hari-sikchi / gym-gazebo

A toolkit for developing and comparing reinforcement learning algorithms using ROS and Gazebo.

RL experiments on Husky integrated with OpenAI gym

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

Requirement

• ROS Indigo
• Gazebo 7

Build and install gym-gazebo

• Install the gym-gazebo following the installation instructions given in the repository.

For installation on ROS Indigo:-

1. Getting the ros packages necessary for our experiments

Get the following packages individually from their git respository(indigo branch) and replace/insert(if not already present) them in the <INSTALLATION_ROOT>/gym_gazebo/envs/installation/catkin_ws/src/ directory.

• ar_track_alvar_msgs
• ecl_core
• ecl_navigation
• ecl_tools
• ecl_lite
• mav_comm
  Now, for the husky environment we require packages provided by Clearpath Robotics for model import.
• husky
  Note: Replace the urdf and launch files in husky/husky_description with those given in additional_packages/husky_description.
• Copy all the packages given in the additional_packages folder in the path <INSTALLATION_ROOT>/gym_gazebo/envs/installation/
• Navigate to <INSTALLATION_ROOT>/gym_gazebo/envs/installation/catkin_ws/
• Run catkin_make

2. Getting the plugins ready.

• The plugins for gazebo world allow us to define behaviour in gazebo world(Random pose, Random velocity of obstacles, etc).

• Plugins required for our experiment are located in <INSTALLATION_ROOT>/gym_gazebo/envs/husky/husky_plugin . Run the following commands to build the plugin. For more details regarding usage, see the README in the husky_plugin folder.

  mkdir build
  cd build
  cmake ..
  make
  

c. Add the following line to your bashrc.(Set the path for compiled plugins)

export GAZEBO_PLUGIN_PATH=${GAZEBO_PLUGIN_PATH}:<INSTALLATION_ROOT>/gym_gazebo/envs/husky/husky_plugin/build 

Running the experiments

  cd <INSTALLATION_ROOT>/examples/husky    
  python husky_reactive_control_qlearn.py  

Name	Middleware	Description	Observation Space	Action Space	Reward range
GazeboHuskyWallMovingObstacles-v0	ROS		Continous	Continuos
GazeboHuskyReactiveControlEnv	ROS		Discrete	Discrete

Visualize the Learning Process

---