The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9607

Title:"Allo Kazam, do You Follow Me?" or Learning to Speak Through Imitation for Social Robots
Date: 1996
Abstract:This project grew out of our interest in building a system able to develop communication capabilities. Communication is part of social behaviour. Our work belongs to the field of research working towards the construction of autonomous, intelligent and social robots. An architecture of connectionist models was developed to allow the learning of simple syntactical language. Concepts, as a word and its meaning, and numerical quantities are understood separately. The learning of the language includes understanding of the meaning of a word and a memorisation of its elements, which allows to use it in return. A physical system of two mobile autonomous robots, teacher and student, was built to implement the architecture. Concepts relative to an agent's movements, such as being stopped and moving, and concepts relative to objects such as a battery were taught. Extension of the architecture to cover the understanding of quantities was tested by teaching the notion of distance. Teaching was done via a teacher agent to a student. However, the learning was unsupervised; no reward of any kind was provided by the teacher. The success of the learning depended upon the right circumstances. Emphasis was put on simplicity and generality while developing the learning architecture, so that it would not be specific to the robot's type nor to the environment used for the implementation.

[Search These Pages] [DAI Home Page] [Comment]