As organisers of the Third International Planning Competition we chose to address several new challenges that we believe are important to the ambitions of the planning community: the management of problems with metric constraints on numerically valued variables, temporal planning (including managing concurrency and scheduling of activities) and the construction of plans subject to specified optimisation criteria other than a simple count of the numbers of steps.
Setting these goals has obvious implications for the potential competitors in terms of extended expressive power and additional problem-solving required to manage the extensions. In order to control the extent to which competitors would be required to handle all of these extensions successfully, we constructed variants, or levels, of all of the domains used in the competition. For most of the domains we included a STRIPS level, a NUMERIC level (using STRIPS and metric variables only), a SIMPLETIME level in which actions have duration but there are no other metric values in the domain and a full temporal level, TIME, using durative actions with durations determined by the context of their usage (so duration depends on the parameter values and not only on the action name). The SIMPLETIME and TIME levels did not involve numeric resources other than time. To address this combination we introduced some additional domain variants, as we discuss below. Each of the four levels corresponds to a particular degree of expressive power of PDDL2.1, and there are different challenges posed by the versions of each domain at each level.
A secondary goal was to assess the relative effort of generating and encoding control rules for planners. Unfortunately, we failed to find any way to usefully quantify this effort. We discuss this question further in the following sections.