Left: Global Coverage Path Planning, Right: Quality-driven Local Path Planning
- ROS Noetic (Ubuntu 20.04)
- NVIDIA RTX 3070Ti (Single GPU)
- CUDA 11.6
- cuDNN 8.9.0
- LibTorch 1.12.1-cu116
- PCL 1.7
- Eigen3
- gcc9
- [Pre-requisites] Make sure 50GB space in your disk.
- Install Unreal Engine
git clone -b 4.25 [email protected]:EpicGames/UnrealEngine.git
cd UnrealEngine
./Setup.sh
./GenerateProjectFiles.sh
make
- Install AirSim
git clone https://github.com/Microsoft/AirSim.git
cd AirSim
./setup.sh
./build.sh
Download the setting.json and move it to ~/Documents/AirSim/settings.json.
- Install cuDNN and LibTorch
https://developer.nvidia.com/rdp/cudnn-download
https://download.pytorch.org/libtorch/cu116/libtorch-cxx11-abi-shared-with-deps-1.12.0%2Bcu116.zip
You should change Torch_DIR in all CMakeLists.txt as your own LibTorch path, e.g. /home/albert/3rdparty/libtorch/share/cmake/Torch.
- Download environments in AirSim and pre-trained model of SPM in C++ version
Palace Env Village Env (Note: merge village_house_1.zip and village_house_2.zip into village house environment.) spm.pt
Note: Put trained model into address using your own file address.
- Complie the planner
cd ${YOUR_WORKSPACE_PATH}
catkin_make -DCMAKE_CUDA_COMPILER=/usr/local/cuda/bin/nvcc
- Change prediction model path
You should change surf_pred/model_ in src/predrecon/exploration_manager/launch/algorithm.xml as your own spm checkpoint path.