Automated Planning: Landmarks And Heuristics.pdf

Automated Planning - Landmarks

This section explores landmarks, a technique for approximating optimal heuristics in automated planning.

Key Concepts:

Landmark: An action that must be part of every solution to a planning problem.
Landmark Cost: Estimated solution cost based on the number of unachieved landmarks.
Delete Relaxation: A technique where delete operations (removing actions from plans) are ignored during heuristic computation, leading to more accurate heuristics.

Process:

1. Define Landmarks: For a STRIPS planning task, we first compute its delete-free version (removing delete effects). Then, landmarks are identified within this relaxed task.

2. Landmark Heuristic: The goal is to find an accurate approximation of the optimal "delete relaxed" heuristic. This is achieved by using algorithms like Optimal Cost Partitioning (Karpas & Domshlak, 2009) to compute a hitting set - a minimal subset of actions covering all possible plan alternatives.

3. Admissibility: The proposed landmark heuristics are proven admissible, meaning they never overestimate the cost required to reach a goal.

Benefits:

Provides a more accurate heuristic than simple cost summation or maximum cost.
* Approximation algorithms exist for efficient computation (e.g., Optimal Cost Partitioning).