A Reinforcement Learning agent that learns to sail the Vendée Globe, the solo non-stop round-the-world race, using real ERA5 wind data.
Project for the Artificial Intelligence course, MSc in Computer Science, University of Bologna (2026). Grade: 30 cum laude.
🌐 Project page: https://alessandrocampedelli.github.io/VendeeGlobeRL/
📖 Full documentation: docs/README.md
📄 Report: report.pdf
- PPO agent (Stable-Baselines3) trained in a custom Gymnasium environment
- Real ERA5 wind data from ECMWF
- Trained on the 2016–17 and 2020–21 editions, tested on the unseen 2024–25 edition
| File | Content |
|---|---|
VendeeGlobe_RL_ERA5.ipynb |
Notebook with environment, training and simulation (run it on Google Colab) |
Vendee_RL_ERA5_weights.zip |
Pre-trained model weights |
report.pdf |
Project report |
docs/ |
Interactive episode viewer and full documentation |
Alessandro Campedelli, Matteo Boscherini, Nicolas Cola