Analytical Study of the A* Heuristic Search Algorithm Used to Solve Efficiently a Puzzle Game

Abstract

The puzzle game presented in this paper consists in polyhedra (prisms, pyramids or pyramidal frustums) which can be moved using the free available spaces. The problem requires to be found the minimum number of movements in order the game reaches to a goal configuration starting from an initial configuration. Because the problem is enough complex, the principal difficulty in solving it is given by dimension of search space, that leads to necessity of a heuristic search. The improving of the search method consists into determination of a strong estimation by the heuristic function which will guide the search process to the most promising side of the search tree. The comparative study is realized among Hamming heuristic, Manhattan heuristic, and the Chebyshev heuristic using A* search algorithm implemented in Java. This paper also presents the necessary stages in object oriented development of a software used to solve efficiently this puzzle game. The modelling of the software is achieved through specific UML diagrams representing the phases of analysis, design and implementation, the system thus being described in a clear and practical manner. With the purpose to confirm the theoretical results which demonstrates that Manhattan heuristic is more efficient was used space complexity criterion. The space complexity was measured by the number of generated nodes from the search tree, by the number of the expanded nodes and by the effective branching factor. From the experimental results obtained by using the Chebyshev heuristic, improvements were observed regarding space complexity of A* algorithm versus Hamming and Manhattan heuristics.