This site contains links and offers some downloads of articles about Stratego programming. Sharing information by providing references and downloads is the only purpose of this site. As a consequence it has been kept very simple and straightforward.

About Stratego

Stratego is one of the ultimate challenges for programmers of artificial intelligence in board games. It is very difficult to let the computer play a good game of Stratego. The best current Stratego programs play on the level of a moderate advanced human Stratego player. If you like to win computer games then do not worry too much about the possibility of a humilating loss. If you are a moderate advanced Stratego player then probably you will win most battles against any Stratego program.

The challenge

Why is it so difficult to make a good playing Stratego program? The most important reasons are:

This classifies Stratego as a game of imperfect-information. That category of games is more complex than games where all information is directly available. But there is more to it. Conventional algorithms like alfa-beta move tree search, forward pruning and quiescence search work well in games like chess but do not give look ahead enough to evaluate tactical issues in Stratego.

Research has been done to overcome these barriers by other methods than move tree search:

but also these alternatives have lead to not more than a mediocre playing strength (for references see the Links tab).

Domain-dependent knowledge

There is lack of literature about good ways to play Stratego. In available articles game knowledge only has been described globally and fragmentary. Most academic research has been spent to universally applicable game algorithms where game knowledge does not play a significant role. Will programs play better if they use structured and detailed game knowledge? An answer to this question requires that at first thorough game knowledge should be made available.

Sharing game knowledge

There is a great source of game knowledge available: the Gravon database. Game knowledge can be extracted by analysis of positions and moves and can be made applicable for use by programs. But this will require much effort and time. The publication of such studies is necessary to prevent duplication of work and effort in this area and to stimulate the creation of real strong playing programs. This is the main reason to publish articles and the results of studies on this site. Everyone may use the contents of the research results on this site without any restriction.