robo-gym V2.1: An Open Source Toolkit for Distributed Deep Reinforcement Learning on Real and Simulated Robots - Now with Modular Environment Classes and Support for Training in Isaac Lab

robo-gym is a toolkit for distributed Reinforcement Learning (RL) based on controlling simulated and real robots via ROS and gRPC.

Environments in robo-gym are compatible with the Env API from Gymnasium and thus offer compatibility with a wide range of RL algorithm implementations.

The archive contains master snapshots of the three individual repositories that need to work together. Such a setup typically consists of an agent side on the one hand, in which RL agents are implemented and any required libraries for this purpose are installed, and the server side on the other hand, which uses a ROS-enabled Python environment and runs the robot servers, simulations or interfaces to real robots, and required auxiliaries. Both sides may be run in a common environment (using the ROS-enabled Python to execute everything) if that does not result in conflicting dependencies.

• robo-gym: Python package containing the agent-side implementation of environments
• robo-gym-robot-servers: ROS packages that provide the server-side parts of environments and simulation elements
• robo-gym-server-modules: infrastructure package required on both sides

Mind the readme files of the individual archives and further documentation for the robo-gym package.