AE4350: Final Project

Mult-Agent Reinforcement Learning For Drone Flocking

Name: Nikhil Sethi

Student Number: 5711428

Date: 11/07/23

This repository contains code for the final project completed for the course Bio-inspired Learning for Aerospace Applications. My aim was to use MADDPG to train 4 drones to flock together in a confined environment.

Results

Video at test time

Learning curve:

Prerequisites

The repository was tested with the following versions. Even though they are fairly old, it is highly recommended that you create a virtual environment and use the same versions because they include dependencies from the authors of the main MADDPG repository.

• Ubuntu 18.04
• Python 3.5.10
• Tensorflow 1.8.0
• gym 0.10.5
• matplotlib 3.0.3
• protobuf 3.19.6

Setup

git clone git@github.com:nikhil-sethi/marl_flocker.git
cd marl_flocker
git submodule update --init --recursive
cd maddpg/
pip3 install -e .
cd ..
cd multiagent-particle-envs
pip3 install -e .
cd ..

Training

cp scenarios/flocking.py multiagent-particle-envs/scenarios/

cd maddpg/experiments
python train.py --scenario flocking --num-episodes 20000 --save-rate 100 --save-dir <path/to/this/repo>/results/policy

Testing

python train.py --restore --display --scenario flocking --load-dir <path/to/this/repo>/results/policy

Plots

To reproduce the plots from the paper:

pip install seaborn
python stat.py