Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HTN Plan Repair Algorithms Compared: Strengths and Weaknesses of Different Methods

Paul Zaidins, Robert P. Goldman, Ugur Kuter, Dana Nau, Mark Roberts

TL;DR

The paper addresses dynamic disruptions in hierarchical HTN planning and compares three plan-repair methods: Rewrite (RW), SHOPFixer (SF), and IPyHOPPER (IPH). It provides a formal framework for plan repair, showing that the algorithms differ in their search spaces and the repairs they permit, with RW being more restrictive, SF leveraging causal links, and IPH relying on external simulation. The authors prove relationships among the solution sets of the three methods and empirically evaluate them on IPC-derived domains (Satellite, Rovers, Openstacks), revealing domain-dependent performance: SF and IPH often outperform RW, while RW can solve a distinct subset of problems. The study highlights how stability, domain structure, and repair definitions influence practical outcomes, and it discusses opportunities for optimization and future research in HTN repair and planning under uncertainty.

Abstract

This paper provides theoretical and empirical comparisons of three recent hierarchical plan repair algorithms: SHOPFixer, IPyHOPPER, and Rewrite. Our theoretical results show that the three algorithms correspond to three different definitions of the plan repair problem, leading to differences in the algorithms' search spaces, the repair problems they can solve, and the kinds of repairs they can make. Understanding these distinctions is important when choosing a repair method for any given application. Building on the theoretical results, we evaluate the algorithms empirically in a series of benchmark planning problems. Our empirical results provide more detailed insight into the runtime repair performance of these systems and the coverage of the repair problems solved, based on algorithmic properties such as replanning, chronological backtracking, and backjumping over plan trees.

HTN Plan Repair Algorithms Compared: Strengths and Weaknesses of Different Methods

TL;DR

The paper addresses dynamic disruptions in hierarchical HTN planning and compares three plan-repair methods: Rewrite (RW), SHOPFixer (SF), and IPyHOPPER (IPH). It provides a formal framework for plan repair, showing that the algorithms differ in their search spaces and the repairs they permit, with RW being more restrictive, SF leveraging causal links, and IPH relying on external simulation. The authors prove relationships among the solution sets of the three methods and empirically evaluate them on IPC-derived domains (Satellite, Rovers, Openstacks), revealing domain-dependent performance: SF and IPH often outperform RW, while RW can solve a distinct subset of problems. The study highlights how stability, domain structure, and repair definitions influence practical outcomes, and it discusses opportunities for optimization and future research in HTN repair and planning under uncertainty.

Abstract

This paper provides theoretical and empirical comparisons of three recent hierarchical plan repair algorithms: SHOPFixer, IPyHOPPER, and Rewrite. Our theoretical results show that the three algorithms correspond to three different definitions of the plan repair problem, leading to differences in the algorithms' search spaces, the repair problems they can solve, and the kinds of repairs they can make. Understanding these distinctions is important when choosing a repair method for any given application. Building on the theoretical results, we evaluate the algorithms empirically in a series of benchmark planning problems. Our empirical results provide more detailed insight into the runtime repair performance of these systems and the coverage of the repair problems solved, based on algorithmic properties such as replanning, chronological backtracking, and backjumping over plan trees.

Paper Structure

This paper contains 40 sections, 7 theorems, 20 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. HTN Plan Repair Algorithms
  4. Rewrite (RW): Compile a New Problem
  5. RW algorithm implementation
  6. SHOPFixer (SF): Examine Causal Links
  7. IPyHOPPER (IPH): Repair via Simulation
  8. Theoretical Results
  9. Preliminary Definitions
  10. Plan Repair Definitions
  11. Class 1: normal execution.
  12. Class 2: anomaly repair.
  13. Class 3: predicted-task-failure repair.
  14. Class 4: predicted-action-failure repair.
  15. Theorems
  16. ...and 25 more sections

Key Result

Theorem 1

In every HTN plan-repair problem, the set of Class 3 solutions is the set of SF solutions.

Figures (20)

  • Figure 1: States and tasks relevant for repair of $T$.
  • Figure 2: Venn diagram displaying the relationships among the sets of solutions each algorithm can produce. Note that $\{\text{RW\xspace solutions}\} \cap \{\text{IPH\xspace solutions}\} \, \subseteq \, \{\text{SF\xspace solutions}\}$.
  • Figure 3: Example demonstrating that there are classes of problems that RW can solve but SF and IPH cannot, and vice versa.
  • Figure 4: Success rates for the Rovers repair problems for each of the three algorithms.
  • Figure 5: Each algorithm's runtimes in $\mathrm{msec}$ (semi-log plot) on the Rovers repair problems that the algorithm solved successfully.
  • ...and 15 more figures

Theorems & Definitions (7)

  • Theorem 1
  • Theorem 2
  • Theorem 3
  • Theorem 4
  • Theorem 5
  • Corollary 1
  • Theorem 6