Cooperative Multiagent Attentional Communication for Large-Scale Task Space

With the rapid development of mobile robots, they have begun to be widely used in industrial manufacturing, logistics scheduling, intelligent medical, and other fields. For large-scale task space, communication between multiagents is key affect cooperation productivity, agents can coordinate more effectively with help dynamic communication. However, traditional mechanism uses simple message aggregation broadcast and, some cases, lacks distinction importance information. Multiagent deep reinforcement learning (MDRL) valid solve problem informational coordination strategies. how different messages each agent’s decision-making process remains a challenging for task. To this problem, we propose IMANet (Import Message Attention Network). It divides into two substages: action, where considered part environment. First, an attention based on query vectors introduced. The correlation vector own information current state estimated, then, results are distinguish from agents. Second, LSTM network as unit controller agent, individual rewards guide agent training after Finally, evaluated tasks multi-agent platforms, Predator Prey (PP), traffic junction. show that improve efficiency training, especially when applied success rate 12% higher than CommNet baseline experiments.