UR Sim-to-Real Reinforcement Learning

A ROS 2 package for transferring reinforcement learning policies from IsaacLab to Universal Robots (UR) hardware.

Overview

This package implements a bridge between trained IsaacLab policies and real UR Robots. Beforehand, you need to train a policy using IsaacLab and save it in the models/ directory.

Dependencies

• ROS 2
• Python 3
• PyTorch
• Universal_Robots_ROS2_GZ_Simulation package installed in your ROS 2 workspace

You need to have a working CUDA device with at least version 11.8 to run this package. For pytorch CUDA support, you can follow the instructions here.

Installation

1. Clone this repository to your ROS 2 workspace:

   cd ~/ros2_ws/src
   git clone https://github.com/kyavuzkurt/ur_sim_to_real_rl_games.git
   

2. Install dependencies:

   cd ~/ros2_ws
   rosdep install --from-paths src --ignore-src -r -y
   

3. Build the package:

   colcon build --packages-select ur_sim_to_real_rl_games
   

Usage

Put your trained model from IsaacLab in the models/ directory.

Simulation

To run in simulation mode:

ros2 launch ur_sim_to_real_rl_games launch_gazebo.launch.py

Real Robot

To deploy on a real UR robot, while the ur_robot_driver is running, you can run the following command:

ros2 launch ur_sim_to_real_rl_games real_robot.launch.py

Configuration

Configuration files are available in the config/ directory:

• simulation.yaml: Settings for simulation environment
• real_robot.yaml: Settings for real robot deployment

License

MIT Licenses

Author

Kadir Yavuz Kurt (k.yavuzkurt1@gmail.com)