much trouble (unless it uses a Judean army with lots of light foot that easily outrun horses...).

I had my best ever win against the Judeans, I caused 82 break points for a loss of 0 and outkilled them almost 5 - 1.

The AI and especially the deployment does need some attention, especially when it deploys non barbarian armies. I have no problem at all with the AI deploying armies of barbarians in a hodge podge of units here and there, but when it does the same for Romans, Macedonians etc. I find it very strange to say the least. For me it wouldn't be so bad if the AI armies were simply basic set piece armies and deployments drawn up by the designers, with maybe a few random extras thrown in by the AI to make up the points. At least that way you would get a sense of the army it is supposed to be portraying, and it would look right from the start of a battle.

As for its playing ability, this isn't so bad when it gets close and has to make limited decisions to attack or not etc. but its the build up that makes me smile as it juggles the army around endlessly for no apparent reason. I think this is where the AI really loses it for me, and if this were fixed the battles would be much closer and better.

So to summarise a collaboration between human & AI in army makeup and deployment, and non-twitching AI movement would for me at least be a huge step forward.

[Examinondas]

IIRC, Rome Total War (and perhaps the previous and following titles of the series) used something like that. There were some "standard formations" from which the AI could choose, depending on the composition of the AI and (I think) the human army. The formations allowed for some freedom, but the basic placement of each troop type was fixed. Everything necessary to define a formation was specified in a text file, and thus it was possible to "improve" the AI by creating better standard formations.

[KenPortner]

I'm far from a great player, but the AI has never presented much of a challenge at all, even giving it the advantage.

I wouldn't care, except that I can't get al the multi player games I'd like.

[deeter]

I know from reading remarks by someone on the inside that Take Command 2nd Manassas was heavily scripted (particularly the scenarios) rather than having at brilliant AI. This became obivous in the open play battles. That said the scripting was good so the game was pretty challenging.

Deeter

[TimW]

I know from reading remarks by someone on the inside that Take Command 2nd Manassas was heavily scripted (particularly the scenarios) rather than having at brilliant AI

Yes - the scenarios were scripted so the flow of events required for them made sense. Commanding an "open play" division under the "command" of an AI controlled corps or larger formation is where the AI can really come into its own with TC2M - it can deal you some very nasty surprises indeed.