trqminh / project

Reinforcement Learning with DDPG

For this project my goal is to create a Deep reinforcement learning Code that can avoid obstacles while trying to get to a target

Base Idea

I saw a code of a DQN agent in the repository of ROBOTIS. But to create a better agent to control Robots I didn't see DQN with too much help in project because it only permits discrete actions. So my idea were to create a DDPG algorithm's agent based. A DDPG agent permits continous control for a robot. In my case I have as outputs: linear velocity (0 ~ 0.22m/s) and angular velocity (-1 ~ 1rad/s).

• https://github.com/ROBOTIS-GIT/turtlebot3_machine_learning

Libraries

[Pytorch]

ROS

You can find the packages the I used here:

• https://github.com/ROBOTIS-GIT/turtlebot3
• https://github.com/ROBOTIS-GIT/turtlebot3_msgs
• https://github.com/ROBOTIS-GIT/turtlebot3_simulations

cd ~/catkin_ws/src/
git clone {link_git}
cd ~/catkin_ws && catkin_make

To install my package you will do the same from above.

Set State

In: turtlebot3/turtlebot3_description/urdf/turtlebot3_burger.gazebo.xacro.

<xacro:arg name="laser_visual" default="false"/>   # Visualization of LDS. If you want to see LDS, set to `true`

And

<scan>
  <horizontal>
    <samples>360</samples>            # The number of sample. Modify it to 10
    <resolution>1</resolution>
    <min_angle>0.0</min_angle>
    <max_angle>6.28319</max_angle>
  </horizontal>
</scan>

Run Code

I have four stage as in the examples of Robotis. But I dont know yet my code dont have a geat performance in stage 3.

First to run:

roslaunch turtlebot3_gazebo turtlebot3_stage_{number_of_stage}.launch

In another terminal run:

roslaunch project ddpg_stage_{number_of_stage}.launch