EXPERT CONTROL FOR A ROBOT BODY

Abstract

Mobile robots with dextrous hands and sophisticated sensory systems will require intelligent, knowledge-based, expert controllers. In this paper we develop a design for a robot controller which can acquire task knowledge as it interacts in the world with its human users. The design is based on four reasonable assumptions which lead us to a theoretical framework for robot learning systems. The framework is called a multiple context learning system. It is a production system with multiple templates for forming productions as the system interacts with the world. The paper discusses elaborations of the framework and experimental tests of the system.