Future wireless networks are expected to support a wide range of use cases with diverse requirements. Among these, Ultra-Reliable Low-Latency Communications (URLLC) stand out as a key enabler for transformative applications across industries such as intelligent transportation, smart grids, and industrial automation.
These applications demand extremely high transmission reliability coupled with very low latency, posing stringent quality of service (QoS) challenges. Since the emergence of 5G, numerous enhancements have been introduced to meet URLLC demands.
However, ensuring adequate QoS still requires the design of new medium access control (MAC) protocols and advanced scheduling policies. In this seminar, we present recent research addressing the scheduling of URLLC traffic on the uplink using Modern Random Access (MRA) and Multi-Agent Reinforcement Learning (MARL).
We first formulate the URLLC MRA (uMRA) problem as a Decentralized Partially Observable Markov Decision Process (Dec-POMDP). We then introduce uMRA-HAPPO, a MARL-based solution employing the Heterogeneous Agents Proximal Policy Optimization (HAPPO) algorithm.
Finally, we evaluate our approach in scenarios inspired by 3GPP use cases, demonstrating that uMRA-HAPPO achieves higher reliability than traditional deterministic and random scheduling strategies, while satisfying strict latency constraints.