The game is played on a board of squares, where each square is a floor or a wall. Some floor squares contain boxes, and some floor squares are marked as storage locations. The player is confined to the board and may move horizontally or vertically onto empty squares. The player can move a box by walking up to it and pushing it to the square beyond. Boxes cannot be pulled and cannot be pushed to squares with walls or other boxes. The number of boxes equals the number of storage locations. The puzzle is solved when all boxes are placed at storage locations.
Sokoban was a hit in Japan and had sold over 400,000 units in that country by the time Spectrum HoloByte imported it to the United States. A review in Computer Gaming World praised the game for being "pure and simple, very playable and mentally challenging", citing its addictive qualities. It was also reviewed in Dragon, receiving out of 5 stars.
Sokoban can be studied using the theory ofcomputational complexity. The problem of solving Sokoban puzzles was first proved to be NP-hard. Further work showed that it was significantly more difficult than NP problems; it is PSPACE-complete. This is of interest for artificial intelligence research because solving Sokoban can be compared to the automated planning required by some autonomous robots. Sokoban is difficult not only because of its large branching factor, but also because of its large search tree depth. Some level types can even be extended indefinitely, with each iteration requiring an exponentially growing number of moves and pushes. Skilled human players rely mostly on heuristics and are usually able to quickly discard a great many futile or redundant lines of play by recognizing patterns and subgoals, thereby drastically reducing the amount of search. Some Sokoban puzzles can be solved automatically by using a single-agent search algorithm, such as IDA*; enhanced by several techniques that make use of domain-specific knowledge. This is the method used by Rolling Stone, a Sokoban solver developed by the University of Alberta GAMES Group. However, the more complex Sokoban levels are out of reach even for the best automated solvers.
Variants
Several puzzles can be considered variants of the original Sokoban game in the sense that they all make use of a controllable character pushing boxes around in a maze.
Alternative tilings: In the standard game, the mazes are laid out on a square grid. Several variants apply the rules of Sokoban to mazes laid out on other tilings. Hexoban uses regular hexagons, and Trioban uses equilateral triangles.
Multiple pushers: In the variants Multiban and Interlock, the player can control multiple characters.
Alternative goals: Several variants adjust the requirements for completing a level. For example, in Block-o-Mania the boxes have different colours, and the goal is to push them onto squares with matching colours. Sokomind Plus implements a similar idea, with boxes and target squares uniquely numbered. In Interlock and Sokolor, the boxes also have different colours, but the goal is to move them so that similarly coloured boxes are adjacent. In CyberBox, each level has a designated exit square, and the goal is to reach that exit. In a variant called Beanstalk, the elements of the level must be pushed onto a target square in a fixed sequence.
Additional game elements: Push Crate, Sokonex, Xsok, Cyberbox and Block-o-Mania all add new elements to the basic puzzle. Examples include holes, teleports, moving blocks and one-way passages. The 1982 Sokoban game featured levels with destructible walls.
Character actions: In Pukoban, the character can pull boxes in addition to pushing them.
Reverse mode: The player solves the puzzle backwards, from the end to the initial position by pulling instead of pushing boxes. Standard Sokoban puzzles can be played in reverse mode, and the reverse-mode solutions can be converted to solutions for the standard-mode puzzles. Therefore, reverse-mode gameplay can also be instrumental in solving standard Sokoban puzzles.
