Reasoning under Uncertainty in Darkchess

Abstract

Two player strategy games with partially observable environments face challenges regarding reasoning under uncertainty. The aim of this thesis is to investigate darkchess with focus on searching and evaluating actions based on partial knowledge. This has been approached by risk assessing threats within the unobservable part of the environment. A working darkchess agent has been developed, where multiple tests between different agents has been conducted, as well as a user test. The results, based on statistical analysis, indicate that a modified alpha-beta search algorithm with risk assessment and a simplified evaluation function approach semi-decent playing strength.