Autonomous Drone Swarm Optimisation

This repository reproduces the results from the Master Thesis: "Optimising Group Alignment and Cohesion in Autonomous Drone Swarms for Targeted Military Exploration." The research focuses on training reinforcement learning (RL) agents to manage the alignment and cohesion of drone swarms during military exploration missions.

Repository Structure

The repository contains the following files and folders:

Files

• DronesEnv.py: Defines the custom environment for the single-agent RL agents, where drones move on an LxL grid, aiming to explore specific regions while maintaining group cohesion and alignment.

• train.py: Trains the single-agent RL model using the PPO algorithm from stable_baselines3. Models are saved in the models/ directory, with the best model during training saved in log_dir_model/.

• evaluate.py: Evaluates the trained models, outputting metrics such as the order parameter, Mean Squared Displacement (MSD), and episode rewards. It also generates plots required to reproduce the thesis results.

• utils.py: Contains utility functions for plotting and other helper functions used across the project.

• CallbackClass.py: Implements a custom callback to track and save training progress.

Folders

• plots/: Stores the plots generated during evaluation.
• models/: Contains the trained models.
• log_dir_model/: Stores logs and the best model checkpoint during training.

Dependencies

To run the scripts in this repository, the following Python packages are required:

• stable_baselines3
• numpy
• matplotlib
• jax

You can install these packages using pip:

pip install stable-baselines3 numpy matplotlib jax

Usage

Training

To train the RL agents, run the train.py script. This will:

• Initialize the drone environment defined in DronesEnv.py.
• Train the model using PPO.
• Save the trained model to the models/ directory.
• Periodically save the best model to log_dir_model/.

You can customize parameters such as the number of drones (N) and the swarm_factor and other settings directly in train.py.

Evaluation

To evaluate the trained models, run the evaluate.py script. This will:

• Load the trained models from models/.
• Evaluate the performance using metrics like the order parameter, MSD, and episode rewards.
• Generate and save plots in the plots/ directory, which are essential for reproducing the results in the thesis.

Please make sure to align the settings used to train the model in the evaluate.py file in order to make sure to evaluate the correct model.

Reproducing Thesis Results

Follow these steps to reproduce the results from the thesis:

1. Adjust the environment and model parameters in train.py as needed.
2. Train the model by running train.py.
3. Evaluate the trained model by running evaluate.py.
4. The plots and metrics necessary for the thesis will be available in the plots/ directory.

Contact

For any questions or issues, please contact lindsayspoor (repo maintainer) on GitHub.