tienhoangvan / robotics-rl Goto Github PK

################################################################################
#  About                                                                       #
################################################################################

Author: Lisa Lee
Advisor/mentors: Dr. Eric Eaton, Dr. Haitham Bou Ammar, Christopher Clingerman

This work was supported by the National Science Foundation REU Site at the
GRASP Laboratory in the University of Pennsylvania.

My research paper can be found at www.princeton.edu/~lilee/reu_paper.pdf

################################################################################
#  List of ROS packages                                                        #
################################################################################

* lilee_follower: Follows object inside the rectangle box published in
  /location_data. (See https://youtu.be/bV7wMxJ8f7s)

* lilee_gazebo: Spawns a turtlebot in gazebo.

* lilee_rl_continuous: Simulates policy gradient (PG) learning and PG-ELLA in
  gazebo. The state and action spaces are continuous.

* lilee_rl_discrete: Simulates Q-Learning in gazebo. The state and action
  spaces are discrete.

* lilee_tracker: Tracks an object whose color matches rgbhsv values in
  thresh.txt, and publishes a rectangle box surrounding that object (on the
  camera screen) to /location_data. (See lilee_tracker/lilee_tracker.jpg)

################################################################################
#  Launch instructions                                                         #
################################################################################

* Q-Learning (gazebo)
  1. roslaunch lilee_gazebo launch_maze_qlearning.launch
  2. roslaunch lilee_rl_discrete qlearning.launch

* PG (gazebo)
  1. roslaunch lilee_gazebo launch_maze_pg.launch
  2. In MATLAB:
    i.  Add LifelongRL-Robotics/lilee/MATLAB to path
    ii. Run pg_startup
  3. roslaunch lilee_rl_continuous pg.launch

* PG-ELLA (gazebo)
  1. roslaunch lilee_gazebo launch_maze_pg.launch
  2. In MATLAB:
    i.  Add LifelongRL-Robotics/lilee/MATLAB to path
    ii. Run pg_ella_startup
  3. roslaunch lilee_rl_continuous pg-ella.launch

* Edit maze.world:
  - For Q-learning:
    1. roslaunch lilee_gazebo edit_maze_qlearning.launch
    2. Modify the world in gazebo
    3. "Save As..." lilee_gazebo/worlds/maze_qlearning.world
  - For PG/PG-ELLA:
    1. roslaunch lilee_gazebo edit_maze_pg.launch
    2. Modify the world in gazebo
    3. "Save As..." lilee_gazebo/worlds/maze_pg.world

* color-follower (gazebo)
  1. roslaunch lilee_gazebo turtlebot_maze.launch
  2. rosrun lilee_tracker tracker.py
  3. rosrun lilee_reinforcement agent
  4. roslaunch lilee_follower follower.launch
  5. rosrun lilee_reinforcement environment

* color-follower (turtlebot)
  1. roslaunch turtlebot_bringup minimal.launch
  2. roslaunch lilee_follower follower.launch
  3. rosrun lilee_tracker tracker.py

################################################################################
#  Usage instructions                                                          #
################################################################################

* How to use the reinforcement packages:
  1. Press Down arrow to turn user control ON/OFF. Default is OFF.
  2. When user control is ON, use Up/Right/Left arrows to choose actions for
     the agent (turtlebot).

* To close PG/PG-ELLA
  1. ^C in terminal to kill lilee_gazebo and lilee_rl_continuous
  2. Type `clear' in MATLAB

################################################################################
#  Installation instructions                                                   #
################################################################################

* ROS I/O Package (MATLAB support for ROS): 
  1. Download ROS I/O Package from mathworks.com/ros
  2. sudo apt-get install ia32-libs
  3. Double-click ROSIOPackage-R2014a_v0.1.6.2_XXXXX-Install to launch the
     installer. 

* PG-ELLA 
  - Policy Gradient Toolbox: ias.tu-darmstadt.de/Research/PolicyGradientToolbox
  - ELLA code: http://www.seas.upenn.edu/~eeaton/publications.html#Software

################################################################################
#  Stored files                                                                #
################################################################################

* Q-Value stored files
  - The agent's Q-Value table is saved in ~/.ros/saved_qValue.
  - The agent's User Policy table is saved in ~/.ros/saved_userPolicy.