No abstract

The automatic planning community has developed a defacto standard planning language called PDDL. Using the PDDL tools, the reliability of PDDL descriptions can only be posteriori examined. However, the Event-B method supports a rich refinement technique that is mathematically proven. This allows the step-by-step correct construction of Event-B models. In order to specify and solve the planning problems, a development process based on the combination of Event-B and PDDL is proposed. Our development process begins with modeling the planning problem by an Event-B abstract model. Through successive refinements, an Event-B ultimate model correct by construction is obtained. Then, using our Event-B2PDDL Eclipce plugin, the Event-B ultimate model can be automatically translated into a PDDL description. Thus, the resulting PDDL description can be considered correct by construction. Finally, using the PDDL planner tool on this generated PDDL description, plan-solutions related to the planning problems initially described by an Event-B model can be produced. Our process is successfully experimented on a set of representative case studies.

Automatic planning has a de facto standard language called PDDL for describing planning problems. The dynamic analysis tools associated with this language do not allow sufficient verification and validation of PDDL descriptions. Indeed, these tools, namely planners and validators, allow a posteriori error detection. In this paper, we recommend a formal approach coupling the two languages Event-B and PDDL. Event-B supports a formal development process based on the refinement technique with mathematical proofs. Thus, we propose a refinement strategy for obtaining reliable PDDL descriptions from an ultimate Event-B model that is correct by construction. The correctness is guaranteed via the verification and validation tools supported by Event-B. We have chosen the MICONIC application managing modern elevators to illustrate our approach while recognizing that the MICONIC application is already modeled in PDDL without formal proof of its correctness.

No abstract

No abstract