The game description is in this article by Martin Gardner: https://gwern.net/doc/reinforcement-learning/model-free/1991-gardner-ch8amatchboxgamelearningmachine.pdf
The physical implementation is described at Instructables.
From the article linked above:
Hexapawn is played on a 3 x 3 board, with three chess pawns on each side as shown in Figure 43. Dimes and pennies can be used instead of actual chess pieces. Only two types of move are allowed:
-
A pawn may advance straight forward one square to an empty square
-
A pawn may capture an enemy pawn by moving one square diagonally, left or right, to a square occupied by the enemy. The captured piece is removed from the board. These are the same as pawn moves in chess, except that no double move, en passant capture or promotion of pawns is permitted.
The game is won in any of three ways:
-
By advancing a pawn to the third row.
-
By capturing all enemy pieces.
-
By achieving a position in which the enemy cannot move.
Players alternate moves, moving one piece at a time. A draw clearly is impossible, but it is not immediately apparent whether the first or second player has the advantage.