DRL-robot-navigation

Deep Reinforcement Learning for mobile robot navigation in ROS Gazebo simulator. Using Twin Delayed Deep Deterministic Policy Gradient (TD3) neural network, a robot learns to navigate to a random goal point in a simulated environment while avoiding obstacles. Obstacles are detected by laser readings and a goal is given to the robot in polar coordinates. Trained in ROS Gazebo simulator with PyTorch. Tested with ROS Melodic on Ubuntu 18.04 with python 3.6.9 and pytorch 1.10.

Training example:

Pre-print of the article:

Some more information is given in the article at: https://arxiv.org/abs/2103.07119

Please cite as:
@misc{cimurs2021goaldriven,
title={Goal-Driven Autonomous Exploration Through Deep Reinforcement Learning},
author={Reinis Cimurs and Il Hong Suh and Jin Han Lee},
year={2021},
eprint={2103.07119},
archivePrefix={arXiv},
primaryClass={cs.RO}
}

Main dependencies:

• ROS Melodic
• PyTorch

Clone the repository:

$ cd ~
### Clone this repo
$ git clone https://github.com/reiniscimurs/DRL-robot-navigation

The network can be run with a standard 2D laser, but this implementation uses a simulated 3D Velodyne sensor

Compile the workspace:

$ cd ~/DRL-robot-navigation/catkin_ws
### Compile
$ catkin_make_isolated

Open a terminal and set up sources:

$ export ROS_HOSTNAME=localhost
$ export ROS_MASTER_URI=http://localhost:11311
$ export ROS_PORT_SIM=11311
$ export GAZEBO_RESOURCE_PATH=~/DRL-robot-navigation/catkin_ws/src/multi_robot_scenario/launch
$ source ~/.bashrc
$ cd ~/DRL-robot-navigation/catkin_ws
$ source devel_isolated/setup.bash
### Run the training
$ cd ~/DRL-robot-navigation/TD3
$ python3 velodyne_td3.py

To kill the training process:

$ killall -9 rosout roslaunch rosmaster gzserver nodelet robot_state_publisher gzclient python python3

Gazebo environment:

Rviz: