MarsExplorer-Gymnasium

MarsExplorer-Gymnasium is a Gymnasium-compatible robotic exploration environment for autonomous terrain coverage in unknown, procedurally generated Mars-like landscapes.

It bridges modern Deep Reinforcement Learning methods with exploration and coverage planning under uncertainty.

For full details and evaluation results, see the companion paper: MarsExplorer: Exploration of Unknown Terrains via Deep Reinforcement Learning and Procedurally Generated Environments

---

🚀 Achieved Results with PPO-based RL Agent

MarsExplorer-Gymnasium has been evaluated using PPO-based agents, demonstrating effective exploration behavior and strong coverage performance across diverse terrain configurations.

---

🌍 Strong Generalization Through Procedural Terrain Generation

Terrain diversification is a core attribute of MarsExplorer-Gymnasium.

Each episode is generated through an automated procedural pipeline with configurable variation levels, including:

• Number, size, and placement of obstacles
• Terrain scalability (map size)
• Required exploration coverage threshold
• Bonus rewards for successful completion

This diversity enforces robust generalization — a critical property for real-world robotic deployment where unseen terrain configurations are inevitable.

---

Installation

Quick Start

Install via pip

pip install git+https://github.com/markub3327/MarsExplorer@main

---

Usage Example (Gymnasium API)

import gymnasium as gym
import mars_explorer

env = gym.make("explorer-v1")

obs, info = env.reset()

done = False
while not done:
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    done = terminated or truncated

env.close()

MarsExplorer-Gymnasium follows the standard Gymnasium API, making it directly compatible with modern reinforcement learning libraries.

---

Citation

If you use MarsExplorer-Gymnasium in your research, please cite:

@article{Koutras2021MarsExplorer,
  title={MarsExplorer: Exploration of Unknown Terrains via Deep Reinforcement Learning and Procedurally Generated Environments},
  author={Dimitrios I. Koutras and A. C. Kapoutsis and A. Amanatiadis and E. B. Kosmatopoulos},
  journal={Electronics},
  volume={10},
  number={22},
  pages={2751},
  year={2021},
  doi={10.3390/electronics10222751},
  url={https://www.mdpi.com/2079-9292/10/22/2751}
}