International Journal of Applied Engineering Research ISSN 0973-4562 Volume 9, Number 22 (2014) pp. 11849-11860 © Research India Publications http://www.ripublication.com
Amelioration of Artificial Intelligence using Game Techniques for an Imperfect Information Board Game Geister S. Balakrishnan1*, K.L. Shunmuganathan2, Raja Sreenevasan3 1Research Scholar, Bharath University, Chennai, India;
[email protected] 2Prof & Head, RMK Engineering College, Chennai, India;
[email protected] 3 Project Manager,
[email protected] *Author for correspondence:
Abstract Over the years Artificial Intelligence has made astonishing strides in computer science. Today, computer game is regarded as a new research field of Artificial Intelligence. Here, we use computer game to understand one of the most enigmatic sections of Artificial Intelligence. Game theory is a mathematical study of rational behavior in strategic environment. There are plethora of settings, most notably two-player zero-sum games, game theory provides strong and appealing solution concepts. We have seen the use of these concepts in creation of many game engines for games such as Chess, Tic-Tac-Toe. Various Gaming techniques such as MiniMax, Alpha-Beta pruning, Hash Table and Evaluation are used to create game engine. Artificial Intelligence of game engine can evaluate best moves using deep searching techniques and Probabilistic reasoning can be used to optimize computational capacity of Agent for uncertain and imperfect information environment. Keywords: Agent, Game theory, Geister Game, Minimax.
1. Introduction Game Theory and gamming techniques has always been a useful tool in enhancing the Computational capacity in the field of Artificial Intelligence. In modern time computer game is consider as a new research field and gamming techniques can be used for advancement in technology. Game theory is a mathematical study of rational behavior in strategic environment. There are plethora of settings, most notably two-player zero-sum games,
Paper Code: 27352-IJAER
11850
S. Balakrishnan et al
game theory provides strong and appealing solution concepts. We have seen the use of these concepts in creation of many game engines for games such as Chess, TicTac-Toe. Various Gaming techniques such as MiniMax, Alpha-Beta pruning, Hash Table and Evaluation are used to create game engine. Artificial Intelligence of game engine can evaluate best moves using deep searching techniques but however, the application of these gaming techniques in an imperfect information environment sounds very challenging. In this paper, we implement the various game techniques such as game tree searching and uncertainty prediction techniques in an imperfect information game called Geister. Thereby, we introduce an automated algorithm for this game of imperfect information. We create an autonomous agent that uses the computational Intelligence techniques to play the opponent in a two-player zero-sum game. 1.1 Description Of Geister Game Geister Game is more popularly known as Ghost Game. This game is particularly considered for our study as players do no possesses complete knowledge of the game states (Imperfect Information Game). This game is played by two players and it consist of 6x6 board representation that can hold 16 Ghosts. Each player can have 4 Good Ghost and 4 Bad Ghost and it can be arranged as provided in the Figure 1.
Figure 1. Initial Setup
Both the opponent can see positions of ghost in the board but cannot see the type of ghost their opponent possess. Various strategies can be adopted by different players
Amelioration of Artificial Intelligence using Game Techniques
11851
such as attacking or bluffing. In order to win, a player has to infer the identity of the opponent’s ghosts. This can be achieve by vivid analysis of the players’ behavior. Each player have to place 8 ghost in board. Placement of the good and bad ghost depends on the individual. Information of the ghost is hidden from the opponent player. Ghosts can move in 4 directions (front, back, left and right). If the position of ghost coincides with the opponent’s ghost, the latter is captured by the former. In order to win, different possibilities are available to a player Having all of her/his bad ghosts captured by the opponent player. Capturing all the good ghosts of the opponent player. Moving one of her/his good ghosts off the board from one of the opponent’s corner squares. Prediction function has been used to modify the uncertain and imperfect information environment into certain and perfect information environment. This module has capacity to predict the characteristics of Ghosts by analyzing their behavior. Analysis of characteristics has been measured by using an evilness counter that keep on changing for every ghost based on their movements with respect to the environment. Once prediction is completed, Game Engine starts to analyze the board for an optimum move. In other words, our model aims at implementing bounded rationality to the Agent algorithm, Computational Intelligence has been used such as MiniMax, Alpha-Beta pruning along with Evaluation function to analyze every possible move and to select the winning move for our Agent. The efficiency of the Minimax algorithm has been improved by introducing Rationality component in it. Bounded rationality allow Minimax to consider that every opponent moves is not an optimum moves thus increased the speed of execution. 1.1 Objective The main objective is to introduce an automated algorithm for a game having imperfect information environment by developing an agent which uses Computational Intelligence to automatically play the game.
2. Literature Survey Díez, S.G et.al [11] suggests a mere prolongation of the celebrated MINIMAX algorithm used in zero-sum two-player games, called Rminimax. The Rminimax algorithm allows controlling the intensity of an artificial rival by randomizing its strategy in an optimal manner. In especial, the randomized shortest-path framework is used for biasing the artificial intelligence (AI) adversary toward worse or more beneficial resolutions, thus controlling its strength. In other words, our model aims at introducing/implementing bounded rationality to the MINIMAX algorithm. Equally opposed to other tree-exploration techniques, this new algorithm considers complete paths of a tree (strategies) where a given entropy is spread. The optimal randomized strategy is efficiently computed by means of a simple recurrence relation while keeping the same complexity as the original MINIMAX. As a consequence, the
11852
S. Balakrishnan et al
Rminimax implements a nondeterministic strength-adapted AI opponent for board games in a principled manner, therefore avoiding the assumption of perfect reason. Simulations on two common games show that Rminimax behaves as required. The primary aim is to optimize the efficacy of a Minimax gaming Algorithm for a perfect information games by keeping off the assumption of perfect reason. Suoju He et.al [15] describes about Player Strategy Pattern Recognition (PSPR) is to apply pattern recognition and its approach to designation of the player’s strategic during the game play. Correctly identified player’s strategy, which is called knowledge, could be applied to improve game opponent AI which can be implemented by KB-UCT (knowledge based Upper Confidence bound for Trees). Upper confidence limit for tree has been used here for strategy recognition. KB-UCT improves adaptability of game AI, the challenge level of the gameplay, and the carrying out of the opponent AI; as a result the entertainment of the game is promoted. In this report, the prey and predator game genre of Dead End game is used as a test-bed. During the PSPR, a classification algorithm of KNN (k-nearest neighbor) is taken to analyze off-line data from the simulated gamers who are taking different strategies. Grounded along the information from PSPR, the game AI is promoted through the application of KB-UCT, in this case, domain knowledge is used for UCT tree pruning; as a result the performance of the opponent AI is enhanced. The primary aim of this report is to improve Player Strategy Recognition Pattern using Knowledge based –Upper Confidence Bound for Trees Algorithm to improve the adaptability of the game.
3. System Architecture The gaming techniques are carried out in an imperfect information board game. The Geister game is experienced for its bluffing nature. The basic strategy for the game is suggested. It includes a logics of the game functioning. The gaming techniques involved holds together the gaming engine. Along with game engine, we provide an efficient method for prediction of the characteristics of the ghosts in the game. The minimax algorithm holds the game engine. The power of autonomous Agent to predict the characteristics of the ghost eventually makes the environment certain. Once the environment becomes certain, Minimax algorithm come into gaming and try to opt for best movements that run to the winning nation. .
Amelioration of Artificial Intelligence using Game Techniques
11853
Figure 2. System Architecture
3.1 Description On that point are four mental faculties in our task. They are described as follows Board representation Ghost characteristic prediction Game engine 3.1.1Board representation The board provides a base for Geister Game, it consists of 6x6 Two Dimensional Matrix representation. Board class have the capacity to hold Ghosts objects. It caters proper movements for Ghosts and its orientation.Initial setup of the game state is saved in the Board class and all the ghosts’ position is stored in the Position object in the Ghost class. It revels the characteristics of the ghost that has been eliminated by the latter player. 3.1.2Ghost characteristic prediction This module help to analyze the environment by monitoring Human Ghosts’ movements. By analyzing the opponents movements Autonomous Agent can use Prediction Function to predict the characteristics of Human Players’ Ghost.
11854
S. Balakrishnan et al
An Evilness counter can be provided to count the evilness of opponent’s ghost. Value of evilness counter is based on opponent ghost movements relative to Board and Computer ghosts. Initially we assume that all the Ghost possess by Human are neutral thus Evilness counter is assign to ‘Zero’. Prediction Formula: If ghost moves forward or backward then ghostEvilness=ghostEvilness+(ghostCurrentXposition-ghostXposition)* ghostCurrentXposition. If ghostCurrentYposition=0 or 5 then ghostEvilness=ghostEvilness-(ghostCurrentXposition-ghostXposition)* ghostCurrentXposition*2 Else If ghostYposition
