GGTP, a General Gaming Text Protocol

This article describes GGTP, a new general gaming text protocol enabling to play and record plays of many kinds of games. These games range from classic complete information board games such as Chess to incomplete information and chance games such as Bridge. In theory, even real time strategy games could be managed, but theses games currently imply practical limits. 1 Text Protocol for games Many games are played around the world. For most of them, players training and development of competitions require a way to describe precisely specific plays of the games. So many different play systems have been developed, each time with specific data structures for a specific game. In artificial intelligence, games have always been a good domain for the exploration of the performance of decision methods. Many separated researches are and have been done all around the world on such themas and many researchers have to define games engines. One more time, with the same restrictions than for human players, the specialisation in a specific rule. Fewer researches focusses on managing general gaming engines which are designed to be able to play many games. In such problems games engines have to be separated from the rules definitions so they are decomposed in few distinct programs with their own independents goals. To work together and to manage the plays of a selected game rule, these separate components have to use some protocoled communications. These protocoled communications are inherited from the description language used by human players to manage human training and human competitions. Let’s first focus on what are precisely the different components and goals of such a protocol, then we comment on variety of games and describe some existing protocols. 1.1 Requirements of a text protocol for games A play description protocol has to define instructions for each step of a play. First, initialisation of the play. A general gaming protocol has to allow to select the different following elements to prepare a play: • the used rule, • the engines used to play, • any specific initialisation data from the rule (such as the first player to play), • any parameters specific to the rules or the engines selected. Second, and as many times as the rules define it, the protocol has to specify a way to describe the moves selected by the engines. When moves have to be validated by the rule such as in Kriegspiel, Phantom Go, Mah Jong or when multiple players can play at the same turn, moves have to be validated both by the engine and the referee. Third, at the end of the play, the protocol has to manage final information such as the winners, losers, or null players and optionally scoring data for players if the rule specify any as in card games. Possibly, some control commands such as the ability to navigate between steps of the play (undoing or redoing moves, or selecting different moves) would be useful to define too. Perhaps some ability to manage subtrees of plays, to show for example some specific lines of play that a teacher thinks important. Another kind of commands could be considered, those who enable to manage tournaments. 1 Such protocol can be used to follow and record any play of any game. It would be nice to use a text style protocol to allow easy reading of the steps of the play by human analysts. This format would have the advantage to allow humans to interfere with the evolution of the play, with possibly some appropriate commands to override the automatic players choices. Some of these commands are sent by automatic engines, some are sent by the rule, and some are sent by the humans. Obviously, the protocol has to define who can send which kind of command too. 1.2 Variety of games Many examples of games have been studied individually in artificial intelligence researches. We can cite as examples Solitaire [Beasley, 2008], Sudoku [Geem, 2007], Sokoban In a way, a competition is a kind of super rule. So a correctly defined protocol, with an engine able to manage really various games can define a competition as a specific game rule. [Botea et al., 2002], Rush Hour [Fernau et al., 2003], Checkers [Schaeffer et al., 2007; Khan et al., 2008], Hex [Henderson and Hayward, 2008], Chess [Fernández and Salmerón, 2008], Shogi [Soeda et al., 2006], Go [Bouzy and Helmstetter, 2003], Chinese Checkers [Cooper et al., 2004], Backgammon [Azaria and Sipper, 2005], Can’t Stop [Glenn et al., 2008], Bridge [Galatti et al., 2005], Skat [Kupferschmid and Helmert, 2006], Poker [Gilpin and Sandholm, 2007], Phantom Go [Cazenave, 2006], Kriegspiel [Ciancarini and Favini, 2007], Mah Jong [Iida et al., 2004], some wargames [Santana et al., 2004] and some real time strategy games [Ontañón et al., 2007]. These games, except for a part of the wargames and the real time strategy games, are all discrete and finites games. Discrete games are games with no continuity of any kind. Boards must be discrete, alternate of players moves must be discrete, and so on. Finite games are games with no infinity of any kind. Boards must be finite, list of moves must be finite, and so on. Indeed, all of them have different specificities, Here is a short list of critic ones: • Number of players: Some of studied games are one player games. These puzzle games generally have their own solving class of methods. Some other studied games are two player games. These games use different methods from puzzle games because they have to consider having an opponent whose goals are generally in opposition with the player’s goals. The last part are games with more than 2 players. In these games, methods from two players games could be reused but they generally miss some point, because there are specific parameters to deal with. (For example the possibility of alliances between players or the definition of some priority for the players such the choice to be the only one to win or not.) This kind of games generally require special algorithms for these problems and this make them different to program. Despite theses differences, this distinction between games do not denote specific problem for the point of view of a general gaming protocol.