Improved Robot Design

Abstract

Existing robots lack teachability and adaptability. This paper advances teachable robot design by developing robot internal representations along with the teacher-robot interaction. Three improvements are proposed for the popular, easy to use, leading method, in which a teacher literally leads the robot by the hand through desired action sequences. The improvements do not require the teacher to use programming languages. The only skill he needs is at the task to be taught. The improvements advance teachable robot design by enabling a teacher to carry out three teaching operations with ease: (a) make online verbal corrections to a led robot's sequences, (b) teach individual actions, and (c) set goals. They advance adaptable robot design by enabling a robot to select its own actions for performing movement sequences and for seeking goals. The improvements are demonstrated with a simple simulation of a domestic vacuum cleaning robot.