Vocabulary Learning and Sound Localization in Autonomous Robots

The simulation of vocabulary acquisition in embodied artificial agents (robots) has been studied by various methods and protocols. Here we simulate the emergence of a common repertoire of signs applying unsupervised learning and phonotaxis to enable a hearer robot to locate a speaker robot and the object referred during communication. Results show that the proposed scenario and control mechanisms allow the robots to find the sound source and lead to robust learning and establishment of a shared repertoire of sign by robots, even in the presence of noise and sensor data identification errors.