Grounding Sequences of Actions in an Autonomous Doll Robot

in an autonomous doll robot Aude Billard, Gillian Hayes Department of Arti cial Intelligence University of Edinburgh, 5 Forrest Hill, Edinburgh EH1 2QL, U.K. [email protected] Abstract This paper presents DRAMA, a connectionist model, for learning time series. DRAMA is a fully recurrent neural network without hidden units using Hebbian update rule. We present its functioning and discuss its performance theoretically. We then report on its implementation in Robota, an autonomous doll robot. By imitating the movements of a demonstrator, the robot learns to perform and label sequences of actions (dance patterns). Further, the robot is taught combination of words, which form grammatically correct sentences, to describe its actions and perceptions. 1 The DRAMA architecture