Advanced Robotics, Vol. 20, No. 4, pp. 391–412 (2006) VSP and Robotics Society of Japan 2006. Also available online - www.vsppub.com Full paper Reasoning of abstract motion of a target object through task order with natural language — pre-knowledge of object-handling-task programming for a service robot RIE KATSUKI 1, , ROLAND SIEGWART 2 , JUN OTA 1 and TAMIO ARAI 1 1 The University of Tokyo, Tokyo 113-8656, Japan 2 Swiss Federal Institute of Technology Lausanne (EPFL), CH-1015 Lausanne, Switzerland Received 26 November 2004; accepted 26 March 2005 Abstract—In this paper, an attempt is made to reason an abstract motion of a target object when the object is manipulated through a task order. The reasoning algorithm is used as a function in the system that navigates layman’srobot programming. The task order consists of two words: Task and Target (e.g., ‘switch on’ and ‘light’). There are three kinds of motions: Linear, Circular and Point-To-Point motion. The system chooses a suitable motion. In the reasoning, it is important to be able to reason from various input words using a limited knowledge base. Therefore, a knowledge base is proposed that consists of a thesaurus and minimum knowledge. The knowledge defines only words that directly stand for the motions (e.g., ‘turn’ means Circular motion). The knowledge is propagated through hypernyms and hyponyms in the thesaurus. A motion is reasoned using the propagated knowledge in Task and Target. Moreover, learning, which results from on-site updating of the knowledge from the user, achieves reinforcement/customization of the knowledge base. The system successfully reasoned motions from various task orders. Moreover, for the robot programming of a door-opening task, a robot with the reasoning system reasoned a motion of the door and realized the task. Keywords: Reasoning; semantics; thesaurus; robot programming; service robot. 1. INTRODUCTION Realization of various object-handling tasks by a service robot has been expected for a long time. Since it is currently difficult for a robot to acquire a task autonomously, programming by a user is a feasible approach. In particular, teaching-by-showing (programming-by-demonstration) is a powerful method of robot programming. In teaching-by-showing, a system extracts pre-defined task features from its visual To whom correspondence should be addressed. E-mail: rie.katsuki@toshiba.co.jp