Acquisition of shared symbols in multi-agent cooperative tasks

In this paper a novel mechanism for acquiring shared symbols in multi-agent cooperative task is introduced. Inspired by human communication, a technique is suggested in which learning the behaviors and learning how to communicate are decomposed. Decomposing the shared symbol acquisition into two separate learning phases not only simplifies the learning algorithm but also it speeds up the process. Moreover, utilizing the gained information about the environment in the behavior learning phase, agent communication is learned easily. A couple of simulations are conducted to support the idea. Simulation results show that agents could assign meaning to symbols and transfer information among themselves using the learned symbols. Roughly speaking, they could form a language.