PoseidonRL

A Gymnasium environment for maritime vessel navigation and path planning.

Overview

PoseidonRL aim to formulates maritime navigation as a reinforcement learning problem. An agent controls a vessel moving ocean grid, and must learn to reach a goal while avoiding land and responding to sea state. The environment exposes a standard Gymnasium API, so it can be trained with any RL library that supports Gymnasium.

Usage

Download the data

Download the bathymetry, grid, and monthly wave forcing files:

python download_maritime_data.py --year 2025

Play manually

Control the boat yourself with the keyboard:

python free_play.py ./data 

Train an agent

Train a PPO agent with Stable-Baselines3:

python train.py  --data-dir ./data/ 

Watch a trained agent

Run a saved model and watch it navigate:

python play_agent.py   runs/boatworld_ppo/best_model/best_model.zip  ./data  

License

This project is licensed under the MIT License.

Status

This project is still in progress to improve performance.

Citation

If you use PoseidonRL in your work, please cite it as:

@software{poseidonrl,
  title  = {PoseidonRL: A Gymnasium Environment for Maritime Vessel Navigation and Path Planning},
  author = {Kardous Youssef},
  year   = {2026},
  url    = {https://github.com/kardSIM/PoseidonRL}
}