Deploying C++ OT model on ROS

[SiqianLi]

Hello, thank you for your great work!

Have you tried deploying the C++ OT model on ROS? I am having a hard time running the OT model with ROS Kinetic in C++. The libtorch breaks the ROS when I try to include it in my package CMakelist.txt. But this "fast_ot.cpp" works in my environment compiled by cmake.

My Environment: Ubuntu 20.04, CUDA 11.1, Pytorch 1.10.0, Libtorch 1.8.0/1.9.0/1.10.0 (all I have tried, but with no success on ROS)

[BIT-MJY]

Hello @SiqianLi,

Thanks for using our code!

Actually we have not implemented OT with ROS. If possible, could you provide the log related to your error here?

[SiqianLi]

Here is my log at catkin_make:

[ 8%] Building CXX object lidar_localization/CMakeFiles/data_pretreat_node.dir/src/mapping/loop_closing/loop_closing_flow.cpp.o
In file included from /home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/data/dataloader_options.h:4,
from /home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/data/dataloader/base.h:3,
from /home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/data/dataloader/stateful.h:3,
from /home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/data/dataloader.h:3,
from /home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/data.h:3,
from /home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/all.h:8,
from /home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/torch.h:3,
from /home/ubuntu/learn/14nn_ws/src/lidar_localization/src/mapping/loop_closing/loop_closing.cpp:17:
/home/ubuntu/Downloads/Tools/libtorch/include/torch/csrc/api/include/torch/types.h:39:16: error: conflicting declaration ‘constexpr const auto torch::kUInt8’
39 | constexpr auto kUInt8 = at::kByte;
| ^~~~~~
In file included from /home/ubuntu/learn/14nn_ws/src/lidar_localization/third_party/libtorch/include/torch/script.h:3,
from /home/ubuntu/learn/14nn_ws/src/lidar_localization/src/mapping/loop_closing/loop_closing.cpp:16:
/home/ubuntu/learn/14nn_ws/src/lidar_localization/third_party/libtorch/include/torch/csrc/api/include/torch/types.h:39:16: note: previous declaration as ‘constexpr const c10::ScalarType torch::kUInt8’
39 | constexpr auto kUInt8 = at::kByte;
| ^~~~~~

Following is my CMakelists.txt:

cmake_minimum_required(VERSION 2.8.3)
project(lidar_localization)

SET(CMAKE_BUILD_TYPE "Release")
#SET(CMAKE_BUILD_TYPE "Debug")
SET(CMAKE_CXX_FLAGS_RELEASE "$ENV{CXXFLAGS} -O3 -Wall")

add_compile_options(-std=c++14)
add_definitions(-std=c++14)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TORCH_CXX_FLAGS}")

set(CMAKE_PREFIX_PATH "/home/ubuntu/Downloads/Tools/libtorch/share/cmake/Torch;/opt/ros/noetic")

set(Torch_DIR /home/ubuntu/Downloads/Tools/libtorch/share/cmake/Torch)
find_package(Torch REQUIRED)
include_directories(/home/ubuntu/Downloads/Tools/libtorch/share/cmake/Torch/include/)
list(APPEND ALL_TARGET_LIBRARIES ${TORCH_LIBRARIES})

add_executable(loop_closing_node src/apps/loop_closing_node.cpp ${ALL_SRCS})
add_dependencies(loop_closing_node ${catkin_EXPORTED_TARGETS} ${PROJECT_NAME}_generate_messages_cpp)
#target_link_libraries(loop_closing_node PUBLIC ${TORCH_LIBRARIES} ${catkin_LIBRARIES} ${ALL_TARGET_LIBRARIES})
target_link_libraries(loop_closing_node PUBLIC ${catkin_LIBRARIES} ${ALL_TARGET_LIBRARIES})

And attached is my running result of ./fast_ot which outputs the similar between frame 000000.bin and 000020.bin.

[smalltheater]

I deployed this work on ROS, It might caused by wrong version of libtorch. You should not use pre cxx 11 , it conflicted to your ROS c++ standard.

[SiqianLi]

Thanks! This works for me!

[Chen-Xieyuanli]

It seems the issue has been solved. I will therefore close it. Please feel free to ask us to reopen it if needed.