Lecture 9: Classical Planning and Planning Algorithms

Lecture Overview

This lecture introduces classical planning - the problem of finding sequences of actions to achieve goals. We'll explore how to represent planning problems using STRIPS and PDDL, and study different algorithms for solving them: forward search, backward search, and SAT-based planning.

Key Concepts: STRIPS, PDDL, State-Space Search, SAT Planning

Applications: Robotics, Logistics, Scheduling, Resource Allocation

📚 Planning Cheat Sheet Available!

Quick reference guide covering all key concepts, formulas, algorithms, and PDDL syntax. Perfect for studying and quick lookups!

View Cheat Sheet

Main Topics

1. Introduction to Classical Planning

Understanding the planning problem: finding sequences of actions to achieve goals. Key assumptions of classical planning and connection to search algorithms.

Planning Problem Deterministic Fully Observable Static Environment

2. State Representation

How to represent states using ground atomic fluents. Understanding database semantics, Closed-World Assumption (CWA), and Unique Names Assumption (UNA).

Ground Fluents CWA UNA Database Semantics

3. STRIPS Action Representation

Classic representation for actions: Preconditions, Add List, and Delete List. Understanding state transitions and the Result function.

STRIPS Preconditions Add/Delete Lists Result Function

4. PDDL (Planning Domain Definition Language)

Modern standard for representing planning problems. Syntax for domain and problem files, types, predicates, actions, initial states, and goals.

PDDL Syntax Domain Files Problem Files Types & Predicates

5. Planning Problems and Examples

Classic planning problems: Spare Tire Problem and Saudi Air Cargo Transport. Understanding problem formulation and sequential reasoning.

Spare Tire Air Cargo Problem Formulation Solutions

6. Forward State-Space Search

Progression from initial state to goal. Understanding applicable actions, grounding, unification, and heuristics for efficient search.

Progression Applicable Actions Grounding Heuristics

7. Backward (Regression) Search

Working backward from the goal to initial state. Relevant actions, regression formulas, and goal-directed planning.

Regression Relevant Actions Goal-Directed Lower Branching

8. Planning as Boolean Satisfiability (SAT)

Converting planning problems to Boolean formulas. Propositionalization, exclusion axioms, precondition axioms, and successor-state axioms.

SAT Encoding Propositionalization Axioms SAT Solvers

9. Python Programming for Automated Planning

From PDDL representation to automated reasoning with Python. Using pddlpy library and building planners from scratch for practical implementation.

PDDL Files pddlpy Library Python Implementation Automated Reasoning

10. Heuristics & Comparing Planning Approaches

Domain-independent heuristics and trade-offs between planning algorithms. Relaxed planning, completeness, optimality, and practical considerations.

Heuristics Relaxed Planning Completeness Trade-offs

Interactive Demonstrations

Visualize and interact with planning algorithms and PDDL in real-time.

PDDL Syntax Interactive Demo

Forward Search Visualizer

Backward Search Visualizer

SAT-Based Planning Demo

STRIPS Action Execution Demo

Air Cargo Planning Demo

Practical Exercises

Work through these exercises to master planning concepts and algorithms.

PDDL Problem Formulation

Learn to formulate planning problems in PDDL syntax with domain and problem files.

Forward Search Exercise

Practice forward state-space search with applicable actions and state transitions.

Backward Search Exercise

Master regression and goal-directed planning with relevant actions.

SAT-Based Planning Exercise

Convert planning problems to SAT formulas and solve with Boolean satisfiability.

Key Concepts Summary

Classical Planning:

Deterministic, fully observable environment
Discrete states and actions
Goal: sequence of actions to reach goal state

STRIPS & PDDL:

STRIPS: Precond, Add List, Delete List
PDDL: Modern standard with types
Result(s, a) = (s - Del(a)) ∪ Add(a)

Planning Algorithms:

Forward: Progression from initial state
Backward: Regression from goal
SAT: Convert to Boolean satisfiability

Key Properties:

Completeness: finds solution if exists
Optimality: finds shortest plan
Efficiency: heuristics reduce search

SE444: Artificial Intelligence | Lecture 9: Classical Planning and Planning Algorithms