Plan Recovery in Reactive HTNs Using Symbolic Planning

Building formal models of the world and using them to plan future action is a central problem in artificial intelligence. In this work, we combine two well-known approaches to this problem, namely, reactive hierarchical task networks HTNs and symbolic linear planning. The practical motivation for this hybrid approach was to recover from breakdowns in HTN execution by dynamically invoking symbolic planning. This work also reflects, however, on the deeper issue of tradeoffs between procedural and symbolic modeling. We have implemented our approach in a system that combines a reactive HTN engine, called Disco, with a STRIPS planner implemented in Prolog, and conducted a preliminary evaluation.