DRL-MachineIntelligence

Deep Reinforcement Learning (DRL) project for the Machine Intelligence course

Assumed Environment

• Ubuntu 20.04
• Ubuntu 18.04

(Maybe other versions are available because we use docker)

Requirements

• docker
• nvidia-docker

How to install

Installation

git clone https://github.com/eliasenseirb/DRL-MachineIntelligence.git
cd Docker
chmod +x build.sh
./build.sh

Usage

Run docker container and enter its terminal

cd Docker
chmod +x run.sh
./run.sh

Run simulation

source setup_melodic.bash
source setup_turtlebot.bash
cd_src
python cafe_turtlebot_lidar_qlearn.py

Note: In the first time, bash setup_melodic.bash builds the ROS packages and it takes long time. But once it is built, the execution files are retained and the build time will be shortened from the next time.

Watch the simulation in Gazebo

Create a new terminal and enter the existing docker container's terminal

cd Docker
chmod +x new_terminal.sh
./new_terminal.sh

Setup the environment and launch Gazebo

source setup_melodic.bash
source setup_turtlebot.bash
source setup_display.sh
gzclient

Tips

Change initial robot position

export "ROBOT_INITIAL_POSE" environment variable like this

export ROBOT_INITIAL_POSE="-x 2.7 -y 2.7 -z 0.25 -R 0 -P 0 -Y -1.57"

Change goal position

export "GOAL_X" and "GOAL_Y" environment variables like this

export GOAL_X="-3.0"
export GOAL_Y="-9.0"

Q. How to create own environment?

A. See this commit

Reference

Gym Documentation

https://www.gymlibrary.ml/

Gazebo Gym Paper

https://arxiv.org/pdf/1608.05742.pdf%C3%AF%C2%BC%E2%80%B0