turtlebot-gym-gazebo

Training a Turtlebot to move around in circles using ROS + Gazebo and OpenAI Gym.

Report.

Installation

Requirements:

• have Docker installed.
• install nvidia-docker2 (sudo apt-get install nvidia-docker2)

Copy run_demo.bash somewhere, make sure it is executable (chmod +x run_demo.bash).

bash run_demo.bash

You're now inside the container.

Navigate to ~/gym-gazebo/gym_gazebo/envs/installation, then execute the following:

• bash setup_kinetic.bash (might take a few minutes to finish compiling)
• bash turtlebot_nn_setup.bash

You should now be ready to execute any script!

To bring up Gazebo and visualize the robot, in a separate (host) terminal, run:

• docker exec -it --env DISPLAY --env QT_X11_NO_MITSHM=1 --privileged CONTAINER_ID bash

  CONTAINER_ID can be obtained by running docker ps.

• After logging into the container, run gzclient.

Summary

Main script to check out:

• turtlebot_circle_dqn_clean.ipynb

  IPython Notebook without output displayed.

  Has the main script for training a DQN for making a Turtlebot go in a circle.

• turtlebot_circle_dqn_with_results.ipynb

  IPython Notebook with output displayed

• test_drive_turtlebot.py

  Script for a quick demo of what the DQN has learned. To run this script, first source the environments as described in "Installation" and then:

  python test_drive turtlebot.py turtlebot_dqn_2

  turtlebot_dqn_2 loads weights of the DQN trained for "Exploration without Directed Bias" (refer report). Replace with turtlebot_dqn_1 for "Exploration with Directed Bias". Run gzclient in a separate terminal to view the robot moving in circles.

Weight files are located ./models.

To launch jupyter-notebook hosted inside the container on the host machine's browser, after the container has been launched using run_demo.bash:

jupyter notebook --ip 0.0.0.0 --no-browser --allow-root