Researchers often encode procedural control knowledge in Hierarchical Task
Networks (HTNs) in order to effectively solve automated planning problems.
Recent advances in heuristic search for domain independent planning,
however, have remained largely unincorporated into HTN planners. On the
flip side, a common technique in plan recognition is to match an observed
agent's behavior against HTN-like structures in order to predict the agents
next actions, and yet incorporating plan recognition into an agent's
planning process is woefully underdeveloped. In this presentation, we start
to close this circle of planning with procedural knowledge. We discuss how
recent theoretical results impact efficient search in both current and
future HTN applications, and how Monte-Carlo sampling techniques can be
used to integrate planning and recognition.