[TB3] Machine Learning Tutorial


(Gil kijong) #1

Do you want to try the machine learning?

We provide machine learning tutorial with TurtleBot3.

First, we wanted to try machine learning, but we provided an installation tutorial for people who were frustrated at the installation stage.

Second, This tutorial is reinforcement learning using DQN. We provide basic theory of DQN and algorithms that we use to help users understand reinforcement learning so that they can apply their own learning. What is important in learning and what to do to get the learning you want.

Finally, In this tutorial, TurtleBot3 learns to go to the goal in gazebo maps:
No obstacle, Static obstacle, Moving obstacle, Combination obstacle

Each map has different learning time and reward function.
Challenge to create more efficient learning models and create agents with different goals in different maps.
We are by your side.

For more information, please refer to the manuals below.
[Source code] : https://github.com/ROBOTIS-GIT/turtlebot3_machine_learning
[Document] : http://emanual.robotis.com/docs/en/platform/turtlebot3/machine_learning/#installation


(Kriz Chong) #2

Hi, Gilbert. This is really a great tutorial to start learning machine learning in ROS. I am very new in machine learning in ROS and my knowledge on machine learning is also not professional. But I have few questions:

  1. Every time I launch “roslaunch turtlebot3_dqn turtlebot3_dqn_stage_#.launch”, it will start a new learning, right? Is it possible to start the learning with learned Q-Value?

  2. Where is result of learning stored? Can we just run with the learned Q-Value without learning? How?


(Gil kijong) #3

Hi @krizchong :slight_smile:
Sure, you can do it. This tutorial stores the model for each tenth episode as stage_#EP#.h5 in save_model folder. When you want to restart launch using save_model, Change self.load_model = false to true and self.load_episode = 0 to episode you wanted in turtlebot3_dqn_stage# file.

And then… now you can’t just run without learning. But, I’ll update switch function between learning mode and non-learning mode.