Subtopic Deep Dive
Classical Planning Heuristic Search
Research Guide
What is Classical Planning Heuristic Search?
Classical Planning Heuristic Search develops admissible heuristics like LM-cut, pattern databases, and abstractions to guide optimal state-space search in STRIPS planning problems.
This subtopic focuses on heuristic functions such as downward, h_add, and merge-and-shrink for efficient optimal planning (Helmert 2006). Key systems like Fast Downward integrate these heuristics for PDDL2.2 problems (Helmert 2006, 447 citations). Landmark heuristics unify critical paths and abstractions (Helmert and Domshlak 2009, 393 citations). Over 20 IPC benchmarks evaluate these approaches.
Why It Matters
Heuristic search advances solve IPC domains for automated planning in robotics and space missions (Fox and Long 2003). Fast Downward's heuristics enable scalable optimal planning for real-world PDDL tasks (Helmert 2006). Landmark techniques improve search guidance across delete relaxation, critical paths, and abstraction heuristics (Helmert and Domshlak 2009). SHOP2 demonstrates heuristic efficiency in competition domains (Nau et al. 2003).
Key Research Challenges
Admissibility vs Informativeness
Admissible heuristics guarantee optimality but often provide weak guidance in large state spaces (Helmert 2006). Balancing delete relaxation accuracy with computational cost remains difficult. LM-cut addresses this but scales poorly in IPC domains (Helmert and Domshlak 2009).
Pattern Database Scalability
Precomputing pattern databases for optimal STRIPS planning requires massive memory for complex domains. Compression techniques help but lose admissibility (Helmert 2006). Merge-and-shrink abstractions struggle with high-dimensional problems.
IPC Benchmark Generalization
Heuristics perform well on standard IPC domains but fail to generalize to temporal extensions (Fox and Long 2003). Evaluating across PDDL2.1 temporal domains challenges downward and h_add (Helmert 2006).
Essential Papers
PDDL2.1: An Extension to PDDL for Expressing Temporal Planning Domains
Maria Fox, Derek Long · 2003 · Journal of Artificial Intelligence Research · 1.7K citations
In recent years research in the planning community has moved increasingly toward s application of planners to realistic problems involving both time and many typ es of resources. For example, inter...
Explanation-Based Generalization: A Unifying View
Tom M. Mitchell, Richard M. Keller, Smadar T. Kedar-Cabelli · 1986 · Machine Learning · 1.2K citations
Decision-Theoretic Planning: Structural Assumptions and Computational Leverage
Craig Boutilier, Taraneh Dean, Steve Hanks · 1999 · Journal of Artificial Intelligence Research · 1.1K citations
Planning under uncertainty is a central problem in the study of automated sequential decision making, and has been addressed by researchers in many different fields, including AI planning, decision...
SHOP2: An HTN Planning System
Dana Nau, Tsz-Chiu Au, Okhtay Ilghami et al. · 2003 · Journal of Artificial Intelligence Research · 922 citations
The SHOP2 planning system received one of the awards for distinguished performance in the 2002 International Planning Competition. This paper describes the features of SHOP2 which enabled it to exc...
40 years of cognitive architectures: core cognitive abilities and practical applications
Iuliia Kotseruba, John K. Tsotsos · 2018 · Artificial Intelligence Review · 488 citations
In this paper we present a broad overview of the last 40 years of research on cognitive architectures. To date, the number of existing architectures has reached several hundred, but most of the exi...
The Fast Downward Planning System
M. Helmert · 2006 · Journal of Artificial Intelligence Research · 447 citations
Fast Downward is a classical planning system based on heuristic search. It can deal with general deterministic planning problems encoded in the propositional fragment of PDDL2.2, including advanced...
Landmarks, Critical Paths and Abstractions: What's the Difference Anyway?
Malte Helmert, Carmel Domshlak · 2009 · Proceedings of the International Conference on Automated Planning and Scheduling · 393 citations
Current heuristic estimators for classical domain-independent planning are usually based on one of four ideas: delete relaxations, critical paths, abstractions, and, most recently, landmarks. Previ...
Reading Guide
Foundational Papers
Read Helmert (2006) Fast Downward first for core heuristic framework (downward, h_add) and PDDL2.2 implementation (447 citations). Then Fox and Long (2003) PDDL2.1 for temporal extensions enabling realistic benchmarks (1703 citations).
Recent Advances
Study Helmert and Domshlak (2009) for landmark unification of critical paths and abstractions (393 citations). SHOP2 (Nau et al. 2003) shows heuristic application in IPC competition (922 citations).
Core Methods
Core techniques: delete relaxation (downward, h_add), pattern databases, LM-cut, merge-and-shrink abstractions, landmark heuristics for optimal search guidance.
How PapersFlow Helps You Research Classical Planning Heuristic Search
Discover & Search
Research Agent uses searchPapers to find Helmert (2006) Fast Downward, then citationGraph reveals 447 downstream works on LM-cut heuristics, and findSimilarPapers identifies landmark unification papers like Helmert and Domshlak (2009). exaSearch queries 'admissible heuristics IPC optimal planning' to surface 50+ IPC-evaluated approaches.
Analyze & Verify
Analysis Agent applies readPaperContent to extract heuristic formulas from Helmert (2006), then verifyResponse with CoVe checks admissibility proofs against PDDL2.2 semantics. runPythonAnalysis simulates h_add vs downward on IPC domains using NumPy, with GRADE scoring evidence strength for optimality guarantees.
Synthesize & Write
Synthesis Agent detects gaps in landmark coverage for temporal PDDL (Fox and Long 2003), flags contradictions between h_add and merge-and-shrink. Writing Agent uses latexEditText for heuristic pseudocode, latexSyncCitations for IPC benchmarks, latexCompile for planning diagrams, and exportMermaid for state-space search graphs.
Use Cases
"Benchmark h_add vs LM-cut on IPC-2006 domains"
Research Agent → searchPapers('IPC heuristics') → Analysis Agent → runPythonAnalysis(NumPy simulation of heuristics) → researcher gets CSV of expansion counts and optimality ratios.
"Compare Fast Downward heuristics to SHOP2"
Research Agent → citationGraph(Helmert 2006) → Synthesis Agent → gap detection → Writing Agent → latexEditText + latexSyncCitations + latexCompile → researcher gets compiled LaTeX report with benchmark tables.
"Find GitHub repos implementing pattern databases"
Research Agent → searchPapers('pattern databases planning') → Code Discovery → paperExtractUrls → paperFindGithubRepo → githubRepoInspect → researcher gets verified code implementations with README analysis.
Automated Workflows
Deep Research workflow scans 50+ papers via searchPapers on 'classical planning heuristics IPC', structures report with Helmert (2006) as core, outputs GRADE-scored review. DeepScan applies 7-step analysis: readPaperContent(Fox and Long 2003) → CoVe verification → Python heuristic simulation. Theorizer generates new admissible heuristic hypotheses from landmark patterns in Helmert and Domshlak (2009).
Frequently Asked Questions
What defines Classical Planning Heuristic Search?
Development of admissible heuristics like LM-cut, pattern databases, and abstractions for optimal STRIPS state-space search, benchmarked on IPC domains using Fast Downward (Helmert 2006).
What are core heuristic methods?
Downward, h_add, merge-and-shrink, and landmarks unify delete relaxations, critical paths, and abstractions (Helmert 2006; Helmert and Domshlak 2009).
What are key papers?
Fast Downward (Helmert 2006, 447 citations) introduces explicit heuristics; Landmarks paper unifies approaches (Helmert and Domshlak 2009, 393 citations); PDDL2.1 enables temporal benchmarks (Fox and Long 2003, 1703 citations).
What are open problems?
Scaling pattern databases to high-dimensional IPC domains without losing admissibility; generalizing landmarks to PDDL2.1 temporal planning (Fox and Long 2003; Helmert 2006).
Research AI-based Problem Solving and Planning with AI
PapersFlow provides specialized AI tools for Computer Science researchers. Here are the most relevant for this topic:
AI Literature Review
Automate paper discovery and synthesis across 474M+ papers
Code & Data Discovery
Find datasets, code repositories, and computational tools
Deep Research Reports
Multi-source evidence synthesis with counter-evidence
AI Academic Writing
Write research papers with AI assistance and LaTeX support
See how researchers in Computer Science & AI use PapersFlow
Field-specific workflows, example queries, and use cases.
Start Researching Classical Planning Heuristic Search with AI
Search 474M+ papers, run AI-powered literature reviews, and write with integrated citations — all in one workspace.
See how PapersFlow works for Computer Science researchers