Metadata-Version: 2.4
Name: petrirl
Version: 0.2.1
Summary: PetriRL : Advanced Scheduling with Petri Nets and Reinforcement Learning
Home-page: https://www.fh-swf.de/de/forschung___transfer_4/labore_3/labs/labor_fuer_automatisierungstechnik__soest_1/standardseite_57.php
Author: Sofiene Lassoued
Author-email: lassoued.sofiene@fh-swf.de
Project-URL: Repository, https://github.com/Sofiene-Uni/PetriRL-Framework
Description-Content-Type: text/markdown
Requires-Dist: gymnasium
Requires-Dist: pandas
Requires-Dist: numpy
Dynamic: author
Dynamic: author-email
Dynamic: description
Dynamic: description-content-type
Dynamic: home-page
Dynamic: project-url
Dynamic: requires-dist
Dynamic: summary

# PetriRL: Explainable Reinforcement Learning with Petri Nets  

**PetriRL** is a research framework that integrates **Colored-Timed Petri Nets (CTPNs)** with **Reinforcement Learning (RL)** to solve complex scheduling and optimization problems in **Flexible Manufacturing Systems (FMS)** and beyond.  

Developed as part of my PhD on *Explainable Model-Based Reinforcement Learning for Discrete Event Systems*, PetriRL provides a **formal, modular, and reproducible** approach to dynamic scheduling, intralogistics, and decision-making under uncertainty.  

## Core Idea  

PetriRL bridges **Petri net modeling** and **RL-based decision-making**:  

- **Petri Nets** provide formal semantics, modular scalability, and explainability via token flow and graphical structure.  
- **Reinforcement Learning** provides adaptability, generalization, and efficient online decision-making under dynamic conditions.  
- **Dynamic Action Masking** (from Petri net guard functions) reduces the RL action space, improving sample efficiency and convergence.  

This combination enables **interpretable, scalable, and high-performing RL agents** for scheduling, dispatching, and optimization in discrete event systems.  

---

## PetriRL Environments  

The PetriRL framework is organized into multiple environments, each targeting a different class of problems:  

- 🔹 [**PetriRL Intralogistics**](https://github.com/Sofiene-Uni/PetriRL_Intralogistics)  
  *Dynamic scheduling of AGVs, tool sharing, and shop floor optimization in Flexible Manufacturing Systems.*  

- 🔹 [**PetriRL DJJSP**](https://github.com/Sofiene-Uni/PetriRL_DJSSP)  
  *Dynamic Job Shop Scheduling Problem (DJJSP) with machine breakdowns, variable job arrivals, and stochastic environments.*  

- 🔹 [**PetriRL Hyperheuristics**](#) *(Planned for 2025)*  
  *A super-heuristic framework where RL learns to combine and select dispatching rules for adaptive scheduling.*  

- 🔹 [**PetriRL Multi-Objective Optimization**](#) *(Planned 2026)*  
  *Combines expert agents and actor-critic RL to solve scheduling problems with multiple objectives (e.g., makespan, tardiness, energy consumption).*  

Each environment is **OpenAI Gym-compatible**, comes with **benchmarks** (Taillard + custom large-scale instances), and is designed for **reproducibility and extension**.  

---

## Why PetriRL?  

✔ **Explainable**: Token flows provide a clear semantic interpretation of RL decisions.  
✔ **Scalable**: Modular Petri net design supports complex and large-scale systems.  
✔ **Efficient**: Action masking reduces search space and improves training efficiency.  
✔ **Adaptive**: RL agents generalize to unseen problems and adapt to disruptions in real time.  
✔ **Research-Ready**: Benchmarks, reproducible pipelines, and Gym compatibility.  

---

## Installation  

Install the core PetriRL package via pip:  

```bash
pip install petrirl
```

Specific environments can be created by selecting the corresponding Gym environment ID:

1) Logistics: 
```bash
 env = gym.make("petrirl-djssp-v0")
```
2) Fault tolerant:
```bash
 env = gym.make("petrirl-dft-v0")
```
---

## Research Context  

PetriRL has been applied to:  
- Flexible Manufacturing Systems intralogistics (AGV + tool sharing).  
- Dynamic job shop scheduling with stochasticity.  
- RL-based hyperheuristics for dispatching rule selection.  
- Multi-objective optimization in scheduling.  

These works are part of my PhD and published in international journals:  

- **Flexible Manufacturing Systems intralogistics: Dynamic optimization of AGVs and tool sharing using Colored-Timed Petri Nets and actor–critic RL with actions masking**  
  [ScienceDirect Link](https://www.sciencedirect.com/science/article/pii/S0278612525001694)  

- **Introducing PetriRL: An innovative framework for JSSP resolution integrating Petri nets and event-based reinforcement learning**  
  [ScienceDirect Link](https://www.sciencedirect.com/science/article/pii/S0278612524000943)  

---

## Citation  

If you use this framework, please cite the associated papers:  

```bibtex

@article{
title = {Introducing PetriRL: An innovative framework for JSSP resolution integrating Petri nets and event-based reinforcement learning},
author = {Sofiene Lassoued and Andreas Schwung},
journal = {Journal of Manufacturing Systems},
volume = {74},
pages = {690-702},
year = {2024},
issn = {0278-6125},
doi = {https://doi.org/10.1016/j.jmsy.2024.04.028},
url = {https://www.sciencedirect.com/science/article/pii/S0278612524000943},
}

@article{
title = {Flexible Manufacturing Systems intralogistics: Dynamic optimization of AGVs and tool sharing using Colored-Timed Petri Nets and actor–critic RL with actions masking},
author = {Sofiene Lassoued, Laxmikant Shrikant Baheti, Nathalie Weiß-Borkowski, Stefan Lier, Andreas Schwung}
journal = {Journal of Manufacturing Systems},
volume = {82},
pages = {405-419},
year = {2025},
issn = {0278-6125},
doi = {https://doi.org/10.1016/j.jmsy.2025.06.017.},
url = {https://www.sciencedirect.com/science/article/pii/S0278612525001694},

```

---

## Contributions  

This repository provides the **main entry point** to PetriRL, including:  
- The **core PetriRL package** (`pip install petrirl`).  
- Links to specialized environments.  
- Documentation and research context.  
- Benchmarks and reproducible pipelines for evaluation.  
