Safe Action Model Learning

SAM learner is an algorithm to efficiently and safely learn action models of STRIPS problems.

This repository contains two variations of the algorithm:

• The classic SAM learning algorithm.
• Lightweight version of ESAM learning algorithm.

The code is also available as a package to download in PyPI.

Requirements:

The repository's code is currently running using Python >= 3.8. In addition, This code is currently depending on the on version 1.3 of pyperplan.

Installation:

From the Python package index (PyPI):

pip install sam-learner

Features:

The repository offers two variants of the SAM learning algorithm, the classical SAM learning algorithm as well as a lightweight version of the Extended SAM learning algorithm.

Classical SAM Learning:

• Learn lifted action model of classical PDDL problems.
• Currently, supports PDDL-1.2 only.
• Export learned domain via domain export functionality.

Extended SAM Learning:

• Learn lifted action model of classical PDDL problems.
• Extends SAM to also learn action models in case that there is an ambiguity in the actions' execution process.
• Currently, supports PDDL-1.2 only.
• Export learned domain via domain export functionality.

New Feature:

Now the code outputs trajectories in a format that FAMA action model learning can read and translate.