• Setup to train a local planner with reinforcement learning approaches from stable baselines integrated ROS
• Training in a simulator fusion of Flatland and pedsim_ros
• local planner has been trained on static and dynamic obstacles: video

• standart ROS setup (Code has been tested with ROS-kinetic on Ubuntu 16.04)

sudo apt install screen

sudo apt install virtualenv

1. Setup repository:

   • Clone this repository in your src-folder of your catkin workspace

   cd <path_to_catkin_ws>/src/drl_local_planner_ros_stable_baselines
   cp .rosinstall ../
   cd ..
   rosws update
   cd <path_to_catkin_ws>
   catkin_make -DCMAKE_BUILD_TYPE=Release
   

   (please install required ros packages)

2. Setup virtual environment to be able to use python3 with ros

   virtualenv <path_to_venv>/venv_p3 --python=python3
   source <path_to_venv>/venv_p3/bin/activate
   <path_to_venv>/venv_p3/bin/pip install \
           pyyaml \
           rospkg \
           catkin_pkg \
           exception \
           numpy \
           tensorflow \
           gym \
           pyquaternion \ 
           cloudpickle \
           mpi4py \
           matplotlib
   cd <path_to_catkin_ws>/src/drl_local_planner_forks/stable_baselines/
   <path_to_venv>/venv_p3/bin/pip install -e .[docs,tests]
   

3. Set system-relevant variables

   • Modify all relevant pathes rl_agent/config/path_config.ini

Copy the example_agents in your "path_to_checkpoint"

1. Train static agent

   • Open first terminal:

   roscore
   

   • Open second terminal:

   cd <path_to_catkin_ws>/src/drl_local_planner_ros_stable_baselines/start_scripts
   ./entrypoint_ppo.sh ppo_1
   

   • Two screen sessions are created:
     • ROS setup with Flatland and Pedsim is started in:

       screen -r ros
       

     • Python3 implementation is started in the virtual environement.

     screen -r python
     

     It includes all RL-relevant as well as the communication with move_base
     • To detach from screen press: Strg+a, d.
   • Open third terminal:

   roslaunch rl_bringup rviz.launch ns:="sim1"
   

   • Rviz with appropriate configuration starts and you should see the robot (grey rectangle) train.

2. Train dynamic agent with parallel multiple simulation envs

   • Open first terminal:

   roscore
   

   • Open second terminal:

   cd <path_to_catkin_ws>/src/drl_local_planner_ros_stable_baselines/rl_agent/start_scripts
   ./entrypoint_ppo2.sh ppo2_1 2
   

   • Two screen sessions are created:
     • ROS setup with simulation 1 (Flatland and Pedsim) is started.

     screen -r ros_sim1
     

     • ROS setup with simulation 2 (Flatland and Pedsim) is started

     screen -r ros_sim2
     

     • Python3 implementation is started in virtual environements.

     screen -r python
     

     It includes all RL-relevant as well as the communication with move_base
     • To detach from screen press: Strg+a, d.
   • Open third terminal to see what happens in simulation 2:

    roslaunch rl_bringup rviz.launch ns:="sim2" 
   

   • Rviz with appropriate configuration starts and you should see the robot (grey rectangle) train.

3. Execute trained ppo1-agent with static objects

   • set train_mode: 0 in rl_bringup/config/rl_params_img_dyn.yaml
   • Open first terminal:

   roscore
   

   • Open second terminal:

   roslaunch rl_bringup setup.launch ns:="sim1" rl_params:="rl_params_img"
   

   • Open third terminal:

   source <path_to_venv>/bin/activate 
   roslaunch rl_agent run_ppo_agent.launch mode:="train"
   

   • Open fourth terminal:

   roslaunch rl_bringup rviz.launch ns:="sim1"
   

   • Set 2D Navigation Goal in rviz

4. Execute trained ppo2-agent with pedestrians

   • set train_mode: 0 in rl_bringup/config/rl_params_img_dyn.yaml
   • Open first terminal:

   roscore
   

   • Open second terminal:

   roslaunch rl_bringup setup.launch ns:="sim1" rl_params:="rl_params_img_dyn"
   

   • Open third terminal:

   source <path_to_venv>/bin/activate 
   roslaunch rl_agent run_ppo2_agent.launch mode:="train"
   

   • Open fourth terminal: roslaunch rl_bringup rviz.launch ns:="sim1"
   • Set 2D Navigation Goal in rviz