Learning to Communicate Through Imitation in Autonomous Robots

Communication is a desirable skill for robots. We describe a method of how these skills could be learned. A control architecture of con-nectionist model combining lifelong learning and predeened behaviours is developed and implemented in a physical system of two autonomous robots. A teaching scenario based on movement imitation is used to teach a basic non grammatical language. Teaching is done from a teacher robot to a student robot. While following the teacher, the student robot learns a basic vocabulary concerning its movements and location. 1 Teaching and learning a language Communication is a social skill and as such may be desirable for autonomous agents that are expected to interact with other agents or humans. In particular, communication seems a desirable skill for learning agents to possess as it would provide us with an easy means for teaching or directing them. Moreover, a exible language which is acquired through a learning process would be advantageous over a xed communication protocol as it would make it more adaptable to the tasks the robots can carry out and the types of robots engaged in those tasks. This works investigates how a common communication protocol could be transmitted from a teacher agent to a learner agent, in terms of the agents' behavioural and cognitive capabilities. By opposition to previous simulation studies on language acquisition ((1], 6], 8]), we take a physical approach to the problem of the language understanding. We focus on the learning and understanding of a language at the level of the individual rather than at the level of the society and propose a method, in terms of the agent's behavioural capacities for transmitting the language from a teacher agent to a student agent. Movement imitation is the means by which the language is taught. A student robot follows and thus imitates the teacher's movements. The student attaches a meaning to the teacher's words in terms of its sensor measurements and motor states. In this context, thèlanguage' consists of a simple vocabulary, where thèwords' are (radio) signals that label speciic conngurations of sensor-motor state. However, it is our belief that the method