EVA-planner

EVA-planner: an EnVironmental Adaptive Gradient-based Local Planner for Quadrotors.

0. Overview

Author: Lun Quan, Zhiwei Zhang, Xingguang Zhong, Chao Xu and Fei Gao from ZJU FAST Lab.

Related Paper: EVA-Planner: Environmental Adaptive Quadrotor Planning, Lun Quan, Zhiwei Zhang, Chao Xu and Fei Gao accepted by 2021 IEEE International Conference on Robotics and Automation (ICRA).

Video Links: Google, Bilibili(for Mainland China)

1. File Structrue

All planning algorithms along with other key modules, such as mapping, are implemented in adaptive_planner
- path_searching: includes multi-layer planner (A*, low-MPC and high-MPCC).
- path_env: includes online mapping algorithms for the planning system (grid map and ESDF(Euclidean signed distance filed)).
- path_manage: High-level modules that schedule and call the mapping and planning algorithms. Interfaces for launching the whole system, as well as the configuration files are contained here

2. 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.

sudo apt-get install libarmadillo-dev

Step 2. We use NLopt to solve the non-linear optimization problem. Please follow the Installation process in NLopt Documentation.

Step 3. Clone the code from github.

git clone https://github.com/ZJU-FAST-Lab/EVA-planner.git

Step 4. Compile.

cd EVA-planner
catkin_make

3. Run a simple example.

Open rviz:

source devel/setup.bash
roslaunch plan_manage rviz.launch

Then, open another terminal and run code:

source devel/setup.bash
roslaunch plan_manage simulation.launch

Then you can enter G with the keyboard and use the mouse to select a target.

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

set(ENABLE_CUDA false)

in the CMakeList.txt in local_sensing packages, to

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

    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_CUDA true, it mimics the depth measured by stereo cameras and renders a depth image by GPU. If ENABLE_CUDA false, 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

Acknowledgements

The framework of this repository is based on Fast-Planner by Zhou Boyu who achieves impressive proformance on quaorotor local planning.
We use NLopt for non-linear optimization.
The hardware architecture is based on an open source implemation from Teach-Repeat-Replan.
The benchmark compared in our paper is ICRA2020_RG_SDDM.

Licence

The source code is released under GPLv3 license.

Maintaince

For any technical issues, please contact Lun Quan (lunquan@zju.edu.cn) or Fei GAO (fgaoaa@zju.edu.cn).

For commercial inquiries, please contact Fei GAO (fgaoaa@zju.edu.cn).