VideoBots: Robotic avatars for desktop teleconferencing

The VideoBot system is designed to enhance the remote presence of teleconference participants. Each remote user is represented by a life-size “talking head” video display mounted on a robotic pedestal with three degrees of freedom. A video camera affixed to the display provides a local image to the remote participant. Moving the robotic display moves the camera in lockstep, and preserves the gaze affordances of face-to-face meetings, where people naturally perceive gaze direction by head orientation. Because the display and camera are mechanically linked, it is easy to tell where the remote participant is looking, because the display is pointed in that direction, which naturally cues local users to the remote user’s gaze direction. A high-quality video link delivers an exceptionally realistic face image and voice quality, increasing the realism and allowing subtle nuances of expression to be conveyed. In addition, the display can move dynamically to attract attention, change gaze direction, indicate degree and direction of interest, and even give motion cues like nodding and bowing.

It is impossible to look at a VideoBot display and not look in the direction of the camera. Studies have shown that the perception of eye contact is asymmetrical, and that users are much less sensitive to gaze direction in the vertical plane. Thus the gaze problem is mitigated to a large extent. Pointing the display in the same direction as the camera naturally cues local users to the remote participant’s gaze direction. As the remote user steers the camera to different regions, the display follows giving immediate feedback to local users about what the remote user can see. If a VideoBot is directed away from a local user, he or she will know that the remote user cannot see them.
VideoBot prototypes consist of 17'' diagonal XVGA color LCD displays mounted on custom robotic platforms. The power supply has been removed from the display and installed in a separate enclosure to save weight. Each of the three axis (rotation, up/down, and in/out) is driven with a powerful 24V DC motor, using a Solutions Cubed MicroPID servocontroller. Each motor is equipped with a reduction gear and an optical position encoder. The servocontroller programmatically drives the motor to the desired position in a feedback loop. Limit switches detect when the motion of a particular axis reaches the maximum or minimum extent. They allow the optical encoders to be calibrated to an absolute position, and also serve as a fail-safe that inactivates the motor in the event it is erroneously requested to move beyond its permitted range.
We use a blend of custom and off-the-shelf software in the VideoBot system. Because the life-size display emphasizes the shortcomings of typical compressed video, we use the Digital Video format (DV) from a consumer Sony camcorder via the IEEE 1994 (``FireWire'') interface,  transmitted over the IP protocol using the WIDE DVTS software. This provides superlative stereo 48 kHz audio and full frame rate 720 X 480 video. Though there is some latency, it is not enough to impact usability. To match the portrait format of the VideoBot display, we resort to the simple expedient of rotating the capture camcorder sideways. For the return audio and video, image fidelity is less important than small camera size (as the camera must be mounted on the moveable display). So for the local-to-remote video, we use a
off-the-shelf  USB video camera with 640 X 480 resolution. Return audio and video is transmitted via the open-source VideoLAN streaming software.
A WXPython front end is used to remotely steer the VideoBot. Users see a conventional video window showing the view from the VideoBot camera, and use a joystick to control VideoBot movement. Left/righ joystick movement rotates the VideoBot, while forward/back joystick movement controls either the up/down axis or, if the trigger button is held, the the in/out axis. If available, the joystick “rudder control” also controls in/out axis movement.

An artist's conception of VideoBot use in a meeting. (image courtesy Surapong Lertsithichai)

A video demonstration will be available Real Soon Now; in the meanwhile you may watch this VideoBot video (5.4 Mb mpg) taken during system development.
Thanks to Tim Black at Quantalink and Conor McQuaid for help with design and mechanical fabrication.
contact: Jonathan Foote

Related Publications

Copyright ©1999-2014 FX Palo Alto Laboratory | Send feedback to the webmaster