Agora: Supporting Multi-participant Telecollaboration Jun Yamashita a, Hideaki Kuzuoka a, Keiichi Yamazaki b, Hiroyuki Miki c, Akio Yamazaki b, Hiroshi Kato d and Hideyuki Suzuki d a Institute of Engineering Mechanics and Systems, University of Tsukuba b Faculty of Liberal Arts, Saitama University c Media Laboratories, Oki Electric Industry Co. Ltd. d C&C Media Research Laboratories, NEC Corporation 1 INTRODUCTION One way in which collaboration can take place is that several people gather around a table to tackle a shared problem. We have designed the Agora Telecollaboration System to allow several spatially separated collaborators to work as though they were deliberating on and cooperating in the solution of a problem gathered around one table (Kuzuoka et al. 1999). We first clarified which conditions must be met to allow collaboration on a problem on the same desktop, by observing how collaboration actually works. We then designed a telecollaboration system to fulfill each of these conditions. Finally a series of experiments to find out if the system did indeed successfully support remote collaboration was conducted. 2 DESIGN IMPLICATIONS Ethnomethodological studies have pointed out that when people are collaborating, participants should monitor each others conduct to determine their own conduct. In case of conversation, for example, Goodwin revealed that a recipient expressed that he/she was ready, to listen to a speaker s utterances by turning his/her gaze toward a speaker. When a speaker did not monitor that the recipient was ready he/she could not proceed with the conversation (Goodwin 1981). In case of a task that participants should deal with shared artifacts, more resources should be monitored mutually.
Camera to take participants feedback elimination circuit Camera to take desktop Figure 1. Schematic diagram of Agora system. When people communicate through mediated communication systems, however, it has been pointed out that gaze, gestures, and other body movements are generally not as effective as in normal face-to-face communication. Heath and Luff found asymmetries in interpersonal communication through mediated presence system, which make the resources, such as gestures and gaze that a speaker might ordinarily use to shape the way in which a recipient should participate unreliable (Heath and Luff 1992). Consequently, it is necessary to devise a system that virtually embodies participants conduct. Observations of actual multi-participant collaboration and the previous studies made clear that the following elements are essential for collaboration support. A) Several people must be able to create a shared workspace. B) Each participant must be able to orient himself or herself towards the shared workspace and this orientation of each worker must be visible to all other participants. C) Body movements and the use of tools in the shared workspace must occur in an interactive fashion. D) Participants must be able to work with different partners and must be able to organize workspaces accordingly. The other participants must be able to see the work that is being done in such partnerships. 3 DESIGN OF AGORA The AGORA Telecollaboration System consists, roughly speaking, of two parts. One part is the shared space of images and actual objects on a desk, which are created using projection. The other part consists of large screens onto which life-size images of remote participants are projected. For the current ex-
periment the same setup was installed in two locations which were connected by audio and circuits. Schematic diagram of Agora is shown in Fig. 1. Images of remote participants were projected onto two 80-inch screens, which were placed in an L-shape around the desk. Remote participants thus appeared in front on the other side of the desk or to one side (either right or left). The screens, for which we used back-projection, showed the upper body of the participants at about life-size. The pictures were captured by CCD cameras mounted on transparent stands extending from the bottom of each screen. They were placed at a position close to the projected eyes of the remote participants, so that it was, if not perfectly, possible to transmit at least to some extent where the remote participants were looking. The pictures on the screens and the projection on the desktop were adjusted to appear seamless. That is, in this setup, the projected picture of a remote participant's hands performing work on the desktop was continuous with the picture of that person as it was projected onto the front (or side) screen. Desktop sharing was achieved by using a similar configuration to DoubleDigitalDesk (Wellner 1993). At each site a camera and were installed above the desk. An image of the desktop at one location was projected onto the desktop at the other, and vice versa. It is know that the direct connection between two desktops in this way causes infinite feedback (the image of your own desk transmitted back from the remote desk). Existing systems use polarized films (Ishii and Kobayashi1992, Tang and Minneman 1990) and image thresholding (Wellner 1993) to avoid feedback. Polarized films cannot be used for our system, however, since polarized light is diffused when reflected on a normal desk surface. Image thresholding is also problematic since the brightness of the projected image should be limited, relatively dark objects are thretholded by mistake, and image procdessing time causes delay in transmission. Thus we developed new hardware to eliminate feedback in realtime. It is electric hardware that inserts a single-color frame once every two frames of the images from the remote desktop camera. Thus the remote desktop images are projected only pnce every two frames. The local camera captured only when remote desktop is not displayed are projected back onto the remote desktop. In this way feedback is eliminated with no processing delay. This method reduces the frame rate from 30 frames/s (NTSC) to 15 frames/s but it does not disturb participants. 4 EXPERIMENT We conducted experiments (Fig. 2) with Agora Telecollaboration System and more then 60 subjects served as participants so far. Various seat arrengements
Cameras to capture participants Remote participant s hand Figure 2. Agora in use. Remote participant s gesture is outlined in this figure. and number of participants (up to six) were tried. The experiments were recorded on tape, observed, and analyzed with respect to the four conditions listed in Section 2. A) We were able to confirm that multiple participants were needed using the desktop as a shared workspace. B) Since the desktop and the directions where the participants were sitting around the table were clearly distinguished, collaborators were able to discern where (at the desktop or another worker) a participant was looking, from the gaze and posture of the picture projected onto the front or side screen. From the movement of the image of hands projected onto the desktop, collaborators were able to determine that other workers were oriented towards the shared workspace. C) We were able to confirm that interactive work was taking place in the shared workspace such that, after one participant did something or made a verbal comment, other participants responded by moving hands or things accordingly in the shared workspace. D) In the case of 3 or more participants, a variety of shared workspaces have to be created in response to each participant's relation with other participants. Furthermore, it is necessary for participants to be able to freely reorganize workspaces, as, for instance, in the case where work first proceeds with a division of labor, but in it the end all have to work together. Our observations of the experiments confirm that all this is possible with the given setup.
5 SYSTEM LIMITATIONS AND FUTURE WORK We have developed the Agora Telecollaboration system to allow several spatially separated participants to work as though they were collaborating face to face. The analysis of the experiments, however, teaches us some limitations of Agora listed as follows: Since the remote participants gaze directions are not completely accurate, eye contact may not be maintained for more than six participants. Due to the camera settings, images are not continuous at the edge where two screens adjoin each other. And because of monochrome cameras, images on the desktop and L-shaped screens are low contrast. Consequently participants sometimes could not distinguish papers put on the desktop. There will be inconvenience if complex 3D objects should be shared. Since Agora Telecollaboration System enabled mutual monitoring of participants conducts, we think that various DigitalDesk type applications can effectively be used cooperatively. However, sometimes participants wanted to manipulate objects at a remote site, so we also plan to integrate other mechanism to remotely manipulate physical objects (Brave 1998). We are thinking of using Agora as a platform for not only digital but also physical applications. 6 REFERENCES Brave, S., Ishii, H., & Dahley, A. (1998), Tangible Interfaces for Remote Collaboration and Communication. In Proc. of CSCW 98, pp.169-178. Goodwin, C. (1981), Conversational Organization: Interaction between speakers and hearers. Academic Press. Ishii, H. & Kobayashi, M. (1992), Clearboard: A Seamless Medium for Shared Drawing and Conversation with Eye Contact. Proc. of CHI 92, 525-532. Kuzuoka, H., Yamashita, J., Yamazaki, K., & Yamazaki, A. (1999), Agora: A remote collaboration system that enables mutual monitoring, Proc. of CHI 99 Heath, C. & Luff, P. (1992), Disembodied Conduct: Interactional Asymmetries in Video-Mediated Communication. Rank Xerox Tech. Rep., EPC-1992-119. Tang, J. & Minneman, S. (1990), Videodraw: A Video Intrerface for Collaborative Drawing. Proc. of CHI 90, 313-320. Wellner, P. (1993), Interacting with Paper on the DigitalDesk. Communications of the ACM, 36(7), 86-96.