Attention-Based Mechanisms for Cognitive Reinforcement Learning

In this paper, we propose a cognitive reinforcement learning method based on an attention mechanism (CRL-CBAM) to address the problems of complex interactive communication, limited range, and time-varying communication topology in multi-intelligence collaborative work. The not only combines efficient decision-making capability learning, representational deep self-learning but also inserts convolutional block module increase by using focus important features suppress unnecessary ones. use two modules, channel spatial axis, emphasize meaningful main dimensions can effectively aid flow information network. Results from simulation experiments show that has more rewards is than other methods formation control, which means greater advantage when dealing with scenarios large number agents. group containment, agents learn sacrifice individual maximize rewards. All tasks are successfully completed, even if scenario changes training scenario. therefore be applied new environments effectiveness robustness.