Demonstration of the Multi-Agent Reinforcement Learning Testbed for Cognitive Radio Applications

Radio Frequency Reinforcement Learning (RFRL) is a growing area of interest in the development of next-generation wireless communication systems, and has strong potential for applications in 6G and military communications. To foster research and innovation in this field, we have previously developed an open-source simulation tool, the RFRL Gym, to design and test reinforcement learning (RL) algorithms that carry out dynamic spectrum access (DSA) and adversarial jamming tasks in RF scenarios. We propose to demonstrate the latest version of the RFRL Gym, which now supports multi-agent reinforcement learning (MARL) through integration with the RLlib framework and the Gymnasium API. The demo will showcase how users can define complex multi-agent RF scenarios using JSON configurations and run simulations where agents compete for spectrum resources or engage in jamming/anti-jamming tasks. We will present a two-part demo: (1) creating user-defined multi-agent scenarios that align with high-level RF situations, and (2) training and evaluating MARL algorithms on these scenarios along with visualizations. This demonstration aims to solicit feedback from experts in both RF and RL research and to highlight the potential of the newly modified RFRL Gym as a valuable tool for testing and developing RL algorithms in complex RF environments.