Planning in Adversity: a Computational Model of Strategic Planning in the Game of Go.

摘要

A general procedure for representing plans in games such as go and chess is presented. This representation is simlar to the types of plans generated by robot problem solving programs, in that it uses a goal tree with the actions to be taken represented in the terminal nodes. It is shown how the information returned from both tactical and strategic lookaheads can be represented using this procedure. A program for doing strategic lookahead in go is described. The program determines the strategic side effects that will occur as a result of trying to achieve a strategic goal by only examining a few of the many move sequences that will achieve that goal. It then generalizes the results to the untested branches.