A Simulation-Based General Game Player
Yngvi Bj, Hilmar Finnsson · 2009
Abstract—The aim of General Game Playing (GGP) is tocreate intelligent agents that can automatically learn how to playmany different games at an expert level without any humanintervention. The traditional design model for GGP agents hasbeen to use a minimax-based game-tree search augmented withan automatically learned heuristic evaluation function. The firstsuccessful GGP agents all followed that approach. In here wedescribe C ADIA P LAYER , a GGP agent employing a radically dif-ferent approach: instead of a traditional game-tree search it usesMonte-Carlo simulations for its move decisions. Furthermore,we empirically evaluate different simulation-based approacheson a wide variety of games; introduce a domain-independent en-hancement for automatically learning search-control knowledgeto guide the simulation playouts; and show how to adapt thesimulation searches to be more effective in single-agent games.C ADIA P LAYER has already proven its effectiveness by winningthe 2007 and 2008 AAAI GGP competitions.Index Terms—Artificial intelligence, Games, Monte Carlomethods, Search methods.