Structured Modeling Language for Representing Active Template Libraries (Causal Modeling)

摘要

In this report we give a high-level description of the computational approach for the Causal Modeler (CModeler) tool. The tool provides a capability for capturing the cause/effect constraints in a Special Operations plan and for reasoning tasks in support of plan execution. The input to the tool is a plan created by a human planner in a mixed initiative environment using custom graphical interface (a program called SOFTools TPE) and the output is a minimal directed acyclic graph (DAG) representing a parsimonious potential causality graph. The nodes of the DAG are the actions and the directed arcs represent potential causal links. The term 'potential' emphasizes the uncertainty in the abduced links since no requirement is placed on the availability of domain theory. Our approach relies on the structural information only; namely the temporal ordering of the actions and the task hierarchy of the plan. We describe one main application of the tool for the Special Operations domain to support the task of run-time replanning. The replanning task takes the unexpected events in the execution of the plan (e.g., late or aborted actions) and uses the causal model to compute the impact on future actions and reconfigure the plan. We summarize at the end of the report our views of the lessons learned and give concluding remarks about future directions for developing this technology.