The EmMAC project explores different AI methods, based on Multi-Agent Reinforcement Learning and the new “Learning to Communicate” paradigm, to enable autonomous wireless nodes to negotiate new protocols that outperform man-made protocols. The long-term vision is to enable the emergence of an entirely new medium-access control plane. Such protocols, which are learned and evolved by the machines themselves based on their experience, will better cope with complex communication situations (such as the outdating of channel state information, optimal timing and coordination between multiple nodes, etc.), which are difficult to handle by purely human engineering.
