Velmwheel Gym

Environment for developing and testing reinforcement learning algorithms for WUT Velmwheel robot. Environment is compliant with OpenAI Gym's API (https://github.com/openai/gym)

Setup up

Build and run WUT Velmwheel project.
Open another terminal, go to root of the Velmwheel project and run

source source_me.bash

NOTE: This is important because some of the packages needed to create the Gym environment are not installable using pip. Alternatively you can install ROS2 Humble system-wide.

Create and activate Python's virtual environment

python3 -m venv venv
source venv/bin/activate

Install packages

pip install -r requirements.txt

NOTE: The Gym environment depends on custom Gazebo message for passing contact state which is built as a part of the Velmwheel project.

TODOs

Refactoring

• ros2_numpy - install from pip or move to extern (consider using a git submodule)
• move Gazebo models to velmwheel_gym/gazebo_env/models
• allow to choose between weights & biases and local tensorboard
• clean up available CLI options (remove useless ones)
• update README.md
• align all algorithms parameters
• add more tests
• add more comments

Improvements

• [] allow to specify custom maps in a file