Controlling an Entertainment Robot through Intuitive Gestures

Abstract

This paper presents a novel system which learns how the users want to control the entertainment robot with intuitive gestures. Two major problems are focused on this paper. One is how to distinct the user intuitive gestures, another is how to make the correspondence between the gestures and the control commands. Users want to use their own gestures which are not known a priori. These gestures must be classified by the systems by themselves. After the classification, the system must interpret the user intention to the control command which the user want to input. For instance, one may use waving hand gesture as the shutdown command, and another may assign the same gesture to the different command such as turning around. The primitive systems are implemented on the entertainment robot named AIBO. AIBO recognizes the gestures which are classified using Self-creating and organizing neural network. AIBO also learns how to interpret them into its command using Q-Learning technique. Some experimental results show the feasibility of the proposed approach.