Resilient UAV Swarm Communications with Graph Convolutional Neural Network

This repository contains the source codes of

Resilient UAV Swarm Communications with Graph Convolutional Neural Network

Zhiyu Mou, Feifei Gao, Jun Liu, and Qihui Wu


Problem Descriptions

In this paper, we study the self-healing of communication connectivity (SCC) problem of unmanned aerial vehicle (UAV) swarm network (USNET) that is required to quickly rebuild the communication connectivity under unpredictable external destructions (UEDs). Firstly, to cope with the one-off UEDs, we propose a graph convolutional neural network (GCN) and find the recovery topology of the USNET in an on-line manner. Secondly, to cope with general UEDs, we develop a GCN based trajectory planning algorithm that can make UAVs rebuild the communication connectivity during the self-healing process. We also design a meta learning scheme to facilitate the on-line executions of the GCN. Numerical results show that the proposed algorithms can rebuild the communication connectivity of the USNET more quickly than the existing algorithms under both one-off UEDs and general UEDs. The simulation results also show that the meta learning scheme can not only enhance the performance of the GCN but also reduce the time complexity of the on-line executions.

Display of Main Results Demo

One-off UEDs

randomly destruct 150 UAVs                             randomly destruct 100 UAVs

150 100

General UEDs

general UEDs with global information           general UEDs with monitoring mechanism

general_global_info general

Note: these are gifs. It may take a few seconds to display. You can refresh the page if they cannot display normally. Or you can view them in ./video.

Environment Requirements

cuda supports and GPU acceleration

Note: other versions of the required packages may also work.

The machine we use

CPU: Intel(R) Core(TM) i7-10700K CPU @ 3.80GHz
GPU: NVIDIA GeForce RTX 3090

Necessary Supplementary Downloads

As some of the necessary configuration files, including .xlsx and .npy files can not be uploaded to the github, we upload these files to the clouds. Anyone trying to run these codes need to download the necessary files.

Download initial UAV positions (necessary)

To make the codes reproducible, you need to download the initial positions of UAVs we used in the experiment from or Upzip the download files to ./Configurations/.

Download Trained Meta Parameters (alternative, but if using meta learning without training again, then necessary)

Since the total size of meta parameters is about 1.2GB, we have uploaded the meta parameters to and You need to download the file from either two links and unzip them to ./Meta_Learning_Results/meta_parameters/if you want to use the trained meta parameters. Otherwise, you need to train the meta parameters again (directly run

Download Meta Learning Loss Functions Pictures (alternative)

The loss function pictures of meta learning are available on and You can store them in ./Meta_Learning_Results/meta_loss_pic/

Quick Start

Simulate SCC under one-off UEDs

directly run ./


Simulate meta learning process

directly run ./


Simulate SCC under general UEDs

directly run ./


File and Directory Explanations

  • ./Configurations/

the initial positions of 200 UAVs

  • ./Drawing/

the drawing functions

  • ./Experiment_Fig/

the experiment figures and the drawing source codes

  • ./Main_algorithm_GCN/

the proposed algorithms in the paper

  • ./Main_algorithm_GCN/

the CR-MGC algorithm (Algorithm 2 in the paper)

  • ./Main_algorithm_GCN/

the GCO algorithm

  • ./Main_algorithm_GCN/

algorithm of finding the smallest distance to make the RUAV graph a CCN (Algorithm 1 in the paper)

  • ./Meta_Learning_Results/

the results of meta learning

  • ./Meta_Learning_Results/meta_loss_pic

the loss function pictures of 199 mGCNs

  • ./Meta_Learning_Results/meta_parameters

the meta parameters (Since the total size of meta parameters is about 1.2GB, we have uploaded the meta parameters to or

  • ./Traditional_Algorithm/

the implementations of traditional algorithms

  • ./video/

the gif files of one-off UEDs

  • ./

the simulation parameters

  • ./

the Environment generating UEDs

  • ./

the simulation under general UEDs

  • ./

the simulation under one-off UEDs

  • ./

draw the Fig. 12(d) in the simulation under one-off UEDs

  • ./

the meta learning

  • ./

the integration of algorithms under one-off UEDs

  • ./

the integration of algorithms under general UEDs

  • ./

the utility functions

Note that some unnecessary drawing codes used in the paper are not uploaded to this responsitory.


View Github