Transmitting Communication Skills Through Imitation in Autonomous Robots

Abstract

Sharing a common context of perception is a prerequisite in order for several agents to develop a common understanding of a language. We propose a method, based on a simple imitative strategy, for transmitting a vocabulary from a teacher agent to a learner agent. A learner robot follows and thus implicitly imitates the movements of a teacher robot. While bounded by mutual following, learner and teacher agents are set in a position from which they share a common set of perceptions, that they can then correlate with an arbitrary set of signals, a vocabulary. The teacher robot leads the learner robot through a series of situations in which it teaches it a vocabulary to describe its observations. The learner robot grounds the teacher’s ‘words’ by associating them with its own perceptions. Learning is provided by a dynamical recurrent associative memory, an Artificial Neural Network architecture. The system is studied through simulations and physical experiments where the vocabulary concerns the agents’ movements and orientation. Experiments are successful, learning is fast and stable in the face of a significant amount of experimental noise. This work suggests then that a simple movement imitation strategy is an interesting scenario for the transmission of a language, as it is an easy means of getting the agents to share the same physical context.