FACT-Fast-and-Active-Coordinate-Initialization

Open source code for our paper FACT: Fast and Active Coordinate Initialization for Vision-based Drone Swarm.

2024/12 Our paper has been accepted by RAL!!!

IEEE Xplore | Preprint on arxiv

Introduction

This repository contains the code for our paper FACT: Fast and Active Coordinate Initialization for Vision-based Drone Swarm. We provide the code for the simulation. The code base is still under development for easier deployment and use. We will keep updating the code base and provide more detailed instructions.

The simulation and trajectory planning part of the code is based on our previous work Ego-Planner.

If you use our code in your own research, please cite as

@ARTICLE{10803028,
  author={Li, Yuan and Zhao, Anke and Wang, Yingjian and Xu, Ziyi and Zhou, Xin and Xu, Chao and Zhou, Jinni and Gao, Fei},
  journal={IEEE Robotics and Automation Letters}, 
  title={FACT: Fast and Active Coordinate Initialization for Vision-Based Drone Swarms}, 
  year={2025},
  volume={10},
  number={2},
  pages={931-938},
  keywords={Robots;Robot kinematics;Drones;Estimation;Pose estimation;Planning;Collision avoidance;Optimization;Sensors;Global navigation satellite system;Aerial systems;perception and autonomy;swarm robotics;vision-based navigation},
  doi={10.1109/LRA.2024.3518101}}

Prerequisites

We developed the code with Ubuntu 20.04 and ROS Noetic. It might work with older distributions, but we recommend using Ubuntu 20.04 and ROS Noetic for simplicity. We recommend using computers with NVIDIA GPUs with CUDA installed. The code has been tested on two laptops with NVIDIA GTX 1660 Ti-MaxQ and RTX 3070.

Mosek

We use Mosek v10.1 to solve the optimization problem. You can get a free academic license from Mosek. More information on license.

After you get the license, download Mosek v10.1 from here(v10.1.22 and v10.1.23 have been tested) and extract the 'mosek' folder to your home directory. Then put the 'mosek.lic' file in the 'mosek' folder. To install Mosek, open a terminal and navigate to the 'mosek' folder and run the following command:

cd 10.1/tools/platform/linux64x86/src/fusion_cxx/ # the command may vary depending on the version and your platform
make install -j

Detailed instructions of installing Mosek can be found here here.

ROS

We use ROS Noetic for the simulation. You can install ROS Noetic by following the instructions here. We recommend installing the full-desktop version.

Installation

Clone the repository to your workspace and build the code with catkin_make:

git clone https://github.com/ZJU-FAST-Lab/FACT-Coordinate-Initialization.git
cd FACT-Coordinate-Initialization
catkin_make -DMOSEK_DIR=$YOUR_HOME_DIR/mosek/10.1/tools/platform/linux64x86 # same directory as the one you installed Mosek

Most of the possible missing dependencies can be installed with apt-get. If you encounter any problems, please feel free to open an issue or contact us.

Usage

We provide a launch file to run the simulation. You can run the simulation with the following command:

source $YOUR_WORKSPACE/devel/setup.bash
roslaunch ego_planner swarmtakeoff.launch

By default, the simulation will run with 6 drones using the SDP optimization method. If you want to test local optimization methods mentioned in the paper, you can change the launch instruction to:

roslaunch ego_planner swarmtakeoff.launch solver_type:=LM

for Levenberg-Marquardt method, or

roslaunch ego_planner swarmtakeoff.launch solver_type:=GaussNewton

for Gauss-Newton method.

The terminal will print the published goal positions and the calculated observations for all drones at each iteration. If enough observations are received, the terminal will print the solved relative rotations. The simulation will stop when all relative poses are solved.

By default, the results will be saved to the 'results' folder in the workspace directory. To evaluate the results, you can run the script 'process_data.py' in the 'results' folder:

cd $YOUR_WORKSPACE/results
python3 process_data.py

The script will print the MAE of relative rotations and the time used for each method.

If you want to visualize the simulation, you can run the following command in another terminal before or after running the simulation command above:

source $YOUR_WORKSPACE/devel/setup.bash
roslaunch ego_planner rviz.launch

Known Issues

Certain situations may cause the planning or renderer to fail, printing huge amounts of error messages.

Contact Us

Feel free to open an issue if you have any questions or suggestions. If you prefer to contact us directly, you can send an email to Anke Zhao at 1203302828@qq.com for deployment problems, or Yuan Li at yuanli_cse@zju.edu.cn for algorithm problems.