| Automated planning is an important research branch with a long history in the field of Artificial Intelligence.Temporal planning is a sub-area of automated planning which expands the time dimension on the basis of Classical Planning.Probabilistic parallel planning is a new frontier sub-field of automated planning,which has the characteristics of"uncertain effect"and"parallel execution".Planning domain definition language(PDDL),the modeling language for current temporal planning is difficult to express concurrent actions,its effect depends on specific actions but does not have global constraints and effects.Relational dynamic influence diagram language(RDDL),the modeling language for probabilistic parallel planning does not support the expression of time.No method have been proposed to solve the problem that actions have duration,uncertainly effects and parallel execution.This paper study probabilistic parallel temporal planning in a partially-domain-specific way,with a simulation-based framework called KarmaRan ~T(Knowledge based Represent and computational-graph based searching planner with Temporal features)aiming at using automated planning to solve problems closer to real life applications.For KarmaRan ~T,we first propose a new domain knowledge representation,which describes the domain knowledge of planning problem in JSON format.Then,we parse and code the JSON file to generate the point set file,edge set file and the corresponding constant,value function in the calculation graph.Finally,we write code under the Pregel the graph calculation model of Spark platform to solve the problem. |
