General Game Playing is a field of artificial intelligence that seeks to create programs capable of playing any game at an expert-level without the need for human aid. There are two major approaches to general game playing: simulation and heuristic. I focused on the move selection component of a common simulation strategy called Monte Carlo Tree Search. Traditionally, the selection step of Monte Carlo Tree Search uses an algorithm called Upper Confidence Bound Applied to Trees or UCT. In place of this algorithm, I investigated the applicability of a random roulette wheel style of selection. I studied the effectiveness of this roulette wheel style selection using tic-tac-toe and nim. The game player built from Roulette Wheel selection performed well against its opponents. It demonstrated the strengths of a flexible planning strategy throughout these games.
展开▼
