Reverse Planning & Epistemic Reasoning

Project Overview

This research project explores the complex field of Reverse Planning—the ability of an agent to backtrack or “undo” actions to return to a previous valid state.

To make the abstract logic accessible, I modeled a unique problem domain: The preparation of the Sicilian Cannolo 🍩🇮🇹. The goal was to mathematically model the states of assembly (e.g., Empty Shell → Filled → Decorated) and analyze which actions are reversible under specific logic frameworks.

Methodologies Compared

The core objective was to benchmark different Automated Planning languages and solvers to handle Universally Reversible Actions.

1. PDDL (STRIPS) & Plasp

• Approach: Classical planning definition using the Planning Domain Definition Language.
• Workflow: Defined the domain in .pddl files and used Plasp to automatically translate these definitions into Answer Set Programming facts.

2. ASP & Epistemic Reasoning (Eclingo)

• Approach: Moving beyond simple state transitions to model knowledge states.
• Tool: Eclingo.
• Why: To solve problems where the agent must know that an action is reversible, not just that it is physically possible.

3. K Language & DLV^K

• Approach: Using a high-level declarative planning language.
• Tool: DLV^K System.
• Result: Successfully modeled planning queries (e.g., “Find a plan of length 3 that results in a filled Cannolo”).

Technical Execution

The repository contains multiple implementations of the same domain to demonstrate the syntax and solving capabilities of each tool:

• Sequential Horizon: Solving using standard Clingo atoms.
• Epistemic Logic: Solving using Eclingo with knowledge operators ($K$ operator).
• QBF Solving: Mapped to Quantified Boolean Formulas using qasp.jar.

Documentation

The project includes a detailed scientific report (Report.pdf) analyzing the transition graphs and the computational performance of the different solvers (Clingo vs. DLV).

Tools used: Clingo, Eclingo, Plasp, DLV System.