Sunteți pe pagina 1din 3

INTELLIGAMMON OVERVIEW

Introduction
In this project, we choose to focus on Backgammon game. This is a complex game that
combines a lot of game states and contains a random dimension. Tactics in Backgammon is
very different and often contradictory. The random and multiple state behaviour of the game
requires a very high skilled human player, and reminds the environment in which we live, a
world that has quite a bit of randomness and multiplicity of situations.

Goals of Projects
We are going to build different agents with different game playing strategies (different
evaluation functions etc.). There are three basic agents that each one will have different
priorities to make a move. There will be a random agent which makes valid moves according
to the dice rolled, an eater agent which plays aggresively and applies attacking strategies
and a gater agent which plays deffensively. At the end of the development phase of the
projects all of these different agents will play againts each other and the success rates are
going to be examined with the reasons behind.

Approach and Method


Backgammon is a nondeterministic game with chance. In nondeterministic games, chance is
introduced by dice or cardshuffling. In case of Backgammon the chance is introduced by
dice. Playing backgammon is simple: there are two players with 15 pieces each. The final
goal is to move all pieces off the board. The rules are: Dice roll determines number of
moves, players move in opposite directions, pieces cannot put on a point occupied by 2 or
more of opponents pieces, single piece can be hit if it is put on the same line a piece of the
opponent and hit piece must start a new route.

ekil 1: Backgammon Board

We are going to perform Minimax algorithm to solve this problem . Minimax is a


generalization of Alpha-Beta search for minimax trees with chance nodes. In case of games
with chance a variation of Minimax algorithm can be used. The baseline algorithm for trees
with chance nodes analogous to Minimax search is the Expectimax algorithm . Just like
Minimax, Expectimax is a full-width search algorithm. It behaves exactly like Minimax except
it adds an extra component for dealing with chance nodes (in addition to Min or Max
nodes).In this algorithm, the outcome depends of a combination of the player's skill and
chance elements such as dice.
There are two players involved, MAX and MIN. A search tree is generated, depth-first,
starting with the current game position upto the end game position. Then, the final game
position is evaluated from MAXs point of view,. Afterwards, the inner node values of the tree
are filled bottom-up with the evaluated values. The nodes that belong to the MAX player
receive the maximun value of its children. The nodes for the MIN player will select the
minimun value of its children.

ekil 2: Game State Tree

Expected Results
In the scope of our project, we are going to build different types of agents and compare their
performances, analyzing the underlying reasons. There will be three type of agents, a
random agent, an eater agent, and a gater agent. Random agent will play the game like a
very begginner humman player, obeying the rules of the game, playing valid but not deeply
wise. Other two agents may choose different moves for a certain snapshot of the game, in
other words their evaluation for the same state can be different. The difference depends on
the different evaluation functions applied while walkingon the search tree. The results will
show which strategy works better on the other one. Obviously random agent is expected to
do the worst against the other agents.

At the end of the projects game playing data will be collected and the results will be
visualized for better interpretation of the data.

References
https://courses.cs.washington.edu/courses/cse573/12au/slides/04-minmax.pdf
http://www.progtools.org/games/tutorials/ai_contest/minmax_contest.pdf

S-ar putea să vă placă și