Emergence of communication among reinforcement learning agents under coordination environment
Qiong Nian Huang, Eiji Uchibe, Kenji Doya · 2016
We have proposed a way to simulate how communication could emerge in a simple n by n grid world task among multi-agent. We showed that by introducing simple learning communication among multi reinforcement learners, agents could reach a coordinated behavior to obtain a higher reward. We also provided a feasible way to overcome the problem of partial observability with reinforcement learning in multi-agent systems.