Abstrakt
Many modern planning systems are exploiting advantages of multi-valued state variables. The vast majority of these planners is based on the translation to the SAS+ encoding that is done at the level of grounded actions.
It seems contraproductive to first model the problem in a STRIPS-like formalism and then to do reverse engineering to re-discover the state-variables. It would be more natural to directly describe the problem as a synchronized evolution of state variables.
Real-life planning applications are already exploiting this timeline-based approach but there does not exist a generally agreed formal framework for modeling problems using timelines. This paper proposes a modeling framework for describing planning problems using operations changing the values of multi-valued state variables.
The key innovation of this modeling framework in comparison to existing timeline-based models is keeping the description of the planning domain separated from the particular planning problem similarly to PDDL. The formal framework is accompanied by a prototype editor that allows formulation of planning domains and problems (and supports their translation back to PDDL).