victorkich/turtlebot3

Gym module for training and control TurtleBot3 in any environment using ROS as backend.

TurtleBot3 Gym env

This is a gym env to work with the TurtleBot3 gazebo simulations, allowing the use of OpenAI Baselines and Stable Baselines deep reinforcement learning algorithms in the robot navigation training. See the examples folder to check some Python programs.

This project is a part of the development of some gazebo environments to apply deep-rl algorithms.

Research Gazebo environments for TurtleBot3 robot

Installation

git clone https://github.com/ITTcs/gym-turtlebot3
cd ITTcs/gym-turtlebot3
pip install -e .

This is an example using the OpenAI Baselines DDPG algorithm in a TurtleBot3 environment.

.\examples\openai_baselines_ddpg.py:

import gym
import gym_turtlebot3
import rospy
import baselines.run as run
import os

env_name = 'TurtleBot3_Circuit_Simple_Continuous-v0'
alg = 'ddpg'
num_timesteps = '1e4'

name_ref = env_name + '_' + num_timesteps + '_' + alg

my_args = [
    '--alg=' + alg, 
    '--env=' + env_name, 
    '--num_timesteps=' + num_timesteps]

os.environ["OPENAI_LOG_FORMAT"] = "csv"
os.environ["OPENAI_LOGDIR"] = './logs/' + name_ref

rospy.init_node(env_name.replace('-', '_'))

run.main(my_args)

Before running a Python program you first need run the specified environment. To the TurtleBot3_Circuit_Simple_Continuous-v0 env for example, would be:

roslaunch turtlebot3_gazebo turtlebot3_circuit_simple.launch

These resouces are part of a master's thesis where we use the Python fuzzylab library to create fuzzy logic controllers with the implementation of deep reinforcement learning algorithms.

To cite this repository in publications:

@misc{gymturtlebot3,
  author = {Avelar, Eduardo},
  title = {gym-turtlebot3},
  year = {2019},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/ITTcs/gym-turtlebot3}},
}