ZJU-FAST-Lab/ego-planner-swarm: An efficient single/multi-agent trajectory planner for multicopters. (github.com)
Quick Start within 3 Minutes
Compiling tests passed on ubuntu 16.04, 18.04, and 20.04 with ros installed. You can just execute the following commands one by one.
1234567 sudo apt-get install libarmadillo-devgit clone https://github.com/ZJU-FAST-Lab/ego-planner-swarm.gitcd ego-planner-swarmcatkin_make -j1source devel/setup.bashroslaunch ego_planner simple_run.launch
If you find this work useful or interesting, please kindly give us a star ⭐, thanks!😀
- This work extends EGO-Planner to swarm navigation.
EGO-Swarm: A Fully Autonomous and Decentralized Quadrotor Swarm System in Cluttered Environments
EGO-Swarm is a decentralized and asynchronous systematic solution for multi-robot autonomous navigation in unknown obstacle-rich scenes using merely onboard resources.
Video Links: YouTube, bilibili (for Mainland China)
1. Related Paper
EGO-Swarm: A Fully Autonomous and Decentralized Quadrotor Swarm System in Cluttered Environments, Xin Zhou, Jiangchao Zhu, Hongyu Zhou, Chao Xu, and Fei Gao (Published in ICRA2021). Paper link and Science report.
2. Standard Compilation
Requirements: ubuntu 16.04, 18.04 or 20.04 with ros-desktop-full installation.
Step 1. Install Armadillo, which is required by uav_simulator.
12 sudo apt-get install libarmadillo-dev
Step 2. Clone the code from github or gitee. These two repositories synchronize automatically.
12 git clone https://github.com/ZJU-FAST-Lab/ego-planner-swarm.git
Step 3. Compile,
123 cd ego-plannercatkin_make -DCMAKE_BUILD_TYPE=Release -j1
Step 4. Run.
In a terminal at the ego-planner-swarm/ folder, open the rviz for visualization and interactions
123 source devel/setup.bashroslaunch ego_planner rviz.launch
In another terminal at the ego-planner-swarm/, run the planner in simulation by
123 source devel/setup.bashroslaunch ego_planner swarm.launch
Then you can follow the gif below to control the drone.
3. Using an IDE
We recommend using vscode, the project file has been included in the code you have cloned, which is the .vscode folder. This folder is hidden by default. Follow the steps below to configure the IDE for auto code completion & jump. It will take 3 minutes.
Step 1. Install C and CMake extentions in vscode.
Step 2. Re-compile the code using the command
12 catkin_make -DCMAKE_BUILD_TYPE=Release -DCMAKE_EXPORT_COMPILE_COMMANDS=Yes
It will export a compile commands file, which can help vscode to determine the code architecture.
Step 3. Launch vscode and select the ego-planner folder to open.
12 code ~/<......>/ego-planner-swarm/
Press Ctrl Shift B in vscode to compile the code. This command is defined in .vscode/tasks.json. You can add customized arguments after “args”. The default is “-DCMAKE_BUILD_TYPE=Release”.
Step 4. Close and re-launch vscode, you will see the vscode has already understood the code architecture and can perform auto completion & jump.
4. Use GPU or Not
Packages in this repo, local_sensing have GPU, CPU two different versions. By default, they are in CPU version for better compatibility. By changing
12 set(ENABLE_CUDA false)
in the CMakeList.txt in local_sensing packages, to
12 set(ENABLE_CUDA true)
CUDA will be turned-on to generate depth images as a real depth camera does.
Please remember to also change the ‘arch’ and ‘code’ flags in the line of
1234 set(CUDA_NVCC_FLAGS-gencode arch=compute_61,code=sm_61;)
in CMakeList.txt, if you encounter compiling error due to different Nvidia graphics card you use. You can check the right code here.
Don’t forget to re-compile the code!
local_sensing is the simulated sensors. If
ENABLE_CUDAtrue, it mimics the depth measured by stereo cameras and renders a depth image by GPU. If
ENABLE_CUDAfalse, it will publish pointclouds with no ray-casting. Our local mapping module automatically selects whether depth images or pointclouds as its input.
For installation of CUDA, please go to CUDA ToolKit
5. Use Drone Simulation Considering Dynamics or Not
Typical simulations use a dynamic model to calculate the motion of the drone under given commands. However, it requires continuous iterations to solve a differential equation, which consumes quite a lot computation. When launching a swarm of drones, this computation burden may cause significant lag. On an i7 9700KF CPU I use, 15 drones are the upper limit. Therefore, for compatibility and scalability purposes, I use a “fake_drone” package to convert commands to drone odometry directly by default.
If you want to use a more realistic quadrotor model, you can un-comment the node
so3_control/SO3ControlNodeletin simulator.xml to enable quadrotor simulation considering dynamics. Please don’t forget to comment the package
poscmd_2_odomright after the above two nodes.
6. Utilize the Full Performance of CPU
The computation time of our planner is too short for the OS to increase CPU frequency, which makes the computation time tend to be longer and unstable.
Therefore, we recommend you to manually set the CPU frequency to the maximum. Firstly, install a tool by
12 sudo apt install cpufrequtils
Then you can set the CPU frequency to the maximum allowed by
12 sudo cpufreq-set -g performance
More information can be found in http://www.thinkwiki.org/wiki/How_to_use_cpufrequtils.
Note that CPU frequency may still decrease due to high temperature in high load.
The source code is released under GPLv3 license.
We are still working on extending the proposed system and improving code reliability.
For any technical issues, please contact Xin Zhou (email@example.com) or Fei GAO (firstname.lastname@example.org).
For commercial inquiries, please contact Fei GAO (email@example.com).
他们发表的论文 [2008.08835] EGO-Planner: An ESDF-free Gradient-based Local Planner for Quadrotors (arxiv.org)，全文如下：