🚀 go-pugleaf

Okay, here is a non-IBM specific topic that could be interesting.
Back in June, a guy posted an article in rec.games.design about
various AI techniques and whether or not these techniques could
be used effiecently in a computer wargame.  I will post a copy
of his message after I get his permission.

What techniques do others use?  I am looking into some goal based
design in which the developer designs each scenario (ambush,
defend site, convoy, mutual assault etc) as a hierarchy of goals.

For example, the convoy scenerio has the ultimate goal of protecting
the convoy.  To obtain this goal the computer can either make a
run for the border using the escorts to keep the player's units
at bay or the computer could use a portion of the escorts as scouts
and attempt to hide from and move around the player's units.

How would each style be decided?  Maybe the composition of the computer's
escorting units and convoy units could be a factor. Can't really
make a run for the border if there are slow moving transports.  Neither
could the computer do scouting if none of the escort units have good
scouting capabilities.  The computer probably wouldn't want to use
the escorts in a major fire fight if the escorts are weak either.  Terrain
could also play a role in the decision making.  The computer could examine
the terrain that surrounds the planned convoy route.  Does it provide
adequate coverage for concealing the convoy?  

The computer should dynamically modify its goals as the game proceeds.
Suppose the computer choose the run for the border approach.  It then
got into a firefight that severely crippled the escorts.  The computer
may then opt for the hiding approach.

This decision making should not be cast in stone.  Add some randomness to
the decision making (especially if the experience level of the commanding
officer is low).  The computer should be able to make the same mistakes
as the player.

So far I have been talking about team based goals.  The individual unit
goals also have to be added to the system.  These goals will be based on
the current team goal and the role of the specific unit.  

For example, using the escort scenerio and the run for the border goal,
each convoy unit (i.e. transport) should have a goal of running at flank
speed to the next waypoint in the convoy path.  The shortest path should
be used.  Each escort unit has the goal of protecting the convoy.  This 
means keeping the enemy at bay while on the run.

Another example, using the escort scenerio and the hiding goal, each
convoy unit should move based on info the scouts provide.  The same is true
for the escorts that accompany the convoy.  The scouts on the other hand 
have the goal of steering the convoy clear of any enemy units.  They do this
by scanning their area and reporting back to the convoy commander.

A major point about this approach (and any other game AI) is that the
computer should be limited to the same amount of info that the human
player can obtain.  For example, the computer should not have a birds-eye
view of the entire map unless the human opponent does.  The computer
should not use enemy unit damage levels as a decision maker unless the
human can obtain the same info about the computer units.  I hate it when
a computer cheats!  

Well, let me know what you think.  These goals are probably out-of-wack in
some ways but thats the whole point of goal based approach.  Try out a
set of goals.  If they don't work right then add some new goals or modify
the way certain goals are to be reached.

At the moment I am designing my wargame to support modem play and
2 player at the same monitor modes but I am also keeping in mind that AI should
be implemented in the future.  

Mark Wyman