Difference with openai_ros

This package has been constructed following the blueprint of the openai_ros package, which was developed by the construct. Consequently, they share a foundational package structure. Both packages compartmentalize the Gym environments into distinct task, robot, and gazebo classes. Furthermore, they employ corresponding gazebo_connection and controllers_connection classes to establish communication with Gazebo and the controllers.

However, it’s important to note that while this common framework serves as the foundation, substantial deviations from the original openai_ros package have emerged. These modifications have led to a significant divergence in the codebase. Consequently, it’s essential to acknowledge that not all components maintain backward compatibility. To offer a more precise overview of the contrasting features, the table below highlights the most noteworthy differences between the two implementations:

	openai_ros	ros_gazebo_gym
License	ILPG	MIT
Import behaviour	Uses StartOpenAI_ROS_Environment function	Direct environment imports, similarly to other gymnasium environments
Focus	Simulated systems	Simulated and real systems
ROS Noetic compatible	❌	✅
ROS EOL compatible	✅	❌
Uses black/flake8 conventions	❌	✅
Clones environment repository	✅	✅
Installs system dependencies	❌	✅
Builds catkin workspace	❌	✅
Sources catkin workspace	❌	✅
Starts roscore and runs rosinit	❌	✅
Uses failsafe subprocesses	❌	✅
Extensive documentation	✅	✅
Numpy docstrings	✅	❌
Google docstrings	❌	✅
Code API documentation	❌	✅
Welcomes contributions	✅	✅
Passes kwargs to environments	❌	✅
Panda environment	❌	✅
Cartpole environment	✅	❌
Fetch environment	✅	❌
Hopper environment	✅	❌
Husarion environment	✅	❌
Iriwam environment	✅	❌
Moving_cube environment	✅	❌
Parrotdrone environment	✅	❌
Sawyer environment	✅	❌
Shadow_tc environment	✅	❌
Sumit_xl environment	✅	❌
Turtlebot2 environment	✅	❌
Turtlebot3 environment	✅	❌
Wamv environment	✅	❌

As the table above shows, the ros_gazebo_gym package works in ROS noetic, is easier to work with, is fully open-source, and comes with extensive code API documentation. The openai_ros on the other hand, currently contains more robotics environments and is compatible with EOL ROS versions. Nevertheless, new environments can easily be added to the ros_gazebo_gym package by following this guide.

Important

Although the ros_gazebo_gym package diverged significantly from the openai_ros package with some effort, it can be back-ported into the openai_ros package under an MIT license. I’m happy to do this if this achieves enough support from the ROS community.