I wonder - how it would be to write game logic in bare Erlang. From one side it can be tricky, because of immutability. Changing state - is always copying. And in any game state can be changed very quickly.
It turns out, that Erlang pattern-matching facility can greatly simplify game logic stuff. Sometimes code is much better and more clear than in Ruby.
Sokoban was quite pleasure to write. It's most essential part is boiled down to this:
% player can't step on wall
speculate(wall, _) -> [];
% player can go on empty
speculate(empty, _) -> [{player,empty}];
% player can go onto hole
speculate(hole, _) -> [{player,hole}];
% box can't move to wall
speculate({box,_}, wall) -> [];
% two boxes can't move
speculate({box,_}, {box,_}) -> [];
% box can move on any base
speculate({box,B2}, B3) when is_atom(B3) -> [{player,B2}, {box,B3}].
To know, what to do, you need to know about state of two next cells after character. Getting all combinations it can be fully described with simple pattern-matching rules.
For me, it's resembled description of rules in Prolog(which isn't surprise).
Take a look at test/sokoban_game_tests.erl to get an idea, how to run this
app.