[Paper Review] Task Interaction in an HTN Planner

Ilce Georgievski, Alexander Lazovik+1 more|arXiv (Cornell University)|Nov 30, 2011

AI-based Problem Solving and Planning11 references3 citations

TL;DR

This paper extends the JSHOP2 HTN planner to natively reason about task interactions where one task's effects are already achieved by prior tasks, enabling automatic detection and pruning of redundant subplans. By integrating 'task-to-task matching' into the planning process, the approach simplifies domain knowledge requirements and improves performance on real-world automation domains like GreenerBuildings, outperforming JSHOP2 in key cases without relying on manually authored phantom tasks.

ABSTRACT

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.

Motivation & Objective

• To address the challenge of managing task interactions in HTN planning, especially in complex real-world domains like building automation.
• To reduce reliance on manually authored phantom tasks in domain knowledge, which are error-prone and cumbersome to write.
• To simplify domain representation by shifting expressivity from domain knowledge to the planner’s reasoning mechanism.
• To improve planning efficiency by detecting and pruning redundant search paths due to already-achieved effects.
• To evaluate the impact of integrating task interaction reasoning directly into the planning process on benchmark domains.

Proposed method

• Introduces a task-to-task matching mechanism that checks whether a task’s effects are already achieved by previously executed tasks.
• Extends JSHOP2 to detect when a task’s preconditions are satisfied due to prior effects, allowing the planner to skip redundant task execution.
• Implements a condition to ignore a task if its effects are still holding and are achieved by another task in the network.
• Integrates this matching logic into the HTN planning loop, enabling dynamic pruning of redundant search paths.
• Uses a formal representation of task effects and preconditions to enable precise matching and effect persistence checks.
• Employs a modified planning algorithm that evaluates task interactions during decomposition, avoiding the need for explicit phantom tasks in the domain model.

Experimental results

Research questions

• RQ1 Can task interaction reasoning be effectively integrated into the HTN planning process to reduce reliance on complex domain-specific constructs like phantom tasks?
• RQ2 How does native task interaction detection affect planning efficiency and solution quality in real-world automation domains?
• RQ3 To what extent can domain knowledge be simplified when the planner handles task interaction reasoning internally?
• RQ4 Does the proposed method outperform JSHOP2 in terms of planning time and solution optimality on benchmark domains?
• RQ5 Can the planner avoid redundant search paths caused by overlapping task effects without increasing computational overhead?

Key findings

• The extended JSHOP2 planner successfully detects and prunes redundant tasks when their effects are already achieved by prior tasks, reducing unnecessary search paths.
• The approach enables simplified domain representations by shifting the burden of interaction reasoning from domain authors to the planner itself.
• The planner outperforms standard JSHOP2 in several benchmark cases, particularly where task interactions would otherwise lead to redundant planning.
• The method reduces the need for manually encoding phantom tasks, which are error-prone and difficult to maintain in complex domains like GreenerBuildings.
• The integration of task-to-task matching improves planning efficiency without sacrificing soundness or completeness, as the planner remains formally correct.
• The approach demonstrates practical benefits in energy-aware building automation, where minimizing redundant actions is critical for optimal energy consumption.