EmMAC3 - Emerging Protocols for the MAC Layer with Multi-Agent Reinforcement Learning (Phase 3)

This project’s goal is to demonstrate the viability of MAC protocol learning by means of deep reinforcement learning techniques, in the context of native AI components for 6G. The EmMAC3 project focuses on overcoming scalability issues when training an L2-layer reinforcement learning agent who acts upon the L1 layer via a set of PHY API commands. Due to the vast size of the API instruction vocabulary, the challenge consists in devising ways to efficiently train the L2 agent. e.g., by exploiting redundancy and the structure of API commands and mixing data-driven and model-driven approaches.