Task Interaction in an HTN Planner

Hierarchical Task Network (HTN) planning uses task decomposition to plan for an executable sequence of actions as a solution to a problem. In order to reason effectively, an HTN planner needs expressive domain knowledge. For instance, a simplified HTN planning system such as JSHOP2 uses such expressivity and avoids some task interactions due to the increased complexity of the planning process. We address the possibility of simplifying the domain representation needed for an HTN planner to find good solutions, especially in real-world domains describing home and building automation environments. We extend the JSHOP2 planner to reason about task interaction that happens when task’s effects are already achieved by other tasks. The planner then prunes some of the redundant searches that can occur due to the planning process’s interleaving nature. We evaluate the original and our improved planner on two benchmark domains. We show that our planner behaves better by using simplified domain knowledge and outperforms JSHOP2 in a number of relevant cases.

💡 Research Summary

The paper addresses a well‑known limitation of hierarchical task network (HTN) planners such as JSHOP2: the need for domain engineers to explicitly encode “phantom tasks” that represent already‑achieved effects. In real‑world domains—particularly large‑scale home and building automation scenarios—authoring these phantom tasks is tedious, error‑prone, and often infeasible. To alleviate this burden, the authors extend JSHOP2 with a lightweight reasoning mechanism that automatically detects when a primitive task’s effects are already satisfied by previously executed tasks and, if those effects still hold, prunes the redundant task from the plan.

The core of the extension is an “agenda” data structure that accumulates all logical atoms (facts) produced by applied primitive actions. When the interleaving step selects a new primitive task, the algorithm checks whether an identical operator instance has already been applied and whether its effects are present in the agenda. If both conditions are true, the task is deemed “matchable” and is omitted from further consideration. For non‑primitive tasks (methods), the planner records whether a method with the same name has been instantiated earlier; if so, its subtasks are examined recursively, and any branch whose effects are already satisfied is marked “reducible” and excluded from subsequent search. Algorithms 1 and 2 in the paper integrate these checks into the existing JSHOP2 workflow with minimal disruption, essentially adding a matchability test before each primitive execution and a reducibility test before expanding a method’s branch.

The authors argue that this approach yields two major benefits. First, it simplifies domain modeling: designers no longer need to anticipate every possible interaction and write phantom tasks manually, which is especially valuable when domain expertise is limited. Second, it shrinks the search space by eliminating redundant interleavings, leading to faster planning and shorter plans. The paper validates the approach on two benchmark domains. The primary case study is the GreenerBuildings project, a realistic energy‑aware building automation environment involving roughly a thousand devices and hundreds of operators. In this setting, the original JSHOP2 either fails to find a plan or produces plans with many unnecessary actions unless extensive phantom task specifications are added. The enhanced planner, using only the simplified domain description, automatically discards duplicate actions (e.g., turning on a light that is already on) and generates a compact, energy‑optimal schedule. A second, more traditional HTN benchmark further demonstrates reduced planning time and plan length compared with the baseline.

Limitations are acknowledged. The current matching criterion requires exact operator identity and identical parameters; thus, tasks that achieve the same effect with different arguments are not recognized as redundant. Moreover, the agenda assumes that effects persist indefinitely, which may be invalid for temporary or consumable effects, potentially leading to incorrect pruning. Future work is suggested to incorporate richer effect‑duration models and to broaden the matching logic to handle semantic equivalence beyond syntactic identity.

In summary, the paper contributes a practical enhancement to HTN planning that shifts the responsibility for handling task interactions from the domain author to the planner itself. By integrating a simple agenda‑based redundancy check into JSHOP2’s interleaving mechanism, the authors achieve both a reduction in domain modeling effort and measurable performance gains in realistic automation domains. The experimental results substantiate the claim that the extended planner can produce better or equal quality plans while operating on considerably simpler domain specifications.