Akshay Joshi
I.  Title of the Project: 

II. Problem Statement. 
	To create a learning/evolving AI for use in a 3d space simulation FPS (
(First Person Shooter game)

III. Purpose. 
The purpose of this project is to create a genetic algorithm for the NPC's 
(non-playable characters), the computer controlled ships, of a space FPS.  The 
project is worth doing because it leads to a better understanding of what types
of plays by NPC's are easily countered by the player, and what types of moves the
player has more trouble with.  This project would be of use to game designers because 
they can base different difficulty level AI's on how easily players are tripped
up by a certain move, and translate that into a more complex AI.

IV. Scope of Study. 
The project will require the programming of numerous different strategies and 
tactics, such as being in a certain sector with respect to the enemy ship, or a 
certain shell distance, etc; it is not so much of a strategy, but more of a goal
state for the ship to reach, and each move the ship makes is towards achieving 
said state.  Then in each instance, the remote game calls upon the "mother" for
a random array of goal states from which the NPC's can draw upon, every couple 
minutes the states with the stronger record will reproduce while the states with
the weaker record are jettisoned. After several rounds, the remote game will 
send the record of each state to the "mother" and receive another array of 
random goal states.  After many games, the "mother" will be able to identify 
which states are clearly the best, and which ones need to be modified.  Difficulty
levels can be added in by changing the accuracy, aggressiveness, and decision time
of the NPC's

V.  Background and review of current literature/research in this area. 
A popular way of programming game AI is the SOAR method, in which the game 
identifies the current state, sets an ideal state for the game to progress into,
and moves all the NPC's in ways that transform the current state towards the goal
state.  Research being done with this method includes the development of models
to predict human performance, robot control architecture, etc.  

VI.   Procedure and Methodology.
The first task will be to identify as many goal states as possible.  Then to create
functions to analyze the environment that are of direct use to assessing the 
current state and achieving the goal state.  Then to code what responses the 
NPC should have towards achieving each goal state.  The first testing cycle will
be with controlled variables, i.e. a ship that is stationary; the object will 
be to see if the ship reacts appropriately.  Next will be with a ship moving in
a straight line.  Then with one enemy ship or player at a time.  Followed by 
multiple enemy ships and players, and, finally, with multiple factions and 
players.
                            
VII.    Expected Results & Value to Others
This project should clearly show what goal states cause the player a lot of 
trouble, some trouble, and no trouble at all. This could help future researchers
and/or game designers decide what tactics should be packaged as easy, medium and
hard.  Future researchers could also run tests to see why a certain tactic causes
a player more trouble than another tactic.  The final data will be represented as
several charts, e.g. record, how often reproduced, how often jettisoned, etc. 
This project will in all likelihood take the entire year to complete.