robocup-mujoco consists of a single simulation server that communicates with any number of agents & monitors over network.
Idea is to develop a ROS2-based agent that directly runs Booster’s C++ ROS2 nodes; can replace behavior-related nodes (brain) with our own Python nodes to make the agent RL trainable.
Can also develop a Python-only agent for training policies that don’t require realisitc observations (like AbstractSim).