Purpose of this repo

This repo is Baxter learnig pick project, which uses ROS DMP package and RL off-policy Q-learning.

You can follow this README to know how it works.

Dependencies

Before running this project, you need install this dependencies:

1. Install ROS full-desk-version : ROS indigo wiki.

2. Install Baxter simulator workspace : Install Baxter simulator workspace.

   • If you don't have a real Baxter robot, pls run this project in the Baxter gazebo simulator.
   • If you have a real Baxter robot, follow this website and make sure that you can control you robot : Baxter getting start.

3. Install Robotiq force torque driver : force torque sensor driver , force torque sensor sensing the envrionemnt and give a reward to the RL agent. Don't worry, if you don't have this sensor, I will add this force torque sensor model to gazebo simulator, and update.

Let's Go

1. Generating a Baxter joint trajectory

• If you have a real Baxter robot, and follow this tutorial and generating a joint trajectory : Baxter Joint Trajectory Playback Example.
• Or you can navigate to this repo datasets file, the baxter_joint_input_data.csv is the joint trajectory csv file.

2. Generating a DMP trajectory

• This repo are using ROS DMP package to generate DMP trajectorys, move to this website to learn more: ROS DMP wiki.

How to do:

rosrun dmp dmp_server
rosrun dmp baxter_r_ram_dmp.py

[TIPS] : You need rewrite the code.

• This repo are generated the DMP trajectorys already, in the datasets data0 - data4 csv file are the DMP trajectorys.

3. Learnig a optomal DMP trajectroy.

This part are using the Reinforcement leaning off-polociy agent.

• After genarating the DMP trajectroys then running :

rosrun dmp dmp_joint_trajectory_action_server.py
rosrun dmp Baxter_DMP_RL_joint_trajectory_learn_client.py

[TIPS] : You need rewrite the code.