Effective Communications: A Joint Learning and Communication Framework for Multi-Agent Reinforcement Learning Over Noisy Channels

We propose a novel formulation of the “effectiveness problem” in communications, put forth by Shannon and Weaver their seminal work “The Mathematical Theory Communication”, considering multiple agents communicating over noisy channel order to achieve better coordination cooperation multi-agent reinforcement learning (MARL) framework. Specifically, we consider partially observable Markov decision process (MA-POMDP), which agents, addition interacting with environment, can also communicate each other communication channel. The is considered explicitly as part dynamics message agent sends action that take. As result, learn not only collaborate but “effectively” This framework generalizes both traditional problem, where main goal convey reliably channel, “learning communicate” has received recent attention MARL literature, underlying channels are assumed be error-free. show via examples joint policy learned using proposed superior separately from MA-POMDP. very powerful framework, many real world applications, autonomous vehicle planning drone swarm control, opens up rich toolbox deep for design multi-user systems.