A Numeric PDDL Representation of the Good Snowman Planning Problem

The single-agent puzzle game Good Snowman aims to navigate a confined, maze-like environment with the specific goal of constructing snowmen by stacking three snowballs in descending order of size, and can be naturally characterised as a planning problem, where it is asked to find a sequence of actions such as rolling balls on snow, and stacking or unstacking them, so that balls of the appropriate sizes are joined into snowmen. A solution to the planning problem has been provided via classical planning; however, the solution does not scale properly. Motivated by the fact that parts of the problem can be conveniently represented using numeric predicates/fluents, in this paper we present a numeric PDDL representation of the Good Snowman planning problem. Results of running ENHSP as planning engine show that the numeric representation leads to computational advantages over the classical formulation. Further, we define, implement and test a domain heuristic tailored for this problem, that leads to further improvements.