lunar_lander

Lunar Lander

Lunar lander is a classic rocket trajectory optimization problem, which is part of the Box2D environments of gymnasium. For more detail, please refer to: https://gymnasium.farama.org/environments/box2d/lunar_lander/.

Requirements

1. Installation of libtorch: Follow the instructions on the official PyTorch website: https://pytorch.org/cppdocs/installing.html. In this example, we select the GPU version of libtorch. Please modify the paths in the following part of CMakeList (examples/lunar_lander/CMakeList.txt):

   # libtorch gpu
   set(CUDA_TOOLKIT_ROOT_DIR "path/to/cuda")
   find_package(CUDA REQUIRED)
   list(APPEND CMAKE_PREFIX_PATH "path/to/libtorch")
   find_package(Torch REQUIRED)
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${TORCH_CXX_FLAGS}")
   target_link_libraries(lunar_lander PRIVATE "${TORCH_LIBRARIES}")
   set_property(TARGET lunar_lander PROPERTY CXX_STANDARD 17)
   if (MSVC)
   file(GLOB TORCH_DLLS "${TORCH_INSTALL_PREFIX}/lib/*.dll")
   add_custom_command(TARGET example-app
                       POST_BUILD
                       COMMAND ${CMAKE_COMMAND} -E copy_if_different
                       ${TORCH_DLLS}
                       $<TARGET_FILE_DIR:example-app>)
   endif (MSVC)
   

2. Installation of Gymnasium:

   pip install gymnasium[box2d]
   

3. Installation of pybind11:

   cd path/to/project
   mkdir third_party
   cd third_party
   git clone -b stable https://github.com/pybind/pybind11.git
   

Run

1. Compile

   cd path/to/project
   mkdir build
   cd build
   cmake .. -DBUILD_LUNAR_LANDER=ON
   make
   

2. Run

   Train and evaluate DQN.

   ./examples/lunar_lander/lunar_lander dqn
   

   Train and evaluate PPO.

   ./examples/lunar_lander/lunar_lander ppo