What is this repository for?
- 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
- Link to IROS Paper
- Link to Master Thesis for more in depth information.
Installation (Else: Docker below)
-
Standart ROS setup (Code has been tested with ROS-kinetic on Ubuntu 16.04)
-
Install additional packages
apt-get update && apt-get install -y \ libqt4-dev \ libopencv-dev \ liblua5.2-dev \ virtualenv \ screen \ python3-dev \ ros-kinetic-tf2-geometry-msgs \ ros-kinetic-navigation \ ros-kinetic-rviz
-
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 missing packages)
-
Setup virtual environment to be able to use python3 with ros (consider also requirements.txt)
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=="1.13.1" \ gym \ pyquaternion \ mpi4py \ matplotlib cd <path_to_catkin_ws>/src/drl_local_planner_forks/stable_baselines/ <path_to_venv>/venv_p3/bin/pip install -e path_to_catkin_ws>/src/drl_local_planner_forks/stable-baselines/
-
Set system-relevant variables
- Modify all relevant pathes rl_bringup/config/path_config.ini
Example usage
-
Train agent
- Open first terminal (roscore):
roscore
- Open second terminal (simulationI:
roslaunch rl_bringup setup.launch ns:="sim1" rl_params:="rl_params_scan"
- Open third terminal (DRL-agent):
source <path_to_venv>/bin/activate python rl_agent/scripts/train_scripts/train_ppo.py
- Open fourth terminal (Visualization):
roslaunch rl_bringup rviz.launch ns:="sim1"
-
Execute self-trained ppo-agent
- Copy your trained agent in your "path_to_models"
- Open first terminal:
roscore
- Open second terminal:
roslaunch rl_bringup setup.launch ns:="sim1" rl_params:="rl_params_scan"
- Open third terminal:
source <path_to_venv>/venv_p3/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
Run pretrained Agents
Note: To be able to load the pretrained agents, you need to install numpy version 1.17.0.
<path_to_venv>/venv_p3/bin/pip install numpy==1.17
Run agent trained on raw data, discrete action space, stack size 1
- Copy the example_agents in your "path_to_models"
- Open first terminal:
roscore
- Open second terminal for visualization:
roslaunch rl_bringup rviz.launch ns:="sim1"
- Open third terminal:
roslaunch rl_bringup setup.launch ns:="sim1" rl_params:="rl_params_scan"
- Open fourth terminal:
source <path_to_venv>/venv_p3/bin/activate roslaunch rl_agent run_1_raw_disc.launch mode:="train"
Run agent trained on raw data, discrete action space, stack size 3
- Step 1 - 4 are the same like in the first example
- Open fourth terminal:
source <path_to_venv>/venv_p3/bin/activate roslaunch rl_agent run_3_raw_disc.launch mode:="train"
Run agent trained on raw data, continuous action space, stack size 1
- Step 1 - 4 are the same like in the first example
- Open fourth terminal:
source <path_to_venv>/venv_p3/bin/activate roslaunch rl_agent run_1_raw_cont.launch mode:="train"
Run agent trained on image data, discrete action space, stack size 1
- Step 1 - 3 are the same like in the first example
- Open third terminal:
roslaunch rl_bringup setup.launch ns:="sim1" rl_params:="rl_params_img"
- Open fourth terminal:
source <path_to_venv>/venv_p3/bin/activate roslaunch rl_agent run_1_img_disc.launch mode:="train"
Training in Docker
I set up a docker image, that allows you to train a DRL-agent in parallel simulation environments. Furthermore, it simplifies the deployment on a server. Using docker you don't need to follow the steps in the Installation section.
- Build the Docker image (This will unfortunately take about 15 minutes):
cd drl_local_planner_ros_stable_baselines/docker
docker build -t ros-drl_local_planner .
Training from scratch
-
In start_scripts/training_params/ppo2_params, define the agents training parameters.
Parameter Desctiption agent_name Number of timestamps the agent will be trained. total_timesteps Number of timestamps the agent will be trained. policy see PPO2 Doc gamma see PPO2 Doc n_steps see PPO2 Doc ent_coef see PPO2 Doc learning_rate see PPO2 Doc vf_coef see PPO2 Doc max_grad_norm see PPO2 Doc lam see PPO2 Doc nminibatches see PPO2 Doc noptepochs see PPO2 Doc cliprange see PPO2 Doc robot_radius The radius if the robot footprint rew_func The reward functions that should be used. They can be found and defined in rl_agent/src/rl_agent/env_utils/reward_container.py. num_stacks State representation includes the current observation and (num_stacks - 1) previous observation. stack_offset The number of timestamps between each stacked observation. disc_action_space 0, if continuous action space. 1, if discrete action space. normalize 0, if input should not be normalized. 1, if input should be normalized. stage stage number of your training. It is supposed to be 0, if you train for the first time. If it is > 0, it loads the agent of the "pretrained_model_path" and continues training. pretrained_model_name If stage > 0 this agent will be loaded and training can be continued. task_mode - "ped" for training on pedestrians only; "static" for training on static objects only; "ped_static" for training on both, static -
There are some predefined agents. As example I will use the ppo2_1_raw_data_disc_0 in the training session.
docker run --rm -d \ -v <folder_to_save_data>:/data \ -v drl_local_planner_ros_stable_baselines/start_scripts/training_params:/usr/catkin_ws/src/drl_local_planner_ros_stable_baselines/start_scripts/training_params \ -e AGENT_NAME=ppo2_1_raw_data_disc_0 \ -e NUM_SIM_ENVS=4 \ ros-drl_local_planner
-
If you want to display the training in Rviz, run the docker container in the hosts network. In order to use rviz, the relevant packages need to be compiled on your machine.
docker run --rm -d \ -v <folder_to_save_data>:/data \ -v drl_local_planner_ros_stable_baselines/start_scripts/training_params:/usr/catkin_ws/src/drl_local_planner_ros_stable_baselines/start_scripts/training_params \ -e AGENT_NAME=ppo2_1_raw_data_disc_0 \ -e NUM_SIM_ENVS=4 \ --net=host \ ros-drl_local_planner
Now you can display the different simulation environments:
- Simulation 1:
roslaunch rl_bringup rviz.launch ns:="sim1"
- Simulation 2:
roslaunch rl_bringup rviz.launch ns:="sim2"
- etc. ...
- Simulation 1:
Train with pre-trained agents
Run agent trained on raw data, discrete action space, stack size 1
```
docker run --rm -d \
-v drl_local_planner_ros_stable_baselines/example_agents:/data/agents \
-v drl_local_planner_ros_stable_baselines/start_scripts/training_params:/usr/catkin_ws/src/drl_local_planner_ros_stable_baselines/start_scripts/training_params \
-e AGENT_NAME=ppo2_1_raw_data_disc_0_pretrained \
-e NUM_SIM_ENVS=4 \
--net=host \
ros-drl_local_planner
```
Run agent trained on image data, discrete action space, stack size 1
```
docker run --rm -d \
-v drl_local_planner_ros_stable_baselines/example_agents:/data/agents \
-v drl_local_planner_ros_stable_baselines/start_scripts/training_params:/usr/catkin_ws/src/drl_local_planner_ros_stable_baselines/start_scripts/training_params \
-e AGENT_NAME=ppo2_1_img_disc_1_pretrained \
-e NUM_SIM_ENVS=4 \
--net=host \
ros-drl_local_planner
```