Proceedings of the 1996 IEEE International Conference on Robotics and Automation Minneapolis, Minnesota zyxwvutsrqpon - April 1996 zyxwvutsrqpo Robotic Office Room to Support Office Work Human Behavior Understanding Function with Networked Machines bY Hiroshi MIZOGUCHI Tomomasa SATO' Tsuyoshi ISHIKAWA' 'Research Center for Advanced Science and Technology, The University of Tokyo 4-6-1 Komaba, Meguro-ku, Tokyo 153, Japan Vokyo Engineering University 1401-1 Katakura-cho, Hachiohji, Tokyo, Japan Abstract This paper proposes a novel human-robot symbiosis system named a Robotic Office Room(R0R). The understanding, recording and retrieving of telephone call behavior are successfully demonstrated as a typical example of an office work-supporting function of ROR. A unique object model is proposed, which accumulates information concerning motion of the object. This model enables the system to understand human behavior, not by monitoring the motion of the human, but by monitoring the motion of the object - typically, a telephone or pen. Since the model requires the system to pay spatial and temporal attention only to the target object and related motions, it saves computing power. Repertoires of understandable behavior can easily be expanded by simply adding new object models. The ROR is equipped with networked machines. The concept of "ubiquitous link" consisting of unique network image links and remote control links is proposed. The network image zyxwvutsr link and the remote control link utilize zyxwvut TV images and either ultrasonic waves or infrared rays, respectively. Such cooperative support of humans by networked machines is realized that when a human starts to talk on the phone, the ROR not only starts to record the image of the human's behavior, utilizing a VCR, but also turns down the volume on an audio set. The ROR is experimentally proved to be a unique distributed robotic system to support human office work. 1. Introduction While human beings live with a lot of machines in modem society, these machines are not easy to use. There are two reasons for this difficulty. One reason is that the machines do not monitor humans and therefore cannot adequately follow human intentions or conditions. If machines monitored humans and understood their intentions through their behavior, machines would be easier to be utilized. In this case, behavior plays an important role as the medium through which to realize seamless interface between machines and human beings. Another reason of inconvenience of the conventional machines is that they are not connected each other and thus cannot work cooperatively. Although computer networks are widely used in office environments these days, computer networks only connect computers that work in a logical world. A new type of network system is required to connect machines which support human physical activity in the real world. To solve the above problems this paper proposes a Robotic Ofice Room (denotes "ROR" hereafter) which is equipped with networked machines. The ROR is a human- machine symbiosis system that supports human activity in the configuration of an office. Understanding human telephone call behavior and recordingiretrieval of the behavior are successfully demonstrated as a model function. The ROR monitors human behavior through a TV camera. When it detects pre-defined features of the motion of the object caused by human behavior, such as use of a telephone or writing of a memo, the ROR starts to record a video of the human behavior. The recorded memo could later heretrieved by human staff. This enables augmentation of human memory. There are several published research papers on machines that monitor and understand the meaning of human behavior. Sato et al.[l] realized a teleoperation system which interprets an operator's maneuvering behavior and uses this information to manage a world model of the telerobot . Kuniyoshi et a1.[2] demonstrated a "teaching by showing" system that understands human pick and place behavior by utilizing a real time vision system. They succeeded in generation of a program sequence that tells a manipulator to imitate the task. Ikeuchi et a1.[3] realized a system which visually understands complicated tasks, such as assembly of polyhedra by humans, and re-executes the same task using a robot. Research on "programming by gesture" or "programming by human demonstration" systems are activated recently [4-71. However, human behaviors that can be understood by these systems are limited to pick and place, grasping and degrasping , or assembly related tasks. A behavior model is necessary for machines to 0-7803-2988-4/96 $4.00 zyxwvut 0 1996 IEEE 2968