Agentic Design

Patterns
🏗️

Hierarchical Task Network (HTN) Planning(HTN)

Automated planning approach that decomposes complex tasks into hierarchically structured networks of simpler tasks using domain knowledge

Complexity: highPlanning

🎯 30-Second Overview

Pattern: Hierarchical decomposition of complex tasks into manageable sub-tasks using domain knowledge

Why: Enables efficient planning through abstraction, reusable expertise, and human-interpretable solutions

Key Insight: Tasks (compound→primitive) + Methods (how to decompose) + Conditions (when to apply) = Hierarchical Plans

⚡ Quick Implementation

1Define Domain:Tasks (compound/primitive), methods, conditions
2Create Methods:Decomposition rules: when & how to break tasks
3Build Network:Hierarchical task structure with constraints
4Plan & Execute:Decompose top-level goal → primitive actions
5Monitor & Adapt:Replan if conditions change or methods fail
Example: analyze_market → [gather_data, process_info, generate_report] → primitive_actions

📋 Do's & Don'ts

✅Encode domain expertise in methods and conditions
✅Use clear task hierarchies with proper abstraction levels
✅Define preconditions and effects for each method
✅Implement backtracking when decomposition fails
✅Cache successful decomposition patterns
✅Design reusable method libraries for domains
❌Create overly deep hierarchies (>5-6 levels)
❌Ignore method ordering constraints and dependencies
❌Use HTN for simple linear task sequences
❌Hard-code solutions without conditional methods
❌Skip validation of method preconditions

🚦 When to Use

Use When

  • • Complex multi-step workflows
  • • Domain expertise can be encoded
  • • Hierarchical task structure exists
  • • Multiple solution approaches possible
  • • Need human-interpretable plans
  • • Reusable planning components

Avoid When

  • • Simple linear task sequences
  • • Unknown or rapidly changing domains
  • • Real-time reactive behaviors
  • • No clear task hierarchy
  • • Purely data-driven approaches needed
  • • Single-shot optimization problems

📊 Key Metrics

Plan Success Rate
% plans that execute successfully
Decomposition Depth
Average levels in task hierarchy
Planning Time
Time to generate executable plan
Method Reuse Rate
% methods reused across plans
Backtrack Frequency
% plans requiring backtracking
Domain Coverage
% domain tasks plannable
Plan Optimality
Quality vs optimal solution
Execution Efficiency
Primitive actions per goal achieved

💡 Top Use Cases

Autonomous Research Agents: Literature review → data extraction → analysis → synthesis
Software Development: Requirements → design → implementation → testing → deployment
Manufacturing: Product planning → resource allocation → production → quality control
Game AI: Strategic goals → tactical objectives → unit actions → movement primitives
Robotics: Navigate → path planning → motion control → actuator commands
Business Process: Strategic planning → resource allocation → task execution → monitoring
Military Planning: Mission objectives → tactical plans → unit assignments → actions
Educational Systems: Learning goals → curriculum design → lesson planning → activities

References & Further Reading

Deepen your understanding with these curated resources

Academic Papers

An Overview of Hierarchical Task Network Planning (ArXiv 2014)

HTN planning: Overview, comparison, and beyond (ScienceDirect 2015)

Hierarchical Task Network Planning for Facilitating Cooperative Multi-Agent Reinforcement Learning (ArXiv 2023)

A Structural Complexity Analysis of Hierarchical Task Network Planning (ArXiv 2024)

Implementation Guides

Hierarchical Task Network (HTN) Planning in AI - GeeksforGeeks

An Introduction to Hierarchical Task Network Planning (University of Ulm)

Exploring HTN Planners through Example (Game AI Pro)

HTN Planning Tutorial (Michigan Tech)

Tools & Implementations

GPT-HTN-Planner - LLM-based HTN Planning

SHOP2 - HTN Planner (University of Maryland)

PYHOP - Python HTN Planning Library

Fluid HTN - Unity Game Engine Implementation

Research Centers & Communities

International Conference on Automated Planning and Scheduling (ICAPS)

University of Maryland AI Planning Research

AIPS Planning Competition

Stack Overflow - HTN Planning Discussions

Contribute to this collection

Know a great resource? Submit a pull request to add it.

Contribute

Patterns

closed

Loading...

Built by Kortexya