Reliable MAC Protocols under Harsh Channel Conditions via Reinforcement Learning

Medium access control (MAC) protocol learning through multi-agent reinforcement learning (MARL) has shown promising results in simple wireless networks, enabling efficient goal-oriented channel access policies. However, to the best of our knowledge, no existing study has applied a MARL framework to extreme environments with harsh channel conditions, where efficient channel access and proper buffer management are essential to ensure both communication reliability and low transmission latency. In this paper, we propose a novel MARL framework tailored for extreme environments to enable the learning of reliable MAC protocols. The proposed framework is designed to support the training phase directly in real-world scenarios, in order to capture specific channel conditions and traffic patterns. The performance of our MARL framework is evaluated against three benchmark solutions, none of which are specifically designed for networks with harsh channel conditions. Experimental results demonstrate that the MAC protocols learned through our framework achieve superior performance, yielding lower delays than legacy benchmarks and outperforming AI-based solutions in both reliability and communication timeliness.