Learning Attentional Communication with a Common Network for Multiagent Reinforcement Learning

Figure 1

Overall framework of multiagent attentional communication with the common network. Right: the agent first obtains consensus information of other agents from the common network. The feature is the hidden layer output of the agent policy at times . The feature is the output of the agent through the attention unit at times . Left: the communication module between agents is implemented via the attention mechanism. The feature is the hidden layer output of the agent policy at times . The feature is the output of the agent through the attention unit at times .