Two Loggers Gazebo Simulation

Tutorials of the contents in this repo are all located at Docs. The order is: system_setup->create_urdf_tutorial->gazebo_ros_tutorial

Setup

• Ubuntu 16.04 or Ubuntu 18.04
• ROS-Kinetic in Ubuntu 16.04 or ROS-Melodic in Ubuntu 18.04
• Python 2.7,
• TensorFlow 2 and TensorFlow Probability The lastest Python2 supported tensorflow2 version is 2.1, tensorflow-probability version is 0.9, please make sure the right version is installed. Example: pip install tensorflow==2.1 and pip install tensorflow-probability==0.9

PyTorch or other deep learning libraries should be no problem working with the environment developed in this repo.

• gazebo_ros_pkgs

sudo apt-get install ros-melodic-gazebo-ros-pkgs ros-melodic-gazebo-ros-control

• Create a catkin workspace, assume your workspace is at ~/ros_ws/
• Clone this repo to your catkin workspace

cd ~/ros_ws/src
git clone https://github.com/linZHank/two_loggers.git

• Build ROS packages (loggers_description, loggers_gazebo, loggers_control)

cd ~/ros_ws/
catkin_make
source devel/setup.bash

Catkin Command Line Tools is a substitution to catkin_make.

• make sure following two lines are in your ~/.bashrc file.

source /opt/ros/melodic/setup.bash
source /home/linzhank/ros_ws/devel/setup.bash

Replace melodic with kinetic in the lines above if you are using ROS-Kinetic. You may want to copy all the contents in two_loggers/loggers_gazebo/models/grey_wall_10/grew_wall to ~/.gazebo/models/ if using Ubuntu 16.04 and ROS-Kinetic. e.g. cp -a ~/ros_ws/src/two_loggers/loggers_gazebo/models/grey_wall_10/ ~/.gazebo/models/

Environments

Two environments are available right now: this_repo/loggers_control/scripts/envs/se.py and this_repo/loggers_control/scripts/envs/de.py both are with discrete action space.

se indicates solo escape, the goal is control the logger robot exiting the room through the only opening on the south wall.

de indicates double escape, the goal is control a two-robot team formed with two logger robots exiting the room while carrying a 2m rod.

Usage

A segment of example code can be found in the end of each script.

Open a terminal and enter the following commands to launch the simulation.

roslaunch loggers_control double_logger_control.launch

In a new terminal (tab), enter the following commands to test the environment with random control signals.

cd this_repo/loggers_control/scripts/envs
python de.py

substitute this_repo with actual repo path.

for solo escape env, roslaunch loggers_control solo_logger_control.launch, then navigate to the envs location and run python se.py

Agents

Two DQN agents are available and tested:

1. DQN
2. PPO

Location: this_repo/loggers_control/scripts/agents

Usage

Open a terminal and enter the following commands to launch the simulation.

roslaunch loggers_control solo_logger_control.launch

In a new terminal (tab), enter the following commands to train a ppo controller in solo escape environment.

rosrun loggers_control train_se_ppo.py

Since the envs are updated, please refer to legacy_20200817 branch for tested scripts training distributed controllers using DQN algorithm.

Under developing scripts include new agents, new envs could be found in devel branch.