Hidden History

There is a lot of hidden information in “Armageddon Empires.”  That’s a design decision that is a double edged blade.  On the positive side, the computer is great at enabling hidden information in a way that a board game simply can’t without some type of record keeping gimmick that more often than not proves akward to implement and execute.  Not seeing every army on the board brings a great sense of tension to the game.  You know you should be looking for the enemy or picketing a flank but you also have other things to do.  So you take a risk and hope to god that nothing too awful wanders in from out of the desert to capture that thinly defended outpost.  You can also use the lack of omniscience to your advantage.  There is nothing more fustrating than trying to orchestrate a “sneak attack” that is doomed from the beginning because the enemy automatically sees it coming. 

From a programmer’s perspective it’s also a bit of a mixed bag.  If an AI can’t see into a particular hex then it can often be left out of an algorithm. Unfortunately, implementing the observation mechanics in computer code can be a real pain.  And the consequences often cascade down into other areas.  Adding something like a stealth mode for an army means a whole new set of rules and mechanics.  The complexity can snowball quickly.

Finally, I soon discovered that a game framework with a lot of hidden information can make the AI players look dumb, diabolically genius or just crooked cheating.  I built the game mechanics, interface and AI architecture from the ground up to avoid a cheating AI. Each AI player keeps a completely seperate copy of the game space and what it knows about it. It updates according to the same observation rules that the human player must work under.  Sometimes an AI army waltzes right past an easy target on its way to accomplish some goal and I have to stop and figure out why. Often the answer is that it simply just didn’t see it.  Coming up with good strategies for the AI to use its recce assets helps things out but in the end observation is about chance and sometimes the dice aren’t nice.

I’ve been doing a lot of testing in 4 player games (3 AI’s and a human) to watch the AI’s perform in a target rich environment and look for ways to fine tune their actions.  I play with a “switch” flipped that allows me to see everybody perfectly (i.e. everything is observed).  Normally a player will only ever see a tiny fraction of this….maybe he has a recce unit next to an enemy outpost when it is attacked by another enemey AI for example.  The AI’s are built to win so they try to gather information on who the best target is for focusing their offensive operations on.  They also keep track of animosity levels.  If an AI starts capturing outposts, destroying resource collectors, bombing armies then animosity builds up and it’s taken into account when deciding who is the biggest threat.  Often I’ll watch the AI’s tear each other to shreds.  Sometimes they are both gimped.  Sometimes one conquers the other and emerges with a better resource base and supply position.  This can be difficult from a game design view because it’s no fun for the human player to waltz right over two punch drunk fighters who have been savaging each other for the last 20 rounds.  All the human player notices is two wimpy opponents that don’t put up much of a fight.  The epic back and forth is never observed or appreciated.  Here is a screenshot of an instance where the Xenopods and Machine Empire found themselves fairly close to each other and with a lot of open terrain to manuever.  Between the two of them was a key special….an abandoned ICBM silo complex that acted as a free outpost. 

XenovsMachines1

After the outpost changed hands several times the Machines eventually came out on top.  The Xenopods had been mauled pretty badly so when the Machine army assaulted the Mother Hive it was a quick affair.  Now the Machines are probing for another conquest………me.

CaptureComplete

Comments are closed.