A model-based action representation for multiagent planning systems
Mary Ellen Weisskopf, James D. Johannes · 1989
The Task Network is an extended knowledge representation for hierarchical, general purpose planning systems. It has explicit features which facilitate the scheduling of parallel actions in multiagent domains and the detection of conflicts between plan steps, two areas where planners have been consistently weak. The formalism is based on the natural division of domain activity into separate tasks. The research leading up to the development of the Task Network consisted of several steps. Initially, the notion of a task was formally defined and an analysis was performed which resulted in a taxonomy of intertask relations. The relations were defined and their properties studied. The analysis led to the construction of a Task Activity Model (the TAM), which captures the salient features of a task. A Domain Description Language (the DDL) was developed for describing domain objects and task attributes. The DDL was used to define a representation for individual tasks and for the connecting links that are used to build the Task Network. Finally, it was shown how the Task Network formalism could be used to implement dependency sets, meta-task definitions, and new plan formulation techniques of task breaking and task overlapping which are important in multiagent domains.