Monte Carlo Simulation im Skat
A lot of AI research about the card game of Bridge was done in the last years. The similarities between the card games Bridge and Skat arose the idea to adopt and enhance the results for AI Bridge players on AI Skat players.
The use of Monte Carlo simulations for finding a play strategy was investigated in the bachelor thesis. In contrast to previous works on this field the incomplete information about the opponent cards was maintained. It was investigated whether keeping the information incomplete leads to better game playing or not. In games with complete information the best strategies could intermix which could result in a suboptimal strategy.
It could be shown that an AI Skat player which keeps the incomplete information about the opponent cards in the simulated worlds during the Monte Carlo simulations plays at the same level as an AI player with complete information. The approach of simulating a possible world and playing the game in this world with skat heuristics was significantly faster than the searching for the best strategy in a game with complete information. This fact leaves space for further improvements or more simulations.
Unfortunately the thesis is written in German and will certainly not be translated into English in the future. For all intrested people the thesis can be downloaded as PDF file: